cli: glibc's allocator carries a lot of overhead on the graphs the be… by kali · Pull Request #2789 · sonos/tract

kali · 2026-09-07T07:59:06Z

…nch suite runs, and a process can settle into a regime where a per-eval intermediate is mapped and unmapped on every pass, costing millions of minor faults. Use jemalloc as the CLI global allocator, wrapped in the same readings instrumentation, behind a jemalloc feature that bench-suite turns on.

github-actions · 2026-09-07T08:48:13Z

🔴 Bench vs main — 4 speed regression(s) · ⚠️ 107 secondary

Reference: 2026-09-07 morning nightly run (0d old) · full report → run

Speed — evaltime · prefill · decode

Δ	metric	device	main → PR
🔴 +7.5%	inceptionv3 _{evaltime · pass}	`apple-m1-max`	59.1 ms → 63.6 ms
🔴 +7.2%	inceptionv3 _{evaltime · pass_mt}	`apple-m1-max`	45.9 ms → 49.2 ms
🔴 +7.2%	mobilenet_v1_1 _{evaltime · pass}	`apple-m1-max`	18 ms → 19.3 ms
🔴 +4.9%	llama_3_2_1B_instruct_q40ef16_541 _{prefill · cpu_mt}	`apple-m1-max`	1.88 ms/tok _{533.2 tok/s} → 1.97 ms/tok _{508.3 tok/s}

Improvements

Δ	metric	device	main → PR
🟢 -6.6%	hey_snips_v4_model17_nnef _{evaltime · pulse8}	`apple-m1-max`	0.116 ms/pulse _{0.00145 RTF} → 0.108 ms/pulse _{0.00135 RTF}
🟢 -6.2%	hey_snips_v4_model17 _{evaltime · pulse8}	`apple-m1-max`	0.116 ms/pulse _{0.00144 RTF} → 0.108 ms/pulse _{0.00135 RTF}
🟢 -5.1%	inceptionv1q _{evaltime · pass}	`apple-m1-max`	41.6 ms → 39.4 ms

⚠️ 107 secondary regression(s)

Δ	metric	device	main → PR
⚠️ +54.6%	en_tdnn_8M_nnef _{RSS @ ready · pulse_240ms}	`i9-11900kb_rtx-4060`	59.5 MB → 92 MB
⚠️ +53.6%	hey_snips_v4_model17 _{RSS @ load · 2sec}	`i9-11900kb_rtx-4060`	33.8 MB → 51.9 MB
⚠️ +48.6%	mdl_en_2019_Q3_librispeech_onnx _{RSS @ load · 2600ms}	`i9-11900kb_rtx-4060`	36 MB → 53.5 MB
⚠️ +46.9%	hey_snips_v4_model17 _{RSS @ load · pulse8}	`i9-11900kb_rtx-4060`	36.1 MB → 53 MB
⚠️ +46.4%	en_tdnn_8M _{RSS @ ready · pulse_180ms}	`i9-11900kb_rtx-4060`	72.6 MB → 106 MB
⚠️ +45.1%	mdl_en_2019_Q3_librispeech_onnx _{RSS @ load · pulse_240ms}	`i9-11900kb_rtx-4060`	37.6 MB → 54.5 MB
⚠️ +44.7%	en_tdnn_8M _{RSS @ ready · pulse_240ms}	`i9-11900kb_rtx-4060`	73.8 MB → 107 MB
⚠️ +44.3%	hey_snips_v4_model17 _{RSS @ ready · 2sec}	`i9-11900kb_rtx-4060`	36.9 MB → 53.3 MB
⚠️ +43.8%	en_tdnn_8M _{RSS @ ready · 2600ms}	`i9-11900kb_rtx-4060`	73.3 MB → 105 MB
⚠️ +43.3%	hey_snips_v4_model17 _{RSS @ ready · pulse8}	`i9-11900kb_rtx-4060`	38 MB → 54.4 MB
⚠️ +41.8%	inceptionv1q _{RSS @ ready · pass}	`i9-11900kb_rtx-4060`	43.4 MB → 61.6 MB
⚠️ +40.8%	en_tdnn_8M _{RSS @ ready · pulse_120ms}	`i9-11900kb_rtx-4060`	75.5 MB → 106 MB
⚠️ +40.0%	parakeet_tdt_600m_v3_f32f32_decoder_pass _{RSS @ load · cuda}	`i9-11900kb_rtx-4060`	69.2 MB → 97 MB
⚠️ +39.9%	voicecom_fake_quant _{RSS @ load · 2sec}	`i9-11900kb_rtx-4060`	27.1 MB → 37.8 MB
⚠️ +39.7%	parakeet_tdt_600m_v3_f32f32_decoder_pass _{RSS @ load · cpu}	`i9-11900kb_rtx-4060`	69.4 MB → 96.9 MB
⚠️ +39.6%	nemotron_3_5_asr_streaming_0_6b_f32f32_decoder_pass _{RSS @ load · cuda}	`jetson-orin-nx`	75.7 MB → 106 MB
⚠️ +36.5%	openelm_270M_q40ef16_541 _{RSS @ ready · cpu}	`i9-11900kb_rtx-4060`	192 MB → 262 MB
⚠️ +36.4%	en_tdnn_8M _{RSS @ load · pulse_180ms}	`i9-11900kb_rtx-4060`	64.4 MB → 87.9 MB
⚠️ +35.5%	en_tdnn_8M _{RSS @ load · 2600ms}	`i9-11900kb_rtx-4060`	64.1 MB → 86.9 MB
⚠️ +35.3%	inceptionv1q _{RSS @ load · pass}	`i9-11900kb_rtx-4060`	30.7 MB → 41.5 MB
⚠️ +35.2%	en_tdnn_8M _{RSS @ load · pulse_240ms}	`i9-11900kb_rtx-4060`	65.6 MB → 88.7 MB
⚠️ +34.8%	voicecom_float _{RSS @ load · 2sec}	`i9-11900kb_rtx-4060`	26.8 MB → 36.1 MB
⚠️ +34.4%	parakeet_tdt_600m_v3_f32f32_decoder_pass _{RSS @ load · cuda}	`jetson-orin-nx`	67.2 MB → 90.3 MB
⚠️ +33.7%	en_tdnn_8M _{RSS @ load · pulse_120ms}	`i9-11900kb_rtx-4060`	65.8 MB → 88 MB
⚠️ +33.7%	parakeet_tdt_600m_v3_f32f32_joint_pass _{RSS @ ready · cpu}	`i9-11900kb_rtx-4060`	51.3 MB → 68.6 MB
⚠️ +33.5%	mdl_en_2019_Q3_librispeech_onnx _{RSS @ ready · 2600ms}	`i9-11900kb_rtx-4060`	41.4 MB → 55.2 MB
⚠️ +32.4%	mdl_en_2019_Q3_librispeech_onnx _{RSS @ ready · pulse_240ms}	`i9-11900kb_rtx-4060`	42.4 MB → 56.1 MB
⚠️ +32.0%	nemotron_3_5_asr_streaming_0_6b_f32f32_decoder_pass _{RSS @ load · cpu}	`i9-11900kb_rtx-4060`	81.6 MB → 108 MB
⚠️ +31.5%	nemotron_3_5_asr_streaming_0_6b_f32f32_decoder_pass _{RSS @ load · cuda}	`i9-11900kb_rtx-4060`	81.9 MB → 108 MB
⚠️ +31.3%	nemotron_3_5_asr_streaming_0_6b_f32f32_preprocessor_1s _{RSS @ load · cuda}	`jetson-orin-nx`	21 MB → 27.6 MB
⚠️ +31.1%	qwen3_1_7B_q40ef16_541 _{RSS @ ready · cuda}	`jetson-orin-nx`	393 MB → 515 MB
⚠️ +30.1%	en_tdnn_15M_nnef _{RSS @ ready · pulse_240ms}	`i9-11900kb_rtx-4060`	96 MB → 125 MB
⚠️ +29.9%	hey_snips_v4_model17_nnef _{RSS @ ready · pulse8}	`i9-11900kb_rtx-4060`	27.9 MB → 36.2 MB
⚠️ +29.0%	parakeet_tdt_600m_v3_f32f32_preprocessor_1s _{RSS @ load · cuda}	`jetson-orin-nx`	21.3 MB → 27.4 MB
⚠️ +28.9%	trunet _{RSS @ load · pulse1_f16}	`i9-11900kb_rtx-4060`	27.6 MB → 35.6 MB
⚠️ +27.6%	hey_snips_v1 _{RSS @ ready · 400ms}	`i9-11900kb_rtx-4060`	25.7 MB → 32.8 MB
⚠️ +27.0%	arm_ml_kws_cnn_m _{RSS @ ready · pass}	`i9-11900kb_rtx-4060`	26.5 MB → 33.7 MB
⚠️ +27.0%	arm_ml_kws_cnn_m _{RSS @ load · pass}	`i9-11900kb_rtx-4060`	23.4 MB → 29.7 MB
⚠️ +26.7%	en_tdnn_pyt_15M _{RSS @ load · pulse_120ms}	`i9-11900kb_rtx-4060`	85.6 MB → 108 MB
⚠️ +26.5%	hey_snips_v31 _{RSS @ load · 400ms}	`i9-11900kb_rtx-4060`	23.6 MB → 29.9 MB
⚠️ +26.3%	trunet _{RSS @ load · pulse1_f32}	`i9-11900kb_rtx-4060`	29.4 MB → 37.2 MB
⚠️ +26.0%	openelm_270M_q40ef16_541 _{RSS @ load · cuda}	`jetson-orin-nx`	200 MB → 252 MB
⚠️ +25.7%	nemotron_3_5_asr_streaming_0_6b_f32f32_preprocessor_1s _{RSS @ load · cpu}	`i9-11900kb_rtx-4060`	23.5 MB → 29.5 MB
⚠️ +25.5%	nemotron_3_5_asr_streaming_0_6b_f32f32_joint_pass _{RSS @ ready · cpu}	`i9-11900kb_rtx-4060`	64.5 MB → 81 MB
⚠️ +24.2%	hey_snips_v1 _{RSS @ load · 400ms}	`i9-11900kb_rtx-4060`	22.6 MB → 28 MB
⚠️ +24.1%	en_tdnn_lstm_bn_q7 _{RSS @ load · 2600ms}	`i9-11900kb_rtx-4060`	28.1 MB → 34.9 MB
⚠️ +23.7%	speaker_id _{RSS @ load · pulse8}	`i9-11900kb_rtx-4060`	24.4 MB → 30.2 MB
⚠️ +23.6%	en_tdnn_lstm_bn_q7 _{RSS @ load · pulse_240ms}	`i9-11900kb_rtx-4060`	29 MB → 35.8 MB
⚠️ +23.5%	parakeet_tdt_600m_v3_f32f32_preprocessor_1s _{RSS @ load · cpu}	`i9-11900kb_rtx-4060`	23.7 MB → 29.3 MB
⚠️ +23.3%	parakeet_tdt_600m_v3_f32f32_preprocessor_1s _{RSS @ load · cuda}	`i9-11900kb_rtx-4060`	23.8 MB → 29.3 MB
⚠️ +23.1%	en_tdnn_15M _{RSS @ load · pulse_240ms}	`i9-11900kb_rtx-4060`	111 MB → 136 MB
⚠️ +23.0%	nemotron_3_5_asr_streaming_0_6b_f32f32_preprocessor_1s _{RSS @ ready · cpu}	`i9-11900kb_rtx-4060`	25.3 MB → 31.2 MB
⚠️ +22.9%	nemotron_3_5_asr_streaming_0_6b_f32f32_preprocessor_1s _{RSS @ load · cuda}	`i9-11900kb_rtx-4060`	23.8 MB → 29.2 MB
⚠️ +22.8%	qwen3_1_7B_q40ef16_541 _{RSS @ ready · cuda}	`i9-11900kb_rtx-4060`	575 MB → 706 MB
⚠️ +22.7%	trunet _{RSS @ ready · pulse1_f16}	`i9-11900kb_rtx-4060`	30.4 MB → 37.3 MB
⚠️ +22.6%	en_tdnn_15M _{RSS @ load · pulse_120ms}	`i9-11900kb_rtx-4060`	111 MB → 136 MB
⚠️ +22.0%	en_tdnn_15M _{RSS @ load · 2600ms}	`i9-11900kb_rtx-4060`	111 MB → 135 MB
⚠️ +21.4%	nemotron_3_5_asr_streaming_0_6b_f32f32_preprocessor_pulse100ms _{RSS @ load · cpu}	`i9-11900kb_rtx-4060`	25.2 MB → 30.6 MB
⚠️ +21.3%	speaker_id _{RSS @ ready · pulse8}	`i9-11900kb_rtx-4060`	26.5 MB → 32.1 MB
⚠️ +21.2%	hey_snips_v31 _{RSS @ ready · 400ms}	`i9-11900kb_rtx-4060`	26.1 MB → 31.6 MB
⚠️ +21.0%	nemotron_3_5_asr_streaming_0_6b_f32f32_preprocessor_pulse100ms _{RSS @ ready · cpu}	`i9-11900kb_rtx-4060`	26.4 MB → 31.9 MB
⚠️ +20.7%	openelm_270M_q40ef16_541 _{RSS @ load · cuda}	`i9-11900kb_rtx-4060`	206 MB → 249 MB
⚠️ +20.6%	trunet _{RSS @ ready · pulse1_f32}	`i9-11900kb_rtx-4060`	32.5 MB → 39.2 MB
⚠️ +20.4%	parakeet_tdt_600m_v3_f32f32_preprocessor_1s _{RSS @ ready · cpu}	`i9-11900kb_rtx-4060`	25.8 MB → 31.1 MB
⚠️ +20.3%	nemotron_3_5_asr_streaming_0_6b_f32f32_preprocessor_pulse100ms _{RSS @ load · cuda}	`i9-11900kb_rtx-4060`	25.5 MB → 30.7 MB
⚠️ +20.3%	openelm_270M_q40ef16_541 _{RSS @ load · cpu}	`i9-11900kb_rtx-4060`	206 MB → 248 MB
⚠️ +19.8%	hey_snips_v4_model17_nnef _{RSS @ load · pulse8}	`i9-11900kb_rtx-4060`	24.5 MB → 29.3 MB
⚠️ +19.6%	llama_3_2_1B_q40ef32_516 _{RSS @ ready · cuda}	`jetson-orin-nx`	356 MB → 426 MB
⚠️ +19.2%	llama_3_2_3B_instruct_q40ef16_541 _{RSS @ ready · cuda}	`jetson-orin-nx`	433 MB → 516 MB
⚠️ +18.6%	en_tdnn_15M _{RSS @ ready · pulse_240ms}	`i9-11900kb_rtx-4060`	127 MB → 151 MB
⚠️ +18.6%	en_tdnn_lstm_bn_q7 _{RSS @ ready · pulse_240ms}	`i9-11900kb_rtx-4060`	31.9 MB → 37.8 MB
⚠️ +18.5%	llama_3_2_1B_instruct_q40ef16_541 _{RSS @ ready · cuda}	`jetson-orin-nx`	357 MB → 423 MB
⚠️ +18.3%	llama_3_2_3B_q40ef32_516 _{RSS @ ready · cuda}	`jetson-orin-nx`	436 MB → 516 MB
⚠️ +17.9%	en_tdnn_lstm_bn_q7 _{RSS @ ready · 2600ms}	`i9-11900kb_rtx-4060`	31.7 MB → 37.3 MB
⚠️ +17.2%	dummy_conmer_12M _{RSS @ load · pulse_120ms}	`i9-11900kb_rtx-4060`	85.9 MB → 101 MB
⚠️ +17.2%	en_tdnn_15M _{RSS @ ready · pulse_120ms}	`i9-11900kb_rtx-4060`	128 MB → 151 MB
⚠️ +17.2%	inceptionv3 _{RSS @ load · pass_mt}	`i9-11900kb_rtx-4060`	151 MB → 176 MB
⚠️ +17.1%	llama_3_2_3B_q40ef32_516 _{RSS @ ready · cuda}	`i9-11900kb_rtx-4060`	617 MB → 722 MB
⚠️ +15.7%	en_tdnn_15M_nnef _{RSS @ load · pulse_240ms}	`i9-11900kb_rtx-4060`	92.3 MB → 107 MB
⚠️ +15.5%	en_tdnn_pyt_15M _{RSS @ ready · pulse_120ms}	`i9-11900kb_rtx-4060`	90.5 MB → 105 MB
⚠️ +15.3%	en_tdnn_8M_nnef _{RSS @ load · pulse_240ms}	`i9-11900kb_rtx-4060`	54.4 MB → 62.7 MB
⚠️ +14.8%	inceptionv3 _{RSS @ load · pass}	`i9-11900kb_rtx-4060`	150 MB → 172 MB
⚠️ +14.5%	llama_3_2_3B_instruct_q40ef16_541 _{RSS @ ready · cuda}	`i9-11900kb_rtx-4060`	618 MB → 708 MB
⚠️ +13.8%	nemotron_3_5_asr_streaming_0_6b_f32f32_decoder_pass _{RSS @ ready · cuda}	`jetson-orin-nx`	215 MB → 245 MB
⚠️ +13.2%	parakeet_tdt_600m_v3_f32f32_decoder_pass _{RSS @ ready · cpu}	`i9-11900kb_rtx-4060`	87.2 MB → 98.7 MB
⚠️ +12.5%	llama_3_2_1B_instruct_q40ef16_541 _{RSS @ ready · cuda}	`i9-11900kb_rtx-4060`	538 MB → 606 MB
⚠️ +12.3%	llama_3_2_1B_q40ef32_516 _{RSS @ ready · cpu}	`i9-11900kb_rtx-4060`	728 MB → 817 MB
⚠️ +12.2%	llama_3_2_1B_instruct_q40ef16_541 _{RSS @ ready · cpu}	`i9-11900kb_rtx-4060`	728 MB → 817 MB
⚠️ +12.0%	llama_3_2_1B_instruct_q40ef16_541 _{RSS @ ready · cpu_mt}	`i9-11900kb_rtx-4060`	727 MB → 814 MB
⚠️ +11.8%	llama_3_2_1B_q40ef32_516 _{RSS @ ready · cuda}	`i9-11900kb_rtx-4060`	539 MB → 603 MB
⚠️ +11.8%	llama_3_1_8B_instruct_q40ef16_541 _{RSS @ ready · cuda}	`i9-11900kb_rtx-4060`	690 MB → 771 MB
⚠️ +11.3%	llama_3_2_1B_q40ef32_516 _{RSS @ load · cuda}	`jetson-orin-nx`	730 MB → 812 MB
⚠️ +10.2%	llama_3_2_1B_instruct_q40ef16_541 _{RSS @ load · cuda}	`jetson-orin-nx`	728 MB → 803 MB
⚠️ +10.1%	llama_3_2_1B_q40ef32_516 _{RSS @ load · cpu}	`i9-11900kb_rtx-4060`	732 MB → 806 MB
⚠️ +10.1%	llama_3_2_1B_q40ef32_516 _{RSS @ load · cuda}	`i9-11900kb_rtx-4060`	732 MB → 806 MB
⚠️ +10.0%	llama_3_2_1B_instruct_q40ef16_541 _{RSS @ load · cpu_mt}	`i9-11900kb_rtx-4060`	731 MB → 804 MB
⚠️ +9.7%	llama_3_2_1B_instruct_q40ef16_541 _{RSS @ load · cpu}	`i9-11900kb_rtx-4060`	732 MB → 803 MB
⚠️ +9.7%	llama_3_2_1B_instruct_q40ef16_541 _{RSS @ load · cuda}	`i9-11900kb_rtx-4060`	731 MB → 802 MB
⚠️ +9.6%	parakeet_tdt_600m_v3_f32f32_decoder_pass _{RSS @ ready · cuda}	`jetson-orin-nx`	219 MB → 240 MB
⚠️ +9.4%	nemotron_3_5_asr_streaming_0_6b_f32f32_decoder_pass _{RSS @ ready · cpu}	`i9-11900kb_rtx-4060`	100 MB → 109 MB
⚠️ +8.3%	llama_3_1_8B_instruct_q40ef16_541 _{RSS @ ready · cuda}	`jetson-orin-nx`	508 MB → 551 MB
⚠️ +6.6%	llama_3_2_3B_instruct_q40ef16_541 _{RSS @ load · cuda}	`i9-11900kb_rtx-4060`	1.85 GB → 1.98 GB
⚠️ +6.5%	llama_3_2_3B_instruct_q40ef16_541 _{RSS @ load · cuda}	`jetson-orin-nx`	1.85 GB → 1.97 GB
⚠️ +6.5%	llama_3_2_3B_q40ef32_516 _{RSS @ load · cuda}	`i9-11900kb_rtx-4060`	1.85 GB → 1.98 GB
⚠️ +6.1%	qwen3_1_7B_q40ef16_541 _{RSS @ load · cuda}	`i9-11900kb_rtx-4060`	1.02 GB → 1.08 GB
⚠️ +5.7%	qwen3_1_7B_q40ef16_541 _{RSS @ load · cuda}	`jetson-orin-nx`	1.01 GB → 1.07 GB
⚠️ +5.1%	arm_ml_kws_cnn_m _{RSS @ load · pass}	`apple-m1-max`	18.6 MB → 19.5 MB

github-actions · 2026-09-07T10:34:02Z

🔴 Bench vs main — 4 speed regression(s) · ⚠️ 432 secondary

Reference: 2026-09-07 morning nightly run (0d old) · full report → run

Speed — evaltime · prefill · decode

Δ	metric	device	main → PR
🔴 +9.0%	mobilenet_v1_1 _{evaltime · pass}	`apple-m1-max`	18 ms → 19.6 ms
🔴 +8.1%	inceptionv3 _{evaltime · pass}	`apple-m1-max`	59.1 ms → 63.9 ms
🔴 +7.7%	inceptionv3 _{evaltime · pass_mt}	`apple-m1-max`	45.9 ms → 49.4 ms
🔴 +5.3%	llama_3_2_1B_instruct_q40ef16_541 _{prefill · cpu_mt}	`apple-m1-max`	1.88 ms/tok _{533.2 tok/s} → 1.98 ms/tok _{506.3 tok/s}

Improvements

Δ	metric	device	main → PR
🟢 -20.8%	speaker_id _{evaltime · pulse8}	`beaglev-ahead`	3.16 ms/pulse _{0.0394 RTF} → 2.5 ms/pulse _{0.0312 RTF}
🟢 -19.9%	hey_snips_v4_model17 _{evaltime · pulse8}	`beaglev-ahead`	5.88 ms/pulse _{0.0735 RTF} → 4.71 ms/pulse _{0.0589 RTF}
🟢 -16.8%	mobilenet_v2_1 _{evaltime · pass_mt}	`beaglev-ahead`	504 ms → 419 ms
🟢 -14.8%	hey_snips_v4_model17 _{evaltime · 2sec}	`beaglev-ahead`	48.8 ms → 41.6 ms
🟢 -13.1%	trunet _{evaltime · pulse1_f32}	`beaglev-ahead`	5.69 ms/pulse _{0.569 RTF} → 4.95 ms/pulse _{0.495 RTF}

+12 more improvement(s)

Δ	metric	device	main → PR
🟢 -10.3%	hey_snips_v4_model17_nnef _{evaltime · pulse8}	`beaglev-ahead`	5.24 ms/pulse _{0.0655 RTF} → 4.7 ms/pulse _{0.0588 RTF}
🟢 -9.7%	inceptionv3 _{evaltime · pass_mt}	`beaglev-ahead`	2.27 s → 2.05 s
🟢 -9.2%	hey_snips_v4_model17_nnef _{evaltime · pulse8}	`cortex-a9`	6.3 ms/pulse _{0.0787 RTF} → 5.72 ms/pulse _{0.0715 RTF}
🟢 -8.9%	trunet _{evaltime · pulse1_f32}	`cortex-a7`	8.07 ms/pulse _{0.807 RTF} → 7.35 ms/pulse _{0.735 RTF}
🟢 -8.7%	hey_snips_v4_model17 _{evaltime · pulse8}	`cortex-a9`	6.32 ms/pulse _{0.079 RTF} → 5.77 ms/pulse _{0.0722 RTF}
🟢 -8.4%	hey_snips_v4_model17 _{evaltime · pulse8}	`cortex-a55`	2.1 ms/pulse _{0.0262 RTF} → 1.92 ms/pulse _{0.024 RTF}
🟢 -7.6%	mobilenet_v2_1 _{evaltime · pass}	`beaglev-ahead`	1.15 s → 1.06 s
🟢 -7.4%	hey_snips_v4_model17_nnef _{evaltime · pulse8}	`cortex-a7`	6.92 ms/pulse _{0.0865 RTF} → 6.4 ms/pulse _{0.0801 RTF}
🟢 -7.0%	trunet _{evaltime · pulse1_f32}	`cortex-a9`	8.1 ms/pulse _{0.81 RTF} → 7.53 ms/pulse _{0.753 RTF}
🟢 -6.7%	hey_snips_v31 _{evaltime · 400ms}	`beaglev-ahead`	6.17 ms → 5.76 ms
🟢 -6.4%	speaker_id _{evaltime · pulse8}	`cortex-a55`	0.781 ms/pulse _{0.00976 RTF} → 0.731 ms/pulse _{0.00913 RTF}
🟢 -5.1%	inceptionv1q _{evaltime · pass}	`apple-m1-max`	41.6 ms → 39.5 ms

⚠️ 432 secondary regression(s)

Δ	metric	device	main → PR
⚠️ +288.9%	en_tdnn_15M _{RSS @ load · pulse_120ms}	`cortex-a55`	94.8 MB → 369 MB
⚠️ +288.1%	en_tdnn_15M _{RSS @ load · 2600ms}	`cortex-a55`	94.9 MB → 368 MB
⚠️ +287.9%	en_tdnn_15M _{RSS @ load · pulse_240ms}	`cortex-a55`	95.1 MB → 369 MB
⚠️ +246.9%	en_tdnn_15M _{RSS @ ready · 2600ms}	`cortex-a55`	110 MB → 383 MB
⚠️ +244.6%	en_tdnn_15M _{RSS @ ready · pulse_120ms}	`cortex-a55`	111 MB → 384 MB
⚠️ +241.9%	en_tdnn_15M _{RSS @ ready · pulse_240ms}	`cortex-a55`	112 MB → 384 MB
⚠️ +239.9%	inceptionv3 _{RSS @ load · pass_mt}	`cortex-a55`	143 MB → 487 MB
⚠️ +238.9%	inceptionv3 _{RSS @ load · pass}	`cortex-a55`	144 MB → 487 MB
⚠️ +221.4%	inceptionv3 _{RSS @ ready · pass}	`cortex-a55`	152 MB → 489 MB
⚠️ +217.9%	inceptionv3 _{RSS @ ready · pass_mt}	`cortex-a55`	154 MB → 490 MB
⚠️ +190.6%	mobilenet_v2_1 _{RSS @ ready · pass}	`cortex-a55`	52.9 MB → 154 MB
⚠️ +189.6%	mobilenet_v2_1 _{RSS @ ready · pass_mt}	`cortex-a55`	53.1 MB → 154 MB
⚠️ +164.7%	en_tdnn_8M _{RSS @ load · 2600ms}	`cortex-a55`	53.8 MB → 142 MB
⚠️ +164.6%	en_tdnn_8M _{RSS @ load · pulse_240ms}	`cortex-a55`	54.3 MB → 144 MB
⚠️ +164.5%	en_tdnn_8M _{RSS @ load · pulse_120ms}	`cortex-a55`	54.4 MB → 144 MB
⚠️ +164.2%	mobilenet_v1_1 _{RSS @ load · pass}	`cortex-a55`	48 MB → 127 MB
⚠️ +164.1%	en_tdnn_8M _{RSS @ load · pulse_180ms}	`cortex-a55`	54.5 MB → 144 MB
⚠️ +139.5%	mobilenet_v1_1 _{RSS @ ready · pass}	`cortex-a55`	53.6 MB → 128 MB
⚠️ +131.8%	en_tdnn_8M _{RSS @ ready · 2600ms}	`cortex-a55`	62.2 MB → 144 MB
⚠️ +130.7%	en_tdnn_8M _{RSS @ ready · pulse_120ms}	`cortex-a55`	63.3 MB → 146 MB
⚠️ +128.2%	en_tdnn_8M _{RSS @ ready · pulse_240ms}	`cortex-a55`	64 MB → 146 MB
⚠️ +127.3%	en_tdnn_8M _{RSS @ ready · pulse_180ms}	`cortex-a55`	64.1 MB → 146 MB
⚠️ +124.9%	mobilenet_v2_1 _{RSS @ load · pass}	`cortex-a55`	67.6 MB → 152 MB
⚠️ +124.4%	mobilenet_v2_1 _{RSS @ load · pass_mt}	`cortex-a55`	67.7 MB → 152 MB
⚠️ +96.9%	mdl_en_2019_Q3_librispeech_onnx _{RSS @ load · 2600ms}	`cortex-a55`	32.3 MB → 63.6 MB
⚠️ +95.0%	mdl_en_2019_Q3_librispeech_onnx _{RSS @ load · pulse_240ms}	`cortex-a55`	32.5 MB → 63.4 MB
⚠️ +85.6%	mdl_en_2019_Q3_librispeech_onnx _{RSS @ ready · 2600ms}	`cortex-a55`	35.1 MB → 65.2 MB
⚠️ +82.3%	hey_snips_v4_model17 _{RSS @ load · 2sec}	`orangepi-rv2`	16.9 MB → 30.8 MB
⚠️ +81.5%	hey_snips_v4_model17 _{RSS @ load · 2sec}	`beaglev-ahead`	16.9 MB → 30.7 MB
⚠️ +80.7%	mdl_en_2019_Q3_librispeech_onnx _{RSS @ ready · pulse_240ms}	`cortex-a55`	36.1 MB → 65.2 MB
⚠️ +74.6%	hey_snips_v4_model17 _{RSS @ ready · 2sec}	`orangepi-rv2`	18.2 MB → 31.8 MB
⚠️ +72.4%	mobilenet_v2_1 _{RSS @ ready · pass}	`cortex-a7`	46 MB → 79.3 MB
⚠️ +72.3%	mobilenet_v2_1 _{RSS @ ready · pass_mt}	`cortex-a9`	45.7 MB → 78.8 MB
⚠️ +69.4%	voicecom_fake_quant _{RSS @ ready · 2sec}	`beaglev-ahead`	18.3 MB → 30.9 MB
⚠️ +69.2%	hey_snips_v4_model17 _{RSS @ load · pulse8}	`orangepi-rv2`	17.7 MB → 29.9 MB
⚠️ +68.7%	mobilenet_v1_1 _{RSS @ load · pass}	`orangepi-rv2`	31.1 MB → 52.5 MB
⚠️ +68.7%	hey_snips_v4_model17 _{RSS @ load · pulse8}	`beaglev-ahead`	17.8 MB → 30 MB
⚠️ +68.2%	mobilenet_v1_1 _{RSS @ load · pass}	`beaglev-ahead`	31.2 MB → 52.5 MB
⚠️ +68.2%	voicecom_fake_quant _{RSS @ ready · 2sec}	`orangepi-rv2`	18.6 MB → 31.2 MB
⚠️ +65.9%	hey_snips_v4_model17 _{RSS @ ready · 2sec}	`beaglev-ahead`	18.5 MB → 30.7 MB
⚠️ +64.5%	mobilenet_v2_1 _{RSS @ load · pass_mt}	`cortex-a9`	42.5 MB → 69.9 MB
⚠️ +64.4%	mobilenet_v2_1 _{RSS @ load · pass}	`cortex-a7`	43 MB → 70.7 MB
⚠️ +63.8%	mobilenet_v2_1 _{RSS @ load · pass}	`cortex-a9`	42.6 MB → 69.8 MB
⚠️ +63.4%	hey_snips_v4_model17 _{RSS @ ready · pulse8}	`orangepi-rv2`	18.9 MB → 30.9 MB
⚠️ +63.0%	mobilenet_v1_1 _{RSS @ ready · pass}	`orangepi-rv2`	32.9 MB → 53.7 MB
⚠️ +62.2%	inceptionv1q _{RSS @ ready · pass}	`beaglev-ahead`	26.2 MB → 42.5 MB
⚠️ +62.2%	voicecom_fake_quant _{RSS @ ready · 2sec}	`cortex-a7`	22.2 MB → 36 MB
⚠️ +62.0%	mobilenet_v2_1 _{RSS @ ready · pass_mt}	`orangepi-rv2`	40.9 MB → 66.2 MB
⚠️ +61.7%	mobilenet_v2_1 _{RSS @ ready · pass}	`orangepi-rv2`	40.9 MB → 66.2 MB
⚠️ +61.2%	mobilenet_v2_1 _{RSS @ ready · pass_mt}	`beaglev-ahead`	41 MB → 66.2 MB
⚠️ +61.1%	mobilenet_v2_1 _{RSS @ ready · pass}	`beaglev-ahead`	41 MB → 66 MB
⚠️ +60.6%	voicecom_float _{RSS @ ready · 2sec}	`beaglev-ahead`	17.7 MB → 28.4 MB
⚠️ +59.3%	mobilenet_v1_1 _{RSS @ ready · pass}	`beaglev-ahead`	33 MB → 52.5 MB
⚠️ +59.1%	mobilenet_v1_1 _{RSS @ load · pass}	`cortex-a9`	35 MB → 55.7 MB
⚠️ +58.9%	voicecom_float _{RSS @ ready · 2sec}	`cortex-a7`	20.9 MB → 33.1 MB
⚠️ +58.7%	hey_snips_v4_model17 _{RSS @ ready · pulse8}	`beaglev-ahead`	19 MB → 30.1 MB
⚠️ +58.0%	voicecom_float _{RSS @ ready · 2sec}	`orangepi-rv2`	18 MB → 28.4 MB
⚠️ +56.2%	voicecom_float _{RSS @ ready · 2sec}	`cortex-a9`	20.8 MB → 32.5 MB
⚠️ +55.1%	mobilenet_v2_1 _{RSS @ ready · pass_mt}	`cortex-a7`	45.7 MB → 70.9 MB
⚠️ +54.6%	en_tdnn_8M_nnef _{RSS @ ready · pulse_240ms}	`i9-11900kb_rtx-4060`	59.5 MB → 92 MB
⚠️ +54.3%	hey_snips_v4_model17 _{RSS @ load · 2sec}	`i9-11900kb_rtx-4060`	33.8 MB → 52.2 MB
⚠️ +53.3%	en_tdnn_8M _{RSS @ ready · 2600ms}	`cortex-a7`	52.7 MB → 80.7 MB
⚠️ +52.9%	en_tdnn_8M _{RSS @ ready · pulse_120ms}	`cortex-a9`	53.7 MB → 82 MB
⚠️ +52.4%	en_tdnn_8M _{RSS @ ready · 2600ms}	`cortex-a9`	52.8 MB → 80.5 MB
⚠️ +52.2%	voicecom_fake_quant _{RSS @ ready · 2sec}	`cortex-a9`	22.3 MB → 34 MB
⚠️ +52.1%	en_tdnn_8M _{RSS @ ready · pulse_120ms}	`cortex-a7`	53.8 MB → 81.9 MB
⚠️ +51.7%	mobilenet_v2_1 _{RSS @ ready · pass}	`cortex-a53`	52.8 MB → 80.1 MB
⚠️ +51.6%	inceptionv1q _{RSS @ ready · pass}	`orangepi-rv2`	26.2 MB → 39.7 MB
⚠️ +51.1%	mobilenet_v2_1 _{RSS @ ready · pass_mt}	`cortex-a53`	53 MB → 80.1 MB
⚠️ +51.1%	en_tdnn_8M _{RSS @ ready · pulse_240ms}	`cortex-a9`	53.8 MB → 81.3 MB
⚠️ +50.2%	en_tdnn_8M _{RSS @ ready · pulse_240ms}	`cortex-a7`	54.2 MB → 81.4 MB
⚠️ +49.5%	inceptionv1q _{RSS @ load · pass}	`orangepi-rv2`	20.2 MB → 30.3 MB
⚠️ +49.4%	inceptionv1q _{RSS @ load · pass}	`beaglev-ahead`	20.2 MB → 30.2 MB
⚠️ +49.4%	en_tdnn_8M_nnef _{RSS @ ready · pulse_240ms}	`orangepi-rv2`	47.5 MB → 70.9 MB
⚠️ +48.5%	mdl_en_2019_Q3_librispeech_onnx _{RSS @ load · 2600ms}	`i9-11900kb_rtx-4060`	36 MB → 53.5 MB
⚠️ +48.5%	en_tdnn_8M_nnef _{RSS @ ready · pulse_240ms}	`beaglev-ahead`	47.8 MB → 70.9 MB
⚠️ +48.3%	mobilenet_v2_1 _{RSS @ load · pass_mt}	`beaglev-ahead`	38.8 MB → 57.5 MB
⚠️ +48.1%	mobilenet_v2_1 _{RSS @ load · pass_mt}	`orangepi-rv2`	38.8 MB → 57.4 MB
⚠️ +48.1%	mobilenet_v2_1 _{RSS @ load · pass}	`orangepi-rv2`	38.7 MB → 57.4 MB
⚠️ +48.0%	mobilenet_v2_1 _{RSS @ load · pass}	`beaglev-ahead`	38.8 MB → 57.4 MB
⚠️ +47.9%	mobilenet_v1_1 _{RSS @ ready · pass}	`cortex-a9`	36.6 MB → 54.2 MB
⚠️ +46.6%	hey_snips_v4_model17 _{RSS @ load · pulse8}	`i9-11900kb_rtx-4060`	36.1 MB → 52.9 MB
⚠️ +46.5%	en_tdnn_8M _{RSS @ ready · pulse_180ms}	`i9-11900kb_rtx-4060`	72.6 MB → 106 MB
⚠️ +46.3%	hey_snips_v4_model17 _{RSS @ ready · 2sec}	`i9-11900kb_rtx-4060`	36.9 MB → 54 MB
⚠️ +46.0%	hey_snips_v4_model17 _{RSS @ load · 2sec}	`cortex-a55`	28.1 MB → 41.1 MB
⚠️ +45.8%	inceptionv1q _{RSS @ ready · pass}	`cortex-a55`	40.6 MB → 59.2 MB
⚠️ +45.4%	mobilenet_v2_1 _{RSS @ load · pass_mt}	`cortex-a7`	42.6 MB → 62 MB
⚠️ +45.3%	en_tdnn_8M _{RSS @ ready · pulse_180ms}	`cortex-a9`	54.1 MB → 78.7 MB
⚠️ +45.1%	mdl_en_2019_Q3_librispeech_onnx _{RSS @ load · pulse_240ms}	`i9-11900kb_rtx-4060`	37.6 MB → 54.5 MB
⚠️ +44.7%	en_tdnn_8M _{RSS @ ready · pulse_240ms}	`i9-11900kb_rtx-4060`	73.8 MB → 107 MB
⚠️ +44.5%	voicecom_float _{RSS @ load · 2sec}	`orangepi-rv2`	16.3 MB → 23.6 MB
⚠️ +44.2%	en_tdnn_8M _{RSS @ ready · 2600ms}	`i9-11900kb_rtx-4060`	73.3 MB → 106 MB
⚠️ +44.0%	voicecom_fake_quant _{RSS @ load · 2sec}	`cortex-a7`	19.2 MB → 27.7 MB
⚠️ +43.2%	mobilenet_v2_1 _{RSS @ ready · pass}	`cortex-a9`	45.9 MB → 65.7 MB
⚠️ +43.1%	voicecom_float _{RSS @ load · 2sec}	`beaglev-ahead`	16.3 MB → 23.4 MB
⚠️ +43.1%	hey_snips_v4_model17 _{RSS @ ready · pulse8}	`i9-11900kb_rtx-4060`	38 MB → 54.3 MB
⚠️ +42.9%	voicecom_fake_quant _{RSS @ load · 2sec}	`cortex-a9`	19.3 MB → 27.6 MB
⚠️ +41.6%	inceptionv1q _{RSS @ ready · pass}	`i9-11900kb_rtx-4060`	43.4 MB → 61.5 MB
⚠️ +41.6%	voicecom_fake_quant _{RSS @ load · 2sec}	`orangepi-rv2`	17.1 MB → 24.2 MB
⚠️ +40.7%	voicecom_fake_quant _{RSS @ load · 2sec}	`i9-11900kb_rtx-4060`	27.1 MB → 38 MB
⚠️ +40.1%	en_tdnn_pyt_15M _{RSS @ load · pulse_120ms}	`orangepi-rv2`	60.3 MB → 84.4 MB
⚠️ +39.9%	parakeet_tdt_600m_v3_f32f32_decoder_pass _{RSS @ load · cuda}	`i9-11900kb_rtx-4060`	69.2 MB → 96.9 MB
⚠️ +39.6%	hey_snips_v4_model17 _{RSS @ ready · pulse8}	`cortex-a55`	31.5 MB → 43.9 MB
⚠️ +39.6%	nemotron_3_5_asr_streaming_0_6b_f32f32_decoder_pass _{RSS @ load · cuda}	`jetson-orin-nx`	75.7 MB → 106 MB
⚠️ +38.8%	en_tdnn_8M _{RSS @ ready · pulse_180ms}	`cortex-a7`	54 MB → 75 MB
⚠️ +38.4%	hey_snips_v4_model17_nnef _{RSS @ ready · pulse8}	`orangepi-rv2`	16.4 MB → 22.7 MB
⚠️ +38.3%	parakeet_tdt_600m_v3_f32f32_decoder_pass _{RSS @ load · cpu}	`i9-11900kb_rtx-4060`	69.4 MB → 96 MB
⚠️ +38.2%	inceptionv3 _{RSS @ ready · pass_mt}	`beaglev-ahead`	119 MB → 164 MB
⚠️ +37.7%	en_tdnn_pyt_15M _{RSS @ load · pulse_120ms}	`beaglev-ahead`	60.3 MB → 83.1 MB
⚠️ +37.5%	hey_snips_v4_model17_nnef _{RSS @ ready · pulse8}	`beaglev-ahead`	16.7 MB → 22.9 MB
⚠️ +37.3%	en_tdnn_8M _{RSS @ ready · pulse_120ms}	`i9-11900kb_rtx-4060`	75.5 MB → 104 MB
⚠️ +36.7%	en_tdnn_8M _{RSS @ load · pulse_180ms}	`i9-11900kb_rtx-4060`	64.4 MB → 88.1 MB
⚠️ +36.6%	voicecom_float _{RSS @ load · 2sec}	`cortex-a7`	19.5 MB → 26.6 MB
⚠️ +36.5%	inceptionv1q _{RSS @ load · pass}	`i9-11900kb_rtx-4060`	30.7 MB → 41.9 MB
⚠️ +36.3%	openelm_270M_q40ef16_541 _{RSS @ ready · cpu}	`i9-11900kb_rtx-4060`	192 MB → 262 MB
⚠️ +36.0%	voicecom_fake_quant _{RSS @ load · 2sec}	`beaglev-ahead`	17.2 MB → 23.3 MB
⚠️ +35.9%	en_tdnn_8M_nnef _{RSS @ ready · pulse_240ms}	`cortex-a55`	55.5 MB → 75.4 MB
⚠️ +35.8%	en_tdnn_8M _{RSS @ load · 2600ms}	`i9-11900kb_rtx-4060`	64.1 MB → 87 MB
⚠️ +35.7%	trunet _{RSS @ load · pulse1_f16}	`i9-11900kb_rtx-4060`	27.6 MB → 37.4 MB
⚠️ +35.7%	en_tdnn_8M_nnef _{RSS @ ready · pulse_240ms}	`cortex-a53`	55.4 MB → 75.2 MB
⚠️ +35.5%	en_tdnn_8M _{RSS @ ready · pulse_180ms}	`orangepi-rv2`	49.8 MB → 67.4 MB
⚠️ +35.5%	en_tdnn_8M _{RSS @ ready · pulse_120ms}	`orangepi-rv2`	49.8 MB → 67.4 MB
⚠️ +35.4%	en_tdnn_pyt_15M _{RSS @ load · pulse_120ms}	`cortex-a7`	63.6 MB → 86.1 MB
⚠️ +35.4%	en_tdnn_8M _{RSS @ ready · 2600ms}	`orangepi-rv2`	49 MB → 66.3 MB
⚠️ +35.3%	hey_snips_v1 _{RSS @ ready · 400ms}	`orangepi-rv2`	14.9 MB → 20.2 MB
⚠️ +35.0%	en_tdnn_8M _{RSS @ load · pulse_240ms}	`i9-11900kb_rtx-4060`	65.6 MB → 88.6 MB
⚠️ +34.8%	voicecom_float _{RSS @ load · 2sec}	`i9-11900kb_rtx-4060`	26.8 MB → 36.1 MB
⚠️ +34.6%	en_tdnn_8M _{RSS @ ready · pulse_180ms}	`beaglev-ahead`	50.1 MB → 67.4 MB
⚠️ +34.5%	en_tdnn_8M _{RSS @ ready · 2600ms}	`beaglev-ahead`	49.3 MB → 66.3 MB
⚠️ +34.4%	parakeet_tdt_600m_v3_f32f32_decoder_pass _{RSS @ load · cuda}	`jetson-orin-nx`	67.2 MB → 90.3 MB
⚠️ +34.3%	en_tdnn_lstm_bn_q7 _{RSS @ load · pulse_240ms}	`cortex-a7`	22.3 MB → 30 MB
⚠️ +34.3%	en_tdnn_pyt_15M _{RSS @ load · pulse_120ms}	`cortex-a9`	64.2 MB → 86.1 MB
⚠️ +34.0%	hey_snips_v1 _{RSS @ ready · 400ms}	`beaglev-ahead`	14.9 MB → 20 MB
⚠️ +33.8%	en_tdnn_15M _{RSS @ ready · pulse_240ms}	`orangepi-rv2`	88 MB → 118 MB
⚠️ +33.7%	mdl_en_2019_Q3_librispeech_onnx _{RSS @ ready · 2600ms}	`i9-11900kb_rtx-4060`	41.4 MB → 55.3 MB
⚠️ +33.4%	voicecom_float _{RSS @ load · 2sec}	`cortex-a9`	19.3 MB → 25.8 MB
⚠️ +33.4%	en_tdnn_15M _{RSS @ ready · pulse_120ms}	`orangepi-rv2`	88 MB → 117 MB
⚠️ +33.3%	mdl_en_2019_Q3_librispeech_onnx _{RSS @ load · pulse_240ms}	`cortex-a9`	29.4 MB → 39.3 MB
⚠️ +33.3%	en_tdnn_15M _{RSS @ ready · 2600ms}	`orangepi-rv2`	87.2 MB → 116 MB
⚠️ +33.1%	en_tdnn_15M _{RSS @ ready · 2600ms}	`beaglev-ahead`	87.6 MB → 117 MB
⚠️ +33.1%	trunet _{RSS @ ready · pulse1_f16}	`orangepi-rv2`	18.5 MB → 24.6 MB
⚠️ +32.9%	en_tdnn_15M _{RSS @ ready · pulse_120ms}	`beaglev-ahead`	88.4 MB → 118 MB
⚠️ +32.8%	en_tdnn_15M _{RSS @ ready · pulse_240ms}	`beaglev-ahead`	88.4 MB → 117 MB
⚠️ +31.8%	mdl_en_2019_Q3_librispeech_onnx _{RSS @ ready · pulse_240ms}	`i9-11900kb_rtx-4060`	42.4 MB → 55.8 MB
⚠️ +31.7%	parakeet_tdt_600m_v3_f32f32_joint_pass _{RSS @ ready · cpu}	`i9-11900kb_rtx-4060`	51.3 MB → 67.6 MB
⚠️ +31.6%	nemotron_3_5_asr_streaming_0_6b_f32f32_decoder_pass _{RSS @ load · cpu}	`i9-11900kb_rtx-4060`	81.6 MB → 107 MB
⚠️ +31.5%	nemotron_3_5_asr_streaming_0_6b_f32f32_preprocessor_1s _{RSS @ load · cuda}	`jetson-orin-nx`	21 MB → 27.6 MB
⚠️ +31.5%	en_tdnn_8M_nnef _{RSS @ ready · pulse_240ms}	`cortex-a7`	51 MB → 67.1 MB
⚠️ +31.4%	trunet _{RSS @ load · pulse1_f16}	`orangepi-rv2`	17.9 MB → 23.6 MB
⚠️ +31.3%	en_tdnn_15M_nnef _{RSS @ ready · pulse_240ms}	`cortex-a55`	91.5 MB → 120 MB
⚠️ +31.3%	trunet _{RSS @ load · pulse1_f16}	`beaglev-ahead`	18.1 MB → 23.8 MB
⚠️ +31.3%	nemotron_3_5_asr_streaming_0_6b_f32f32_decoder_pass _{RSS @ load · cuda}	`i9-11900kb_rtx-4060`	81.9 MB → 107 MB
⚠️ +31.1%	qwen3_1_7B_q40ef16_541 _{RSS @ ready · cuda}	`jetson-orin-nx`	393 MB → 515 MB
⚠️ +30.5%	en_tdnn_lstm_bn_q7 _{RSS @ load · pulse_240ms}	`cortex-a9`	22.4 MB → 29.2 MB
⚠️ +30.5%	en_tdnn_15M_nnef _{RSS @ ready · pulse_240ms}	`i9-11900kb_rtx-4060`	96 MB → 125 MB
⚠️ +30.1%	en_tdnn_8M _{RSS @ load · pulse_120ms}	`i9-11900kb_rtx-4060`	65.8 MB → 85.6 MB
⚠️ +30.0%	hey_snips_v4_model17_nnef _{RSS @ ready · pulse8}	`i9-11900kb_rtx-4060`	27.9 MB → 36.2 MB
⚠️ +29.7%	mdl_en_2019_Q3_librispeech_onnx _{RSS @ load · 2600ms}	`cortex-a7`	29.2 MB → 37.9 MB
⚠️ +29.6%	mobilenet_v1_1 _{RSS @ load · pass}	`cortex-a7`	35.2 MB → 45.7 MB
⚠️ +29.5%	inceptionv3 _{RSS @ ready · pass}	`beaglev-ahead`	119 MB → 154 MB
⚠️ +29.4%	inceptionv1q _{RSS @ load · pass}	`cortex-a7`	23.9 MB → 30.9 MB
⚠️ +29.4%	hey_snips_v1 _{RSS @ ready · 400ms}	`cortex-a9`	18.5 MB → 24 MB
⚠️ +29.4%	inceptionv1q _{RSS @ load · pass}	`cortex-a9`	24.1 MB → 31.1 MB
⚠️ +29.4%	inceptionv3 _{RSS @ ready · pass}	`cortex-a9`	127 MB → 164 MB
⚠️ +29.1%	parakeet_tdt_600m_v3_f32f32_preprocessor_1s _{RSS @ load · cuda}	`jetson-orin-nx`	21.3 MB → 27.5 MB
⚠️ +29.0%	inceptionv3 _{RSS @ load · pass_mt}	`cortex-a7`	116 MB → 150 MB
⚠️ +28.9%	mdl_en_2019_Q3_librispeech_onnx _{RSS @ load · 2600ms}	`cortex-a9`	29.4 MB → 37.9 MB
⚠️ +28.9%	en_tdnn_8M_nnef _{RSS @ ready · pulse_240ms}	`cortex-a9`	50.8 MB → 65.5 MB
⚠️ +28.8%	inceptionv3 _{RSS @ ready · pass}	`orangepi-rv2`	118 MB → 152 MB
⚠️ +28.7%	en_tdnn_pyt_15M _{RSS @ load · pulse_120ms}	`cortex-a55`	74.2 MB → 95.5 MB
⚠️ +28.3%	hey_snips_v1 _{RSS @ ready · 400ms}	`i9-11900kb_rtx-4060`	25.7 MB → 33 MB
⚠️ +28.3%	en_tdnn_lstm_bn_q7 _{RSS @ load · 2600ms}	`cortex-a9`	22.1 MB → 28.4 MB
⚠️ +28.3%	trunet _{RSS @ ready · pulse1_f16}	`i9-11900kb_rtx-4060`	30.4 MB → 39 MB
⚠️ +28.2%	en_tdnn_lstm_bn_q7 _{RSS @ load · 2600ms}	`cortex-a7`	21.9 MB → 28.1 MB
⚠️ +27.5%	trunet _{RSS @ ready · pulse1_f16}	`beaglev-ahead`	18.6 MB → 23.7 MB
⚠️ +27.3%	hey_snips_v31 _{RSS @ load · 400ms}	`i9-11900kb_rtx-4060`	23.6 MB → 30.1 MB
⚠️ +27.2%	mobilenet_v1_1 _{RSS @ ready · pass}	`cortex-a7`	36.9 MB → 47 MB
⚠️ +26.9%	trunet _{RSS @ load · pulse1_f32}	`i9-11900kb_rtx-4060`	29.4 MB → 37.4 MB
⚠️ +26.9%	en_tdnn_pyt_15M _{RSS @ load · pulse_120ms}	`i9-11900kb_rtx-4060`	85.6 MB → 109 MB
⚠️ +26.8%	arm_ml_kws_cnn_m _{RSS @ ready · pass}	`i9-11900kb_rtx-4060`	26.5 MB → 33.6 MB
⚠️ +26.8%	arm_ml_kws_cnn_m _{RSS @ load · pass}	`i9-11900kb_rtx-4060`	23.4 MB → 29.7 MB
⚠️ +26.7%	trunet _{RSS @ ready · pulse1_f32}	`orangepi-rv2`	20.3 MB → 25.7 MB
⚠️ +26.5%	dummy_conmer_12M _{RSS @ ready · pulse_120ms}	`beaglev-ahead`	69.2 MB → 87.5 MB
⚠️ +26.4%	inceptionv1q _{RSS @ load · pass}	`cortex-a55`	27.2 MB → 34.5 MB
⚠️ +26.4%	en_tdnn_pyt_15M _{RSS @ ready · pulse_120ms}	`orangepi-rv2`	61.4 MB → 77.6 MB
⚠️ +26.3%	en_tdnn_8M _{RSS @ load · pulse_120ms}	`cortex-a9`	51.4 MB → 64.9 MB
⚠️ +26.3%	voicecom_fake_quant _{RSS @ load · 2sec}	`cortex-a55`	22.7 MB → 28.7 MB
⚠️ +26.2%	en_tdnn_pyt_15M _{RSS @ ready · pulse_120ms}	`beaglev-ahead`	61.6 MB → 77.7 MB
⚠️ +26.0%	mdl_en_2019_Q3_librispeech_onnx _{RSS @ load · pulse_240ms}	`cortex-a7`	29.4 MB → 37.1 MB
⚠️ +26.0%	trunet _{RSS @ ready · pulse1_f32}	`beaglev-ahead`	20.4 MB → 25.7 MB
⚠️ +25.9%	inceptionv3 _{RSS @ load · pass}	`beaglev-ahead`	114 MB → 143 MB
⚠️ +25.9%	inceptionv3 _{RSS @ load · pass_mt}	`beaglev-ahead`	114 MB → 143 MB
⚠️ +25.8%	hey_snips_v4_model17 _{RSS @ ready · 2sec}	`cortex-a55`	32.6 MB → 41.1 MB
⚠️ +25.6%	hey_snips_v1 _{RSS @ load · 400ms}	`i9-11900kb_rtx-4060`	22.6 MB → 28.3 MB
⚠️ +25.5%	en_tdnn_8M _{RSS @ load · pulse_120ms}	`cortex-a7`	51.8 MB → 65 MB
⚠️ +25.5%	inceptionv3 _{RSS @ ready · pass_mt}	`cortex-a9`	126 MB → 159 MB
⚠️ +25.5%	en_tdnn_lstm_bn_q7 _{RSS @ ready · pulse_240ms}	`cortex-a7`	24.9 MB → 31.2 MB
⚠️ +25.3%	en_tdnn_pyt_15M _{RSS @ ready · pulse_120ms}	`cortex-a55`	81.3 MB → 102 MB
⚠️ +25.1%	en_tdnn_pyt_15M _{RSS @ ready · pulse_120ms}	`cortex-a7`	65.3 MB → 81.8 MB
⚠️ +24.9%	openelm_270M_q40ef16_541 _{RSS @ load · cuda}	`jetson-orin-nx`	200 MB → 250 MB
⚠️ +24.9%	hey_snips_v1 _{RSS @ ready · 400ms}	`cortex-a7`	18.6 MB → 23.2 MB
⚠️ +24.7%	nemotron_3_5_asr_streaming_0_6b_f32f32_joint_pass _{RSS @ ready · cpu}	`i9-11900kb_rtx-4060`	64.5 MB → 80.5 MB
⚠️ +24.7%	en_tdnn_8M _{RSS @ ready · pulse_240ms}	`orangepi-rv2`	49.8 MB → 62 MB
⚠️ +24.6%	voicecom_fake_quant _{RSS @ load · 2sec}	`cortex-a53`	22.8 MB → 28.4 MB
⚠️ +24.6%	en_tdnn_8M _{RSS @ load · 2600ms}	`cortex-a7`	51.3 MB → 64 MB
⚠️ +24.6%	hey_snips_v4_model17_nnef _{RSS @ load · pulse8}	`orangepi-rv2`	14.5 MB → 18.1 MB
⚠️ +24.5%	arm_ml_kws_cnn_m _{RSS @ ready · pass}	`beaglev-ahead`	15.9 MB → 19.8 MB
⚠️ +24.5%	en_tdnn_lstm_bn_q7 _{RSS @ ready · 2600ms}	`cortex-a9`	24.1 MB → 30 MB
⚠️ +24.5%	en_tdnn_8M _{RSS @ load · pulse_180ms}	`cortex-a9`	52.1 MB → 64.8 MB
⚠️ +24.4%	mdl_en_2019_Q3_librispeech_onnx _{RSS @ ready · pulse_240ms}	`cortex-a9`	32.3 MB → 40.2 MB
⚠️ +24.4%	en_tdnn_lstm_bn_q7 _{RSS @ ready · 2600ms}	`cortex-a7`	24 MB → 29.9 MB
⚠️ +24.4%	en_tdnn_8M _{RSS @ load · pulse_180ms}	`cortex-a7`	51.6 MB → 64.2 MB
⚠️ +24.3%	en_tdnn_lstm_bn_q7 _{RSS @ load · 2600ms}	`i9-11900kb_rtx-4060`	28.1 MB → 34.9 MB
⚠️ +24.3%	en_tdnn_pyt_15M _{RSS @ ready · pulse_120ms}	`cortex-a9`	65.8 MB → 81.7 MB
⚠️ +24.2%	en_tdnn_15M _{RSS @ load · pulse_240ms}	`i9-11900kb_rtx-4060`	111 MB → 137 MB
⚠️ +24.2%	hey_snips_v4_model17 _{RSS @ load · 2sec}	`cortex-a7`	24.1 MB → 29.9 MB
⚠️ +24.2%	speaker_id _{RSS @ load · pulse8}	`i9-11900kb_rtx-4060`	24.4 MB → 30.3 MB
⚠️ +24.1%	mdl_en_2019_Q3_librispeech_onnx _{RSS @ ready · 2600ms}	`cortex-a9`	31.3 MB → 38.8 MB
⚠️ +24.1%	en_tdnn_15M _{RSS @ load · 2600ms}	`i9-11900kb_rtx-4060`	111 MB → 137 MB
⚠️ +24.0%	inceptionv3 _{RSS @ load · pass}	`orangepi-rv2`	113 MB → 141 MB
⚠️ +24.0%	nemotron_3_5_asr_streaming_0_6b_f32f32_preprocessor_1s _{RSS @ load · cuda}	`i9-11900kb_rtx-4060`	23.8 MB → 29.5 MB
⚠️ +24.0%	nemotron_3_5_asr_streaming_0_6b_f32f32_preprocessor_1s _{RSS @ load · cpu}	`i9-11900kb_rtx-4060`	23.5 MB → 29.1 MB
⚠️ +24.0%	en_tdnn_8M _{RSS @ ready · pulse_120ms}	`beaglev-ahead`	50.1 MB → 62.1 MB
⚠️ +23.9%	parakeet_tdt_600m_v3_f32f32_preprocessor_1s _{RSS @ load · cpu}	`i9-11900kb_rtx-4060`	23.7 MB → 29.4 MB
⚠️ +23.9%	en_tdnn_8M _{RSS @ load · pulse_240ms}	`cortex-a9`	51.9 MB → 64.3 MB
⚠️ +23.9%	inceptionv3 _{RSS @ ready · pass_mt}	`orangepi-rv2`	118 MB → 147 MB
⚠️ +23.8%	en_tdnn_8M _{RSS @ ready · pulse_240ms}	`beaglev-ahead`	50.1 MB → 62.1 MB
⚠️ +23.8%	mdl_en_2019_Q3_librispeech_onnx _{RSS @ ready · 2600ms}	`cortex-a7`	31.4 MB → 38.8 MB
⚠️ +23.7%	en_tdnn_8M _{RSS @ load · pulse_240ms}	`cortex-a7`	52 MB → 64.3 MB
⚠️ +23.6%	en_tdnn_8M _{RSS @ load · 2600ms}	`cortex-a9`	51.6 MB → 63.8 MB
⚠️ +23.5%	en_tdnn_lstm_bn_q7 _{RSS @ load · pulse_240ms}	`i9-11900kb_rtx-4060`	29 MB → 35.8 MB
⚠️ +23.4%	arm_ml_kws_cnn_m _{RSS @ ready · pass}	`orangepi-rv2`	16.5 MB → 20.3 MB
⚠️ +23.2%	nemotron_3_5_asr_streaming_0_6b_f32f32_preprocessor_pulse100ms _{RSS @ load · cpu}	`i9-11900kb_rtx-4060`	25.2 MB → 31.1 MB
⚠️ +23.2%	en_tdnn_15M _{RSS @ ready · pulse_120ms}	`cortex-a9`	91.7 MB → 113 MB
⚠️ +23.1%	parakeet_tdt_600m_v3_f32f32_preprocessor_1s _{RSS @ load · cuda}	`i9-11900kb_rtx-4060`	23.8 MB → 29.2 MB
⚠️ +23.0%	hey_snips_v4_model17_nnef _{RSS @ load · pulse8}	`beaglev-ahead`	14.6 MB → 17.9 MB
⚠️ +22.9%	qwen3_1_7B_q40ef16_541 _{RSS @ ready · cuda}	`i9-11900kb_rtx-4060`	575 MB → 707 MB
⚠️ +22.7%	hey_snips_v31 _{RSS @ ready · 400ms}	`i9-11900kb_rtx-4060`	26.1 MB → 32 MB
⚠️ +22.7%	speaker_id _{RSS @ ready · pulse8}	`i9-11900kb_rtx-4060`	26.5 MB → 32.5 MB
⚠️ +22.7%	trunet _{RSS @ load · pulse1_f32}	`orangepi-rv2`	19.2 MB → 23.6 MB
⚠️ +22.6%	en_tdnn_15M _{RSS @ ready · pulse_240ms}	`cortex-a9`	92 MB → 113 MB
⚠️ +22.6%	nemotron_3_5_asr_streaming_0_6b_f32f32_preprocessor_pulse100ms _{RSS @ ready · cpu}	`i9-11900kb_rtx-4060`	26.4 MB → 32.3 MB
⚠️ +22.5%	en_tdnn_8M _{RSS @ ready · pulse_180ms}	`cortex-a53`	63.7 MB → 78.1 MB
⚠️ +22.1%	inceptionv3 _{RSS @ ready · pass_mt}	`cortex-a7`	127 MB → 155 MB
⚠️ +22.1%	hey_snips_v31 _{RSS @ load · 400ms}	`orangepi-rv2`	14.6 MB → 17.8 MB
⚠️ +22.0%	inceptionv1q _{RSS @ load · pass}	`cortex-a53`	27.6 MB → 33.6 MB
⚠️ +22.0%	llama_3_2_3B_q40ef32_516 _{RSS @ ready · cuda}	`jetson-orin-nx`	436 MB → 532 MB
⚠️ +22.0%	en_tdnn_8M _{RSS @ load · pulse_240ms}	`orangepi-rv2`	49 MB → 59.8 MB
⚠️ +21.7%	nemotron_3_5_asr_streaming_0_6b_f32f32_preprocessor_1s _{RSS @ ready · cpu}	`i9-11900kb_rtx-4060`	25.3 MB → 30.8 MB
⚠️ +21.7%	hey_snips_v1 _{RSS @ load · 400ms}	`cortex-a7`	16.7 MB → 20.3 MB
⚠️ +21.7%	en_tdnn_8M _{RSS @ load · pulse_180ms}	`orangepi-rv2`	49 MB → 59.7 MB
⚠️ +21.7%	dummy_conmer_12M _{RSS @ ready · pulse_120ms}	`orangepi-rv2`	68.7 MB → 83.6 MB
⚠️ +21.6%	en_tdnn_8M _{RSS @ load · pulse_120ms}	`orangepi-rv2`	49.1 MB → 59.7 MB
⚠️ +21.6%	hey_snips_v1 _{RSS @ load · 400ms}	`cortex-a9`	16.7 MB → 20.3 MB
⚠️ +21.6%	en_tdnn_8M _{RSS @ ready · pulse_120ms}	`cortex-a53`	63.9 MB → 77.7 MB
⚠️ +21.6%	hey_snips_v31 _{RSS @ load · 400ms}	`beaglev-ahead`	14.6 MB → 17.8 MB
⚠️ +21.5%	en_tdnn_lstm_bn_q7 _{RSS @ ready · pulse_240ms}	`cortex-a9`	25.2 MB → 30.7 MB
⚠️ +21.5%	speaker_id _{RSS @ load · pulse8}	`cortex-a9`	18.4 MB → 22.3 MB
⚠️ +21.5%	inceptionv3 _{RSS @ ready · pass}	`cortex-a7`	126 MB → 153 MB
⚠️ +21.4%	en_tdnn_15M _{RSS @ load · pulse_120ms}	`i9-11900kb_rtx-4060`	111 MB → 135 MB
⚠️ +21.2%	trunet _{RSS @ ready · pulse1_f32}	`i9-11900kb_rtx-4060`	32.5 MB → 39.3 MB
⚠️ +21.2%	en_tdnn_8M _{RSS @ ready · pulse_240ms}	`cortex-a53`	64.1 MB → 77.7 MB
⚠️ +21.2%	en_tdnn_8M _{RSS @ load · 2600ms}	`beaglev-ahead`	48.2 MB → 58.4 MB
⚠️ +21.0%	en_tdnn_8M _{RSS @ load · 2600ms}	`orangepi-rv2`	48.2 MB → 58.4 MB
⚠️ +21.0%	hey_snips_v31 _{RSS @ ready · 400ms}	`orangepi-rv2`	15.8 MB → 19.1 MB
⚠️ +20.7%	nemotron_3_5_asr_streaming_0_6b_f32f32_preprocessor_pulse100ms _{RSS @ load · cuda}	`i9-11900kb_rtx-4060`	25.5 MB → 30.8 MB
⚠️ +20.7%	parakeet_tdt_600m_v3_f32f32_preprocessor_1s _{RSS @ ready · cpu}	`i9-11900kb_rtx-4060`	25.8 MB → 31.2 MB
⚠️ +20.7%	voicecom_float _{RSS @ load · 2sec}	`cortex-a55`	23.8 MB → 28.7 MB
⚠️ +20.2%	openelm_270M_q40ef16_541 _{RSS @ load · cuda}	`i9-11900kb_rtx-4060`	206 MB → 248 MB
⚠️ +20.2%	hey_snips_v1 _{RSS @ load · 400ms}	`beaglev-ahead`	13.7 MB → 16.4 MB
⚠️ +20.1%	openelm_270M_q40ef16_541 _{RSS @ load · cpu}	`i9-11900kb_rtx-4060`	206 MB → 248 MB
⚠️ +20.1%	llama_3_2_1B_q40ef32_516 _{RSS @ ready · cuda}	`jetson-orin-nx`	356 MB → 428 MB
⚠️ +20.1%	speaker_id _{RSS @ load · pulse8}	`orangepi-rv2`	15.9 MB → 19.1 MB
⚠️ +20.0%	hey_snips_v4_model17_nnef _{RSS @ load · pulse8}	`i9-11900kb_rtx-4060`	24.5 MB → 29.4 MB
⚠️ +19.9%	hey_snips_v1 _{RSS @ load · 400ms}	`orangepi-rv2`	13.7 MB → 16.4 MB
⚠️ +19.8%	en_tdnn_8M _{RSS @ load · pulse_180ms}	`beaglev-ahead`	49 MB → 58.7 MB
⚠️ +19.7%	en_tdnn_15M _{RSS @ load · 2600ms}	`beaglev-ahead`	86.3 MB → 103 MB
⚠️ +19.6%	en_tdnn_15M _{RSS @ load · pulse_240ms}	`orangepi-rv2`	87.3 MB → 104 MB
⚠️ +19.5%	en_tdnn_15M _{RSS @ ready · pulse_240ms}	`i9-11900kb_rtx-4060`	127 MB → 152 MB
⚠️ +19.5%	mobilenet_v1_1 _{RSS @ load · pass}	`cortex-a53`	48.1 MB → 57.5 MB
⚠️ +19.5%	llama_3_2_1B_instruct_q40ef16_541 _{RSS @ ready · cuda}	`jetson-orin-nx`	357 MB → 427 MB
⚠️ +19.4%	inceptionv1q _{RSS @ ready · pass}	`cortex-a9`	30.2 MB → 36 MB
⚠️ +19.3%	en_tdnn_15M _{RSS @ load · pulse_240ms}	`beaglev-ahead`	87.3 MB → 104 MB
⚠️ +19.3%	trunet _{RSS @ load · pulse1_f16}	`cortex-a7`	21.7 MB → 25.9 MB
⚠️ +19.3%	en_tdnn_lstm_bn_q7 _{RSS @ ready · pulse_240ms}	`i9-11900kb_rtx-4060`	31.9 MB → 38 MB
⚠️ +19.2%	en_tdnn_15M _{RSS @ load · pulse_120ms}	`orangepi-rv2`	87.3 MB → 104 MB
⚠️ +19.2%	dummy_conmer_12M _{RSS @ load · pulse_120ms}	`beaglev-ahead`	67.8 MB → 80.9 MB
⚠️ +19.1%	en_tdnn_8M _{RSS @ load · pulse_120ms}	`beaglev-ahead`	49.1 MB → 58.4 MB
⚠️ +19.1%	inceptionv3 _{RSS @ load · pass_mt}	`orangepi-rv2`	113 MB → 135 MB
⚠️ +19.1%	en_tdnn_8M _{RSS @ load · pulse_240ms}	`beaglev-ahead`	49 MB → 58.4 MB
⚠️ +19.0%	en_tdnn_15M _{RSS @ load · 2600ms}	`orangepi-rv2`	86.4 MB → 103 MB
⚠️ +18.9%	speaker_id _{RSS @ load · pulse8}	`cortex-a7`	18.6 MB → 22.1 MB
⚠️ +18.9%	en_tdnn_15M_nnef _{RSS @ ready · pulse_240ms}	`orangepi-rv2`	85.7 MB → 102 MB
⚠️ +18.8%	en_tdnn_8M _{RSS @ ready · 2600ms}	`cortex-a53`	64.7 MB → 76.9 MB
⚠️ +18.8%	hey_snips_v4_model17_nnef _{RSS @ ready · pulse8}	`cortex-a9`	19.8 MB → 23.5 MB
⚠️ +18.8%	en_tdnn_15M _{RSS @ load · pulse_120ms}	`cortex-a9`	89.5 MB → 106 MB
⚠️ +18.7%	en_tdnn_15M _{RSS @ load · pulse_120ms}	`beaglev-ahead`	87.3 MB → 104 MB
⚠️ +18.7%	mdl_en_2019_Q3_librispeech_onnx _{RSS @ ready · pulse_240ms}	`cortex-a7`	32.2 MB → 38.2 MB
⚠️ +18.7%	en_tdnn_15M_nnef _{RSS @ ready · pulse_240ms}	`beaglev-ahead`	86.1 MB → 102 MB
⚠️ +18.6%	hey_snips_v4_model17 _{RSS @ ready · 2sec}	`cortex-a7`	25.6 MB → 30.4 MB
⚠️ +18.5%	mdl_en_2019_Q3_librispeech_onnx _{RSS @ load · 2600ms}	`cortex-a53`	32.5 MB → 38.5 MB
⚠️ +18.5%	speaker_id _{RSS @ ready · pulse8}	`orangepi-rv2`	17 MB → 20.2 MB
⚠️ +18.3%	llama_3_2_3B_instruct_q40ef16_541 _{RSS @ ready · cuda}	`jetson-orin-nx`	433 MB → 512 MB
⚠️ +18.3%	inceptionv3 _{RSS @ load · pass}	`cortex-a7`	116 MB → 137 MB
⚠️ +18.2%	trunet _{RSS @ load · pulse1_f32}	`beaglev-ahead`	19.2 MB → 22.7 MB
⚠️ +18.2%	hey_snips_v4_model17_nnef _{RSS @ ready · pulse8}	`cortex-a7`	19.7 MB → 23.3 MB
⚠️ +17.9%	en_tdnn_15M _{RSS @ load · 2600ms}	`cortex-a9`	89.8 MB → 106 MB
⚠️ +17.9%	en_tdnn_pyt_15M _{RSS @ load · pulse_120ms}	`cortex-a53`	74.4 MB → 87.8 MB
⚠️ +17.9%	en_tdnn_15M _{RSS @ ready · 2600ms}	`cortex-a9`	91 MB → 107 MB
⚠️ +17.8%	en_tdnn_15M _{RSS @ load · pulse_240ms}	`cortex-a9`	90.2 MB → 106 MB
⚠️ +17.7%	en_tdnn_lstm_bn_q7 _{RSS @ ready · 2600ms}	`i9-11900kb_rtx-4060`	31.7 MB → 37.3 MB
⚠️ +17.6%	trunet _{RSS @ load · pulse1_f16}	`cortex-a9`	21.8 MB → 25.7 MB
⚠️ +17.6%	en_tdnn_15M _{RSS @ load · pulse_120ms}	`cortex-a7`	90.4 MB → 106 MB
⚠️ +17.5%	dummy_conmer_12M _{RSS @ load · pulse_120ms}	`orangepi-rv2`	67.7 MB → 79.6 MB
⚠️ +17.2%	dummy_conmer_12M _{RSS @ load · pulse_120ms}	`i9-11900kb_rtx-4060`	85.9 MB → 101 MB
⚠️ +17.1%	inceptionv3 _{RSS @ load · pass_mt}	`i9-11900kb_rtx-4060`	151 MB → 176 MB
⚠️ +17.0%	trunet _{RSS @ load · pulse1_f32}	`cortex-a7`	21.7 MB → 25.4 MB
⚠️ +17.0%	hey_snips_v4_model17 _{RSS @ load · 2sec}	`cortex-a9`	24.1 MB → 28.3 MB
⚠️ +16.8%	arm_ml_kws_cnn_m _{RSS @ load · pass}	`orangepi-rv2`	14.6 MB → 17 MB
⚠️ +16.8%	speaker_id _{RSS @ load · pulse8}	`beaglev-ahead`	16 MB → 18.7 MB
⚠️ +16.6%	en_tdnn_lstm_bn_q7 _{RSS @ load · pulse_240ms}	`cortex-a53`	24.5 MB → 28.6 MB
⚠️ +16.6%	en_tdnn_15M _{RSS @ load · pulse_240ms}	`cortex-a7`	90.3 MB → 105 MB
⚠️ +16.4%	en_tdnn_15M _{RSS @ ready · pulse_120ms}	`cortex-a7`	92 MB → 107 MB
⚠️ +16.4%	en_tdnn_15M _{RSS @ ready · pulse_120ms}	`i9-11900kb_rtx-4060`	128 MB → 149 MB
⚠️ +16.3%	en_tdnn_15M_nnef _{RSS @ ready · pulse_240ms}	`cortex-a53`	91.6 MB → 107 MB
⚠️ +16.3%	hey_snips_v4_model17 _{RSS @ load · pulse8}	`cortex-a9`	24.3 MB → 28.3 MB
⚠️ +16.2%	llama_3_2_3B_q40ef32_516 _{RSS @ ready · cuda}	`i9-11900kb_rtx-4060`	617 MB → 717 MB
⚠️ +15.9%	hey_snips_v31 _{RSS @ load · 400ms}	`cortex-a7`	18.3 MB → 21.2 MB
⚠️ +15.9%	en_tdnn_15M_nnef _{RSS @ load · pulse_240ms}	`i9-11900kb_rtx-4060`	92.3 MB → 107 MB
⚠️ +15.7%	en_tdnn_15M _{RSS @ load · 2600ms}	`cortex-a7`	90.2 MB → 104 MB
⚠️ +15.5%	trunet _{RSS @ ready · pulse1_f16}	`cortex-a7`	23 MB → 26.6 MB
⚠️ +15.5%	en_tdnn_pyt_15M _{RSS @ ready · pulse_120ms}	`i9-11900kb_rtx-4060`	90.5 MB → 105 MB
⚠️ +15.3%	inceptionv3 _{RSS @ load · pass_mt}	`cortex-a9`	116 MB → 134 MB
⚠️ +15.1%	en_tdnn_8M_nnef _{RSS @ load · pulse_240ms}	`i9-11900kb_rtx-4060`	54.4 MB → 62.6 MB
⚠️ +15.1%	en_tdnn_15M _{RSS @ ready · 2600ms}	`cortex-a7`	91.4 MB → 105 MB
⚠️ +15.0%	trunet _{RSS @ load · pulse1_f32}	`cortex-a9`	22.5 MB → 25.9 MB
⚠️ +14.9%	en_tdnn_15M _{RSS @ ready · pulse_240ms}	`cortex-a7`	92.1 MB → 106 MB
⚠️ +14.8%	hey_snips_v4_model17 _{RSS @ ready · 2sec}	`cortex-a9`	25.7 MB → 29.5 MB
⚠️ +14.8%	inceptionv3 _{RSS @ load · pass}	`i9-11900kb_rtx-4060`	150 MB → 172 MB
⚠️ +14.6%	llama_3_2_3B_instruct_q40ef16_541 _{RSS @ ready · cuda}	`i9-11900kb_rtx-4060`	618 MB → 709 MB
⚠️ +14.6%	en_tdnn_8M _{RSS @ load · pulse_180ms}	`cortex-a53`	54.3 MB → 62.2 MB
⚠️ +14.4%	en_tdnn_15M _{RSS @ load · pulse_120ms}	`cortex-a53`	95 MB → 109 MB
⚠️ +14.3%	hey_snips_v1 _{RSS @ ready · 400ms}	`cortex-a55`	22.1 MB → 25.2 MB
⚠️ +14.3%	hey_snips_v4_model17 _{RSS @ load · pulse8}	`cortex-a7`	24.3 MB → 27.7 MB
⚠️ +14.2%	en_tdnn_8M _{RSS @ load · pulse_240ms}	`cortex-a53`	54.5 MB → 62.2 MB
⚠️ +14.1%	hey_snips_v31 _{RSS @ load · 400ms}	`cortex-a9`	18.3 MB → 20.8 MB
⚠️ +13.9%	dummy_conmer_12M _{RSS @ ready · pulse_120ms}	`cortex-a7`	76.6 MB → 87.3 MB
⚠️ +13.9%	nemotron_3_5_asr_streaming_0_6b_f32f32_decoder_pass _{RSS @ ready · cuda}	`jetson-orin-nx`	215 MB → 245 MB
⚠️ +13.7%	en_tdnn_8M _{RSS @ load · 2600ms}	`cortex-a53`	54.6 MB → 62.1 MB
⚠️ +13.5%	mdl_en_2019_Q3_librispeech_onnx _{RSS @ ready · 2600ms}	`cortex-a53`	35.2 MB → 40 MB
⚠️ +13.5%	arm_ml_kws_cnn_m _{RSS @ ready · pass}	`cortex-a9`	20.4 MB → 23.1 MB
⚠️ +13.4%	trunet _{RSS @ ready · pulse1_f16}	`cortex-a9`	23.3 MB → 26.5 MB
⚠️ +13.4%	dummy_conmer_12M _{RSS @ ready · pulse_120ms}	`cortex-a9`	77.1 MB → 87.4 MB
⚠️ +13.3%	arm_ml_kws_cnn_m _{RSS @ ready · pass}	`cortex-a7`	20.6 MB → 23.3 MB
⚠️ +13.3%	en_tdnn_8M _{RSS @ load · pulse_120ms}	`cortex-a53`	54.7 MB → 62 MB
⚠️ +13.0%	inceptionv3 _{RSS @ load · pass}	`cortex-a9`	116 MB → 131 MB
⚠️ +12.8%	en_tdnn_15M _{RSS @ load · pulse_240ms}	`cortex-a53`	94.8 MB → 107 MB
⚠️ +12.8%	hey_snips_v4_model17_nnef _{RSS @ load · pulse8}	`cortex-a9`	17.9 MB → 20.1 MB
⚠️ +12.8%	speaker_id _{RSS @ ready · pulse8}	`beaglev-ahead`	17 MB → 19.2 MB
⚠️ +12.8%	en_tdnn_15M _{RSS @ load · 2600ms}	`cortex-a53`	94.8 MB → 107 MB
⚠️ +12.7%	llama_3_2_1B_instruct_q40ef16_541 _{RSS @ ready · cuda}	`i9-11900kb_rtx-4060`	538 MB → 607 MB
⚠️ +12.6%	trunet _{RSS @ load · pulse1_f32}	`cortex-a53`	25.2 MB → 28.4 MB
⚠️ +12.5%	hey_snips_v4_model17 _{RSS @ ready · pulse8}	`cortex-a9`	26.9 MB → 30.2 MB
⚠️ +12.4%	mdl_en_2019_Q3_librispeech_onnx _{RSS @ load · pulse_240ms}	`cortex-a53`	32.5 MB → 36.5 MB
⚠️ +12.3%	llama_3_2_1B_instruct_q40ef16_541 _{RSS @ ready · cpu}	`i9-11900kb_rtx-4060`	728 MB → 817 MB
⚠️ +12.3%	llama_3_2_1B_q40ef32_516 _{RSS @ ready · cpu}	`i9-11900kb_rtx-4060`	728 MB → 817 MB
⚠️ +12.2%	hey_snips_v31 _{RSS @ ready · 400ms}	`cortex-a7`	20 MB → 22.4 MB
⚠️ +12.1%	hey_snips_v1 _{RSS @ ready · 400ms}	`cortex-a53`	21.9 MB → 24.6 MB
⚠️ +12.1%	llama_3_2_1B_instruct_q40ef16_541 _{RSS @ ready · cpu_mt}	`i9-11900kb_rtx-4060`	727 MB → 815 MB
⚠️ +12.0%	parakeet_tdt_600m_v3_f32f32_decoder_pass _{RSS @ ready · cpu}	`i9-11900kb_rtx-4060`	87.2 MB → 97.6 MB
⚠️ +11.9%	llama_3_2_1B_q40ef32_516 _{RSS @ ready · cuda}	`i9-11900kb_rtx-4060`	539 MB → 603 MB
⚠️ +11.8%	llama_3_1_8B_instruct_q40ef16_541 _{RSS @ ready · cuda}	`i9-11900kb_rtx-4060`	690 MB → 771 MB
⚠️ +11.7%	voicecom_float _{RSS @ load · 2sec}	`cortex-a53`	23.5 MB → 26.2 MB
⚠️ +11.5%	arm_ml_kws_cnn_m _{RSS @ load · pass}	`cortex-a9`	18 MB → 20 MB
⚠️ +11.3%	voicecom_fake_quant _{RSS @ ready · 2sec}	`cortex-a55`	33 MB → 36.7 MB
⚠️ +11.3%	hey_snips_v4_model17_nnef _{RSS @ load · pulse8}	`cortex-a7`	17.8 MB → 19.8 MB
⚠️ +11.0%	arm_ml_kws_cnn_m _{RSS @ load · pass}	`cortex-a7`	18.1 MB → 20.1 MB
⚠️ +11.0%	llama_3_2_1B_q40ef32_516 _{RSS @ load · cuda}	`jetson-orin-nx`	730 MB → 810 MB
⚠️ +10.8%	hey_snips_v4_model17_nnef _{RSS @ ready · pulse8}	`cortex-a55`	24.6 MB → 27.3 MB
⚠️ +10.6%	hey_snips_v4_model17_nnef _{RSS @ ready · pulse8}	`cortex-a53`	24.5 MB → 27.1 MB
⚠️ +10.6%	trunet _{RSS @ ready · pulse1_f16}	`cortex-a55`	26 MB → 28.8 MB
⚠️ +10.5%	speaker_id _{RSS @ load · pulse8}	`cortex-a53`	19.5 MB → 21.6 MB
⚠️ +10.5%	llama_3_2_1B_instruct_q40ef16_541 _{RSS @ load · cuda}	`jetson-orin-nx`	728 MB → 805 MB
⚠️ +10.3%	voicecom_fake_quant _{RSS @ ready · 2sec}	`cortex-a53`	33 MB → 36.4 MB
⚠️ +10.2%	trunet _{RSS @ load · pulse1_f16}	`cortex-a53`	25.8 MB → 28.5 MB
⚠️ +10.1%	dummy_conmer_12M _{RSS @ load · pulse_120ms}	`cortex-a7`	73.7 MB → 81.1 MB
⚠️ +10.0%	llama_3_2_1B_q40ef32_516 _{RSS @ load · cuda}	`i9-11900kb_rtx-4060`	732 MB → 806 MB
⚠️ +10.0%	llama_3_2_1B_q40ef32_516 _{RSS @ load · cpu}	`i9-11900kb_rtx-4060`	732 MB → 805 MB
⚠️ +10.0%	speaker_id _{RSS @ load · pulse8}	`cortex-a55`	19.7 MB → 21.6 MB
⚠️ +9.9%	llama_3_2_1B_instruct_q40ef16_541 _{RSS @ load · cuda}	`i9-11900kb_rtx-4060`	731 MB → 804 MB
⚠️ +9.8%	en_tdnn_15M _{load · 2600ms}	`cortex-a53`	1.46 s → 1.61 s
⚠️ +9.8%	llama_3_2_1B_instruct_q40ef16_541 _{RSS @ load · cpu_mt}	`i9-11900kb_rtx-4060`	731 MB → 803 MB
⚠️ +9.8%	parakeet_tdt_600m_v3_f32f32_decoder_pass _{RSS @ ready · cuda}	`jetson-orin-nx`	219 MB → 241 MB
⚠️ +9.8%	trunet _{RSS @ ready · pulse1_f32}	`cortex-a53`	27.4 MB → 30.1 MB
⚠️ +9.8%	trunet _{RSS @ ready · pulse1_f32}	`cortex-a55`	27.5 MB → 30.2 MB
⚠️ +9.8%	en_tdnn_15M _{RSS @ ready · pulse_120ms}	`cortex-a53`	111 MB → 122 MB
⚠️ +9.7%	trunet _{RSS @ load · pulse1_f32}	`cortex-a55`	24.6 MB → 27 MB
⚠️ +9.7%	llama_3_2_1B_instruct_q40ef16_541 _{RSS @ load · cpu}	`i9-11900kb_rtx-4060`	732 MB → 803 MB
⚠️ +9.5%	en_tdnn_15M _{RSS @ ready · 2600ms}	`cortex-a53`	110 MB → 121 MB
⚠️ +9.2%	nemotron_3_5_asr_streaming_0_6b_f32f32_decoder_pass _{RSS @ ready · cpu}	`i9-11900kb_rtx-4060`	100 MB → 109 MB
⚠️ +9.1%	dummy_conmer_12M _{RSS @ load · pulse_120ms}	`cortex-a9`	73.6 MB → 80.3 MB
⚠️ +9.0%	inceptionv1q _{RSS @ ready · pass}	`cortex-a7`	30.2 MB → 33 MB
⚠️ +9.0%	llama_3_1_8B_instruct_q40ef16_541 _{RSS @ ready · cuda}	`jetson-orin-nx`	508 MB → 554 MB
⚠️ +8.5%	dummy_conmer_12M _{RSS @ load · pulse_120ms}	`cortex-a55`	79.4 MB → 86.1 MB
⚠️ +8.1%	hey_snips_v31 _{RSS @ ready · 400ms}	`cortex-a53`	21.1 MB → 22.8 MB
⚠️ +8.0%	en_tdnn_15M _{load · 2600ms}	`cortex-a9`	2.5 s → 2.7 s
⚠️ +7.9%	dummy_conmer_12M _{RSS @ load · pulse_120ms}	`cortex-a53`	79.7 MB → 86 MB
⚠️ +7.8%	hey_snips_v4_model17 _{RSS @ ready · pulse8}	`cortex-a7`	26.6 MB → 28.7 MB
⚠️ +7.2%	inceptionv1q _{RSS @ ready · pass}	`cortex-a53`	41.3 MB → 44.3 MB
⚠️ +7.2%	en_tdnn_15M _{load · pulse_240ms}	`cortex-a53`	1.71 s → 1.83 s
⚠️ +7.1%	hey_snips_v31 _{RSS @ ready · 400ms}	`cortex-a55`	21.3 MB → 22.8 MB
⚠️ +7.1%	en_tdnn_15M_nnef _{RSS @ ready · pulse_240ms}	`cortex-a7`	89 MB → 95.3 MB
⚠️ +6.9%	voicecom_float _{RSS @ ready · 2sec}	`cortex-a55`	31.5 MB → 33.7 MB
⚠️ +6.7%	llama_3_2_3B_instruct_q40ef16_541 _{RSS @ load · cuda}	`i9-11900kb_rtx-4060`	1.85 GB → 1.98 GB
⚠️ +6.7%	en_tdnn_15M_nnef _{RSS @ ready · pulse_240ms}	`cortex-a9`	89.4 MB → 95.4 MB
⚠️ +6.5%	en_tdnn_15M _{load · pulse_120ms}	`cortex-a53`	1.72 s → 1.83 s
⚠️ +6.5%	llama_3_2_3B_q40ef32_516 _{RSS @ load · cuda}	`i9-11900kb_rtx-4060`	1.85 GB → 1.98 GB
⚠️ +6.4%	en_tdnn_15M _{load+optimize · 2600ms}	`cortex-a53`	1.78 s → 1.89 s
⚠️ +6.2%	mobilenet_v2_1 _{RSS @ load · pass}	`cortex-a53`	67.5 MB → 71.7 MB
⚠️ +6.1%	mobilenet_v2_1 _{RSS @ load · pass_mt}	`cortex-a53`	67.5 MB → 71.6 MB
⚠️ +6.0%	arm_ml_kws_cnn_m _{RSS @ ready · pass}	`cortex-a53`	23.6 MB → 25.1 MB
⚠️ +6.0%	qwen3_1_7B_q40ef16_541 _{RSS @ load · cuda}	`i9-11900kb_rtx-4060`	1.02 GB → 1.08 GB
⚠️ +5.8%	llama_3_2_3B_instruct_q40ef16_541 _{RSS @ load · cuda}	`jetson-orin-nx`	1.85 GB → 1.95 GB
⚠️ +5.8%	en_tdnn_15M _{load · pulse_120ms}	`cortex-a9`	2.87 s → 3.04 s
⚠️ +5.8%	voicecom_float _{RSS @ ready · 2sec}	`cortex-a53`	31.5 MB → 33.3 MB
⚠️ +5.7%	qwen3_1_7B_q40ef16_541 _{RSS @ load · cuda}	`jetson-orin-nx`	1.01 GB → 1.07 GB
⚠️ +5.6%	arm_ml_kws_cnn_m _{RSS @ ready · pass}	`cortex-a55`	23.4 MB → 24.8 MB
⚠️ +5.3%	en_tdnn_8M_nnef _{RSS @ load · pulse_240ms}	`orangepi-rv2`	46.4 MB → 48.9 MB
⚠️ +5.3%	llama_3_2_3B_q40ef32_516 _{RSS @ load · cuda}	`jetson-orin-nx`	1.85 GB → 1.94 GB
⚠️ +5.2%	arm_ml_kws_cnn_m _{RSS @ load · pass}	`apple-m1-max`	18.6 MB → 19.5 MB
⚠️ +5.2%	en_tdnn_15M _{load · pulse_240ms}	`cortex-a9`	2.87 s → 3.02 s
⚠️ +5.1%	hey_snips_v1 _{RSS @ load · 400ms}	`apple-m1-max`	18.9 MB → 19.8 MB

github-actions · 2026-09-07T12:07:22Z

🔴 Bench vs main — 1 speed regression(s) · ⚠️ 427 secondary

Reference: 2026-09-07 morning nightly run (0d old) · full report → run

Speed — evaltime · prefill · decode

Δ	metric	device	main → PR
🔴 +5.3%	hey_snips_v4_model17 _{evaltime · 2sec}	`cortex-a53`	16.1 ms → 17 ms

Improvements

Δ	metric	device	main → PR
🟢 -21.8%	speaker_id _{evaltime · pulse8}	`beaglev-ahead`	3.16 ms/pulse _{0.0394 RTF} → 2.47 ms/pulse _{0.0309 RTF}
🟢 -19.8%	hey_snips_v4_model17 _{evaltime · pulse8}	`beaglev-ahead`	5.88 ms/pulse _{0.0735 RTF} → 4.72 ms/pulse _{0.059 RTF}
🟢 -16.4%	mobilenet_v2_1 _{evaltime · pass_mt}	`beaglev-ahead`	504 ms → 421 ms
🟢 -14.9%	hey_snips_v4_model17 _{evaltime · 2sec}	`beaglev-ahead`	48.8 ms → 41.6 ms
🟢 -12.6%	trunet _{evaltime · pulse1_f32}	`beaglev-ahead`	5.69 ms/pulse _{0.569 RTF} → 4.98 ms/pulse _{0.498 RTF}

+15 more improvement(s)

Δ	metric	device	main → PR
🟢 -9.4%	inceptionv3 _{evaltime · pass_mt}	`beaglev-ahead`	2.27 s → 2.05 s
🟢 -9.0%	hey_snips_v4_model17_nnef _{evaltime · pulse8}	`cortex-a9`	6.3 ms/pulse _{0.0787 RTF} → 5.73 ms/pulse _{0.0716 RTF}
🟢 -8.7%	hey_snips_v4_model17_nnef _{evaltime · pulse8}	`beaglev-ahead`	5.24 ms/pulse _{0.0655 RTF} → 4.78 ms/pulse _{0.0598 RTF}
🟢 -8.7%	hey_snips_v4_model17 _{evaltime · pulse8}	`cortex-a9`	6.32 ms/pulse _{0.079 RTF} → 5.78 ms/pulse _{0.0722 RTF}
🟢 -8.6%	trunet _{evaltime · pulse1_f32}	`cortex-a7`	8.07 ms/pulse _{0.807 RTF} → 7.38 ms/pulse _{0.738 RTF}
🟢 -8.5%	hey_snips_v4_model17 _{evaltime · pulse8}	`cortex-a55`	2.1 ms/pulse _{0.0262 RTF} → 1.92 ms/pulse _{0.024 RTF}
🟢 -8.4%	hey_snips_v4_model17 _{evaltime · pulse8}	`cortex-a7`	6.96 ms/pulse _{0.0869 RTF} → 6.37 ms/pulse _{0.0797 RTF}
🟢 -8.1%	hey_snips_v31 _{evaltime · 400ms}	`beaglev-ahead`	6.17 ms → 5.68 ms
🟢 -7.8%	hey_snips_v4_model17_nnef _{evaltime · pulse8}	`cortex-a7`	6.92 ms/pulse _{0.0865 RTF} → 6.38 ms/pulse _{0.0797 RTF}
🟢 -7.6%	mobilenet_v2_1 _{evaltime · pass}	`beaglev-ahead`	1.15 s → 1.06 s
🟢 -7.0%	trunet _{evaltime · pulse1_f32}	`cortex-a9`	8.1 ms/pulse _{0.81 RTF} → 7.54 ms/pulse _{0.754 RTF}
🟢 -6.1%	trunet _{evaltime · pulse1_f32}	`i9-11900kb_rtx-4060`	0.22 ms/pulse _{0.022 RTF} → 0.207 ms/pulse _{0.0207 RTF}
🟢 -5.6%	hey_snips_v4_model17 _{evaltime · pulse8}	`cortex-a53`	2.73 ms/pulse _{0.0342 RTF} → 2.58 ms/pulse _{0.0322 RTF}
🟢 -5.1%	mdl_en_2019_Q3_librispeech_onnx _{evaltime · 2600ms}	`cortex-a9`	343 ms → 325 ms
🟢 -3.2%	llama_3_2_1B_instruct_q40ef16_541 _{prefill · cpu_mt}	`i9-11900kb_rtx-4060`	16.1 ms/tok _{62.13 tok/s} → 15.6 ms/tok _{64.17 tok/s}

⚠️ 427 secondary regression(s)

Δ	metric	device	main → PR
⚠️ +289.0%	en_tdnn_15M _{RSS @ load · pulse_120ms}	`cortex-a55`	94.8 MB → 369 MB
⚠️ +288.0%	en_tdnn_15M _{RSS @ load · 2600ms}	`cortex-a55`	94.9 MB → 368 MB
⚠️ +287.1%	en_tdnn_15M _{RSS @ load · pulse_240ms}	`cortex-a55`	95.1 MB → 368 MB
⚠️ +246.5%	en_tdnn_15M _{RSS @ ready · 2600ms}	`cortex-a55`	110 MB → 383 MB
⚠️ +244.4%	en_tdnn_15M _{RSS @ ready · pulse_120ms}	`cortex-a55`	111 MB → 384 MB
⚠️ +241.9%	en_tdnn_15M _{RSS @ ready · pulse_240ms}	`cortex-a55`	112 MB → 384 MB
⚠️ +239.9%	inceptionv3 _{RSS @ load · pass_mt}	`cortex-a55`	143 MB → 487 MB
⚠️ +239.4%	inceptionv3 _{RSS @ load · pass}	`cortex-a55`	144 MB → 487 MB
⚠️ +221.5%	inceptionv3 _{RSS @ ready · pass}	`cortex-a55`	152 MB → 489 MB
⚠️ +217.8%	inceptionv3 _{RSS @ ready · pass_mt}	`cortex-a55`	154 MB → 489 MB
⚠️ +190.8%	mobilenet_v2_1 _{RSS @ ready · pass}	`cortex-a55`	52.9 MB → 154 MB
⚠️ +189.5%	mobilenet_v2_1 _{RSS @ ready · pass_mt}	`cortex-a55`	53.1 MB → 154 MB
⚠️ +165.3%	en_tdnn_8M _{RSS @ load · pulse_240ms}	`cortex-a55`	54.3 MB → 144 MB
⚠️ +164.2%	en_tdnn_8M _{RSS @ load · 2600ms}	`cortex-a55`	53.8 MB → 142 MB
⚠️ +164.2%	mobilenet_v1_1 _{RSS @ load · pass}	`cortex-a55`	48 MB → 127 MB
⚠️ +164.0%	en_tdnn_8M _{RSS @ load · pulse_120ms}	`cortex-a55`	54.4 MB → 144 MB
⚠️ +163.6%	en_tdnn_8M _{RSS @ load · pulse_180ms}	`cortex-a55`	54.5 MB → 144 MB
⚠️ +139.5%	mobilenet_v1_1 _{RSS @ ready · pass}	`cortex-a55`	53.6 MB → 128 MB
⚠️ +131.3%	en_tdnn_8M _{RSS @ ready · 2600ms}	`cortex-a55`	62.2 MB → 144 MB
⚠️ +129.8%	en_tdnn_8M _{RSS @ ready · pulse_120ms}	`cortex-a55`	63.3 MB → 145 MB
⚠️ +128.4%	en_tdnn_8M _{RSS @ ready · pulse_240ms}	`cortex-a55`	64 MB → 146 MB
⚠️ +126.6%	en_tdnn_8M _{RSS @ ready · pulse_180ms}	`cortex-a55`	64.1 MB → 145 MB
⚠️ +124.8%	mobilenet_v2_1 _{RSS @ load · pass}	`cortex-a55`	67.6 MB → 152 MB
⚠️ +124.1%	mobilenet_v2_1 _{RSS @ load · pass_mt}	`cortex-a55`	67.7 MB → 152 MB
⚠️ +102.7%	hey_snips_v4_model17 _{RSS @ load · 2sec}	`beaglev-ahead`	16.9 MB → 34.3 MB
⚠️ +96.8%	mdl_en_2019_Q3_librispeech_onnx _{RSS @ load · 2600ms}	`cortex-a55`	32.3 MB → 63.5 MB
⚠️ +93.8%	mdl_en_2019_Q3_librispeech_onnx _{RSS @ load · pulse_240ms}	`cortex-a55`	32.5 MB → 63.1 MB
⚠️ +85.6%	mdl_en_2019_Q3_librispeech_onnx _{RSS @ ready · 2600ms}	`cortex-a55`	35.1 MB → 65.2 MB
⚠️ +81.3%	hey_snips_v4_model17 _{RSS @ load · 2sec}	`orangepi-rv2`	16.9 MB → 30.6 MB
⚠️ +79.7%	mdl_en_2019_Q3_librispeech_onnx _{RSS @ ready · pulse_240ms}	`cortex-a55`	36.1 MB → 64.9 MB
⚠️ +75.0%	hey_snips_v4_model17 _{RSS @ load · pulse8}	`orangepi-rv2`	17.7 MB → 31 MB
⚠️ +73.9%	hey_snips_v4_model17 _{RSS @ ready · 2sec}	`orangepi-rv2`	18.2 MB → 31.7 MB
⚠️ +71.4%	hey_snips_v4_model17 _{RSS @ ready · 2sec}	`beaglev-ahead`	18.5 MB → 31.7 MB
⚠️ +71.2%	hey_snips_v4_model17 _{RSS @ load · pulse8}	`beaglev-ahead`	17.8 MB → 30.4 MB
⚠️ +69.8%	voicecom_fake_quant _{RSS @ ready · 2sec}	`beaglev-ahead`	18.3 MB → 31 MB
⚠️ +68.8%	hey_snips_v4_model17 _{RSS @ ready · pulse8}	`orangepi-rv2`	18.9 MB → 31.9 MB
⚠️ +68.4%	mobilenet_v1_1 _{RSS @ load · pass}	`orangepi-rv2`	31.1 MB → 52.4 MB
⚠️ +68.2%	voicecom_fake_quant _{RSS @ ready · 2sec}	`orangepi-rv2`	18.6 MB → 31.2 MB
⚠️ +68.0%	mobilenet_v1_1 _{RSS @ load · pass}	`beaglev-ahead`	31.2 MB → 52.5 MB
⚠️ +63.1%	mobilenet_v1_1 _{RSS @ ready · pass}	`beaglev-ahead`	33 MB → 53.8 MB
⚠️ +62.6%	mobilenet_v1_1 _{RSS @ ready · pass}	`orangepi-rv2`	32.9 MB → 53.6 MB
⚠️ +62.3%	inceptionv1q _{RSS @ ready · pass}	`beaglev-ahead`	26.2 MB → 42.5 MB
⚠️ +61.9%	mobilenet_v2_1 _{RSS @ ready · pass_mt}	`orangepi-rv2`	40.9 MB → 66.2 MB
⚠️ +61.7%	mobilenet_v2_1 _{RSS @ ready · pass}	`orangepi-rv2`	40.9 MB → 66.2 MB
⚠️ +61.1%	mobilenet_v2_1 _{RSS @ ready · pass}	`beaglev-ahead`	41 MB → 66 MB
⚠️ +60.8%	mobilenet_v2_1 _{RSS @ ready · pass_mt}	`beaglev-ahead`	41 MB → 66 MB
⚠️ +60.6%	voicecom_float _{RSS @ ready · 2sec}	`beaglev-ahead`	17.7 MB → 28.4 MB
⚠️ +60.3%	hey_snips_v4_model17 _{RSS @ ready · pulse8}	`beaglev-ahead`	19 MB → 30.4 MB
⚠️ +58.0%	voicecom_float _{RSS @ ready · 2sec}	`orangepi-rv2`	18 MB → 28.4 MB
⚠️ +55.6%	mobilenet_v2_1 _{RSS @ ready · pass}	`cortex-a7`	46 MB → 71.5 MB
⚠️ +55.0%	en_tdnn_8M_nnef _{RSS @ ready · pulse_240ms}	`i9-11900kb_rtx-4060`	59.5 MB → 92.2 MB
⚠️ +54.8%	inceptionv1q _{RSS @ ready · pass}	`orangepi-rv2`	26.2 MB → 40.6 MB
⚠️ +53.7%	mobilenet_v1_1 _{RSS @ ready · pass}	`cortex-a9`	36.6 MB → 56.3 MB
⚠️ +53.5%	en_tdnn_8M _{RSS @ ready · 2600ms}	`cortex-a7`	52.7 MB → 80.8 MB
⚠️ +53.4%	hey_snips_v4_model17 _{RSS @ load · 2sec}	`i9-11900kb_rtx-4060`	33.8 MB → 51.9 MB
⚠️ +53.1%	en_tdnn_8M _{RSS @ ready · pulse_120ms}	`cortex-a9`	53.7 MB → 82.2 MB
⚠️ +52.9%	voicecom_fake_quant _{RSS @ ready · 2sec}	`cortex-a7`	22.2 MB → 33.9 MB
⚠️ +52.5%	mobilenet_v2_1 _{RSS @ ready · pass}	`cortex-a53`	52.8 MB → 80.5 MB
⚠️ +52.1%	en_tdnn_8M _{RSS @ ready · pulse_180ms}	`cortex-a7`	54 MB → 82.2 MB
⚠️ +51.9%	mobilenet_v2_1 _{RSS @ load · pass_mt}	`cortex-a9`	42.5 MB → 64.6 MB
⚠️ +51.4%	mobilenet_v2_1 _{RSS @ ready · pass_mt}	`cortex-a53`	53 MB → 80.3 MB
⚠️ +50.9%	voicecom_fake_quant _{RSS @ ready · 2sec}	`cortex-a9`	22.3 MB → 33.7 MB
⚠️ +50.6%	en_tdnn_8M _{RSS @ ready · pulse_120ms}	`cortex-a7`	53.8 MB → 81.1 MB
⚠️ +50.0%	voicecom_float _{RSS @ ready · 2sec}	`cortex-a9`	20.8 MB → 31.2 MB
⚠️ +49.5%	inceptionv1q _{RSS @ load · pass}	`orangepi-rv2`	20.2 MB → 30.3 MB
⚠️ +49.4%	mobilenet_v2_1 _{RSS @ ready · pass_mt}	`cortex-a7`	45.7 MB → 68.3 MB
⚠️ +49.4%	en_tdnn_8M_nnef _{RSS @ ready · pulse_240ms}	`orangepi-rv2`	47.5 MB → 70.9 MB
⚠️ +49.0%	inceptionv1q _{RSS @ load · pass}	`beaglev-ahead`	20.2 MB → 30.1 MB
⚠️ +48.8%	mdl_en_2019_Q3_librispeech_onnx _{RSS @ load · 2600ms}	`i9-11900kb_rtx-4060`	36 MB → 53.5 MB
⚠️ +48.7%	voicecom_float _{RSS @ ready · 2sec}	`cortex-a7`	20.9 MB → 31 MB
⚠️ +48.7%	en_tdnn_8M_nnef _{RSS @ ready · pulse_240ms}	`beaglev-ahead`	47.8 MB → 71 MB
⚠️ +48.1%	mobilenet_v2_1 _{RSS @ load · pass}	`orangepi-rv2`	38.7 MB → 57.4 MB
⚠️ +48.1%	mobilenet_v2_1 _{RSS @ load · pass_mt}	`orangepi-rv2`	38.8 MB → 57.4 MB
⚠️ +48.0%	mobilenet_v2_1 _{RSS @ load · pass}	`beaglev-ahead`	38.8 MB → 57.4 MB
⚠️ +48.0%	mobilenet_v2_1 _{RSS @ load · pass_mt}	`beaglev-ahead`	38.8 MB → 57.4 MB
⚠️ +46.9%	en_tdnn_8M _{RSS @ ready · pulse_180ms}	`i9-11900kb_rtx-4060`	72.6 MB → 107 MB
⚠️ +46.3%	en_tdnn_8M _{RSS @ ready · pulse_120ms}	`orangepi-rv2`	49.8 MB → 72.8 MB
⚠️ +46.1%	mobilenet_v2_1 _{RSS @ load · pass}	`cortex-a7`	43 MB → 62.8 MB
⚠️ +46.1%	hey_snips_v4_model17 _{RSS @ load · 2sec}	`cortex-a55`	28.1 MB → 41.1 MB
⚠️ +46.1%	mobilenet_v1_1 _{RSS @ load · pass}	`cortex-a9`	35 MB → 51.1 MB
⚠️ +45.8%	hey_snips_v4_model17 _{RSS @ ready · 2sec}	`i9-11900kb_rtx-4060`	36.9 MB → 53.8 MB
⚠️ +45.3%	mdl_en_2019_Q3_librispeech_onnx _{RSS @ load · pulse_240ms}	`i9-11900kb_rtx-4060`	37.6 MB → 54.6 MB
⚠️ +45.1%	en_tdnn_8M _{RSS @ ready · pulse_120ms}	`beaglev-ahead`	50.1 MB → 72.7 MB
⚠️ +45.1%	en_tdnn_8M _{RSS @ ready · pulse_240ms}	`cortex-a9`	53.8 MB → 78.1 MB
⚠️ +44.9%	en_tdnn_8M _{RSS @ ready · pulse_240ms}	`beaglev-ahead`	50.1 MB → 72.6 MB
⚠️ +44.9%	mobilenet_v2_1 _{RSS @ ready · pass_mt}	`cortex-a9`	45.7 MB → 66.2 MB
⚠️ +44.8%	en_tdnn_8M _{RSS @ ready · pulse_180ms}	`cortex-a9`	54.1 MB → 78.4 MB
⚠️ +44.5%	voicecom_float _{RSS @ load · 2sec}	`orangepi-rv2`	16.3 MB → 23.6 MB
⚠️ +44.0%	en_tdnn_8M _{RSS @ ready · 2600ms}	`i9-11900kb_rtx-4060`	73.3 MB → 105 MB
⚠️ +43.8%	en_tdnn_8M _{RSS @ ready · pulse_240ms}	`cortex-a7`	54.2 MB → 77.9 MB
⚠️ +43.7%	en_tdnn_8M _{RSS @ ready · pulse_240ms}	`i9-11900kb_rtx-4060`	73.8 MB → 106 MB
⚠️ +43.1%	voicecom_float _{RSS @ load · 2sec}	`beaglev-ahead`	16.3 MB → 23.4 MB
⚠️ +41.6%	voicecom_fake_quant _{RSS @ load · 2sec}	`orangepi-rv2`	17.1 MB → 24.2 MB
⚠️ +40.7%	en_tdnn_8M _{RSS @ ready · pulse_120ms}	`i9-11900kb_rtx-4060`	75.5 MB → 106 MB
⚠️ +40.7%	hey_snips_v4_model17 _{RSS @ load · pulse8}	`i9-11900kb_rtx-4060`	36.1 MB → 50.8 MB
⚠️ +40.6%	voicecom_fake_quant _{RSS @ load · 2sec}	`cortex-a7`	19.2 MB → 27.1 MB
⚠️ +40.5%	inceptionv1q _{RSS @ ready · pass}	`i9-11900kb_rtx-4060`	43.4 MB → 61 MB
⚠️ +40.3%	parakeet_tdt_600m_v3_f32f32_decoder_pass _{RSS @ load · cuda}	`i9-11900kb_rtx-4060`	69.2 MB → 97.1 MB
⚠️ +40.1%	en_tdnn_pyt_15M _{RSS @ load · pulse_120ms}	`orangepi-rv2`	60.3 MB → 84.4 MB
⚠️ +40.0%	voicecom_fake_quant _{RSS @ load · 2sec}	`i9-11900kb_rtx-4060`	27.1 MB → 37.9 MB
⚠️ +39.9%	en_tdnn_8M _{RSS @ ready · 2600ms}	`cortex-a9`	52.8 MB → 73.9 MB
⚠️ +39.9%	parakeet_tdt_600m_v3_f32f32_decoder_pass _{RSS @ load · cpu}	`i9-11900kb_rtx-4060`	69.4 MB → 97.1 MB
⚠️ +39.6%	voicecom_fake_quant _{RSS @ load · 2sec}	`cortex-a9`	19.3 MB → 27 MB
⚠️ +39.4%	mobilenet_v2_1 _{RSS @ load · pass_mt}	`cortex-a7`	42.6 MB → 59.4 MB
⚠️ +39.4%	hey_snips_v4_model17 _{RSS @ ready · pulse8}	`cortex-a55`	31.5 MB → 43.9 MB
⚠️ +39.2%	hey_snips_v4_model17_nnef _{RSS @ ready · pulse8}	`orangepi-rv2`	16.4 MB → 22.8 MB
⚠️ +38.0%	hey_snips_v4_model17 _{RSS @ ready · pulse8}	`i9-11900kb_rtx-4060`	38 MB → 52.4 MB
⚠️ +37.8%	en_tdnn_pyt_15M _{RSS @ load · pulse_120ms}	`beaglev-ahead`	60.3 MB → 83.1 MB
⚠️ +37.7%	hey_snips_v4_model17_nnef _{RSS @ ready · pulse8}	`beaglev-ahead`	16.7 MB → 23 MB
⚠️ +37.3%	en_tdnn_8M _{RSS @ load · pulse_180ms}	`i9-11900kb_rtx-4060`	64.4 MB → 88.5 MB
⚠️ +37.0%	openelm_270M_q40ef16_541 _{RSS @ ready · cpu}	`i9-11900kb_rtx-4060`	192 MB → 263 MB
⚠️ +37.0%	nemotron_3_5_asr_streaming_0_6b_f32f32_decoder_pass _{RSS @ load · cuda}	`jetson-orin-nx`	75.7 MB → 104 MB
⚠️ +36.9%	inceptionv3 _{RSS @ ready · pass}	`beaglev-ahead`	119 MB → 163 MB
⚠️ +36.5%	en_tdnn_8M_nnef _{RSS @ ready · pulse_240ms}	`cortex-a53`	55.4 MB → 75.6 MB
⚠️ +36.5%	voicecom_float _{RSS @ load · 2sec}	`cortex-a9`	19.3 MB → 26.4 MB
⚠️ +36.3%	voicecom_fake_quant _{RSS @ load · 2sec}	`beaglev-ahead`	17.2 MB → 23.4 MB
⚠️ +35.9%	inceptionv1q _{RSS @ load · pass}	`i9-11900kb_rtx-4060`	30.7 MB → 41.7 MB
⚠️ +35.9%	mobilenet_v2_1 _{RSS @ ready · pass}	`cortex-a9`	45.9 MB → 62.3 MB
⚠️ +35.9%	mdl_en_2019_Q3_librispeech_onnx _{RSS @ load · pulse_240ms}	`cortex-a7`	29.4 MB → 40 MB
⚠️ +35.9%	trunet _{RSS @ ready · pulse1_f16}	`orangepi-rv2`	18.5 MB → 25.2 MB
⚠️ +35.8%	en_tdnn_8M_nnef _{RSS @ ready · pulse_240ms}	`cortex-a55`	55.5 MB → 75.4 MB
⚠️ +35.4%	en_tdnn_8M _{RSS @ load · 2600ms}	`i9-11900kb_rtx-4060`	64.1 MB → 86.8 MB
⚠️ +35.4%	en_tdnn_pyt_15M _{RSS @ load · pulse_120ms}	`cortex-a7`	63.6 MB → 86.1 MB
⚠️ +35.4%	en_tdnn_8M _{RSS @ ready · 2600ms}	`orangepi-rv2`	49 MB → 66.3 MB
⚠️ +35.3%	hey_snips_v1 _{RSS @ ready · 400ms}	`orangepi-rv2`	14.9 MB → 20.2 MB
⚠️ +35.2%	inceptionv3 _{RSS @ ready · pass_mt}	`beaglev-ahead`	119 MB → 160 MB
⚠️ +35.1%	en_tdnn_8M _{RSS @ ready · 2600ms}	`beaglev-ahead`	49.3 MB → 66.6 MB
⚠️ +35.1%	voicecom_float _{RSS @ load · 2sec}	`i9-11900kb_rtx-4060`	26.8 MB → 36.2 MB
⚠️ +34.4%	parakeet_tdt_600m_v3_f32f32_decoder_pass _{RSS @ load · cuda}	`jetson-orin-nx`	67.2 MB → 90.2 MB
⚠️ +34.4%	en_tdnn_pyt_15M _{RSS @ load · pulse_120ms}	`cortex-a9`	64.2 MB → 86.2 MB
⚠️ +34.4%	trunet _{RSS @ load · pulse1_f16}	`orangepi-rv2`	17.9 MB → 24.1 MB
⚠️ +34.2%	inceptionv1q _{RSS @ ready · pass}	`cortex-a55`	40.6 MB → 54.5 MB
⚠️ +34.2%	voicecom_float _{RSS @ load · 2sec}	`cortex-a7`	19.5 MB → 26.1 MB
⚠️ +34.1%	mdl_en_2019_Q3_librispeech_onnx _{RSS @ ready · 2600ms}	`i9-11900kb_rtx-4060`	41.4 MB → 55.4 MB
⚠️ +34.0%	hey_snips_v1 _{RSS @ ready · 400ms}	`beaglev-ahead`	14.9 MB → 20 MB
⚠️ +33.9%	en_tdnn_8M _{RSS @ load · pulse_240ms}	`i9-11900kb_rtx-4060`	65.6 MB → 87.8 MB
⚠️ +33.8%	en_tdnn_15M _{RSS @ ready · pulse_120ms}	`orangepi-rv2`	88 MB → 118 MB
⚠️ +33.6%	en_tdnn_8M _{RSS @ load · pulse_120ms}	`i9-11900kb_rtx-4060`	65.8 MB → 87.9 MB
⚠️ +33.3%	en_tdnn_15M _{RSS @ ready · 2600ms}	`orangepi-rv2`	87.2 MB → 116 MB
⚠️ +33.1%	en_tdnn_15M _{RSS @ ready · 2600ms}	`beaglev-ahead`	87.6 MB → 117 MB
⚠️ +33.0%	en_tdnn_15M _{RSS @ ready · pulse_240ms}	`beaglev-ahead`	88.4 MB → 118 MB
⚠️ +33.0%	parakeet_tdt_600m_v3_f32f32_joint_pass _{RSS @ ready · cpu}	`i9-11900kb_rtx-4060`	51.3 MB → 68.3 MB
⚠️ +32.9%	en_tdnn_15M _{RSS @ ready · pulse_120ms}	`beaglev-ahead`	88.4 MB → 118 MB
⚠️ +32.4%	mdl_en_2019_Q3_librispeech_onnx _{RSS @ load · pulse_240ms}	`cortex-a9`	29.4 MB → 39 MB
⚠️ +32.3%	en_tdnn_15M _{RSS @ ready · pulse_240ms}	`orangepi-rv2`	88 MB → 116 MB
⚠️ +31.9%	mdl_en_2019_Q3_librispeech_onnx _{RSS @ ready · pulse_240ms}	`i9-11900kb_rtx-4060`	42.4 MB → 55.9 MB
⚠️ +31.9%	nemotron_3_5_asr_streaming_0_6b_f32f32_decoder_pass _{RSS @ load · cpu}	`i9-11900kb_rtx-4060`	81.6 MB → 108 MB
⚠️ +31.8%	mdl_en_2019_Q3_librispeech_onnx _{RSS @ load · 2600ms}	`cortex-a7`	29.2 MB → 38.5 MB
⚠️ +31.7%	trunet _{RSS @ load · pulse1_f16}	`beaglev-ahead`	18.1 MB → 23.8 MB
⚠️ +31.6%	en_tdnn_8M_nnef _{RSS @ ready · pulse_240ms}	`cortex-a7`	51 MB → 67.1 MB
⚠️ +31.6%	nemotron_3_5_asr_streaming_0_6b_f32f32_preprocessor_1s _{RSS @ load · cuda}	`jetson-orin-nx`	21 MB → 27.6 MB
⚠️ +31.5%	inceptionv1q _{RSS @ load · pass}	`cortex-a7`	23.9 MB → 31.4 MB
⚠️ +31.5%	qwen3_1_7B_q40ef16_541 _{RSS @ ready · cuda}	`jetson-orin-nx`	393 MB → 516 MB
⚠️ +31.3%	en_tdnn_15M_nnef _{RSS @ ready · pulse_240ms}	`cortex-a55`	91.5 MB → 120 MB
⚠️ +31.3%	nemotron_3_5_asr_streaming_0_6b_f32f32_decoder_pass _{RSS @ load · cuda}	`i9-11900kb_rtx-4060`	81.9 MB → 107 MB
⚠️ +31.1%	en_tdnn_lstm_bn_q7 _{RSS @ load · pulse_240ms}	`cortex-a7`	22.3 MB → 29.3 MB
⚠️ +30.6%	inceptionv3 _{RSS @ ready · pass_mt}	`cortex-a7`	127 MB → 165 MB
⚠️ +30.6%	en_tdnn_lstm_bn_q7 _{RSS @ load · pulse_240ms}	`cortex-a9`	22.4 MB → 29.3 MB
⚠️ +30.6%	inceptionv1q _{RSS @ load · pass}	`cortex-a9`	24.1 MB → 31.4 MB
⚠️ +30.5%	hey_snips_v4_model17_nnef _{RSS @ ready · pulse8}	`i9-11900kb_rtx-4060`	27.9 MB → 36.4 MB
⚠️ +30.5%	trunet _{RSS @ ready · pulse1_f32}	`beaglev-ahead`	20.4 MB → 26.6 MB
⚠️ +30.5%	en_tdnn_15M_nnef _{RSS @ ready · pulse_240ms}	`i9-11900kb_rtx-4060`	96 MB → 125 MB
⚠️ +30.3%	mobilenet_v1_1 _{RSS @ load · pass}	`cortex-a7`	35.2 MB → 45.9 MB
⚠️ +29.8%	mdl_en_2019_Q3_librispeech_onnx _{RSS @ load · 2600ms}	`cortex-a9`	29.4 MB → 38.1 MB
⚠️ +29.4%	en_tdnn_lstm_bn_q7 _{RSS @ load · 2600ms}	`cortex-a9`	22.1 MB → 28.6 MB
⚠️ +29.4%	en_tdnn_lstm_bn_q7 _{RSS @ load · 2600ms}	`cortex-a7`	21.9 MB → 28.4 MB
⚠️ +29.2%	parakeet_tdt_600m_v3_f32f32_preprocessor_1s _{RSS @ load · cuda}	`jetson-orin-nx`	21.3 MB → 27.5 MB
⚠️ +28.7%	hey_snips_v1 _{RSS @ ready · 400ms}	`i9-11900kb_rtx-4060`	25.7 MB → 33.1 MB
⚠️ +28.5%	en_tdnn_pyt_15M _{RSS @ load · pulse_120ms}	`cortex-a55`	74.2 MB → 95.4 MB
⚠️ +28.4%	mobilenet_v1_1 _{RSS @ ready · pass}	`cortex-a7`	36.9 MB → 47.4 MB
⚠️ +28.4%	hey_snips_v1 _{RSS @ ready · 400ms}	`cortex-a9`	18.5 MB → 23.8 MB
⚠️ +28.0%	en_tdnn_8M_nnef _{RSS @ ready · pulse_240ms}	`cortex-a9`	50.8 MB → 65 MB
⚠️ +27.9%	trunet _{RSS @ ready · pulse1_f16}	`beaglev-ahead`	18.6 MB → 23.8 MB
⚠️ +27.9%	trunet _{RSS @ load · pulse1_f16}	`i9-11900kb_rtx-4060`	27.6 MB → 35.3 MB
⚠️ +27.4%	inceptionv3 _{RSS @ ready · pass}	`cortex-a7`	126 MB → 161 MB
⚠️ +27.4%	en_tdnn_8M _{RSS @ load · pulse_120ms}	`cortex-a9`	51.4 MB → 65.4 MB
⚠️ +27.0%	en_tdnn_8M _{RSS @ load · pulse_120ms}	`orangepi-rv2`	49.1 MB → 62.3 MB
⚠️ +27.0%	dummy_conmer_12M _{RSS @ ready · pulse_120ms}	`beaglev-ahead`	69.2 MB → 87.8 MB
⚠️ +27.0%	en_tdnn_pyt_15M _{RSS @ load · pulse_120ms}	`i9-11900kb_rtx-4060`	85.6 MB → 109 MB
⚠️ +26.8%	inceptionv3 _{RSS @ ready · pass}	`orangepi-rv2`	118 MB → 150 MB
⚠️ +26.7%	mdl_en_2019_Q3_librispeech_onnx _{RSS @ ready · pulse_240ms}	`cortex-a7`	32.2 MB → 40.8 MB
⚠️ +26.6%	arm_ml_kws_cnn_m _{RSS @ ready · pass}	`i9-11900kb_rtx-4060`	26.5 MB → 33.5 MB
⚠️ +26.2%	en_tdnn_8M _{RSS @ load · pulse_180ms}	`cortex-a7`	51.6 MB → 65.1 MB
⚠️ +26.0%	mdl_en_2019_Q3_librispeech_onnx _{RSS @ ready · 2600ms}	`cortex-a7`	31.4 MB → 39.5 MB
⚠️ +26.0%	hey_snips_v31 _{RSS @ load · 400ms}	`i9-11900kb_rtx-4060`	23.6 MB → 29.8 MB
⚠️ +26.0%	trunet _{RSS @ ready · pulse1_f32}	`orangepi-rv2`	20.3 MB → 25.6 MB
⚠️ +26.0%	en_tdnn_pyt_15M _{RSS @ ready · pulse_120ms}	`orangepi-rv2`	61.4 MB → 77.3 MB
⚠️ +25.9%	arm_ml_kws_cnn_m _{RSS @ load · pass}	`i9-11900kb_rtx-4060`	23.4 MB → 29.5 MB
⚠️ +25.9%	speaker_id _{RSS @ load · pulse8}	`cortex-a9`	18.4 MB → 23.1 MB
⚠️ +25.9%	hey_snips_v4_model17 _{RSS @ ready · 2sec}	`cortex-a55`	32.6 MB → 41.1 MB
⚠️ +25.9%	inceptionv3 _{RSS @ load · pass_mt}	`beaglev-ahead`	114 MB → 143 MB
⚠️ +25.9%	inceptionv3 _{RSS @ load · pass}	`beaglev-ahead`	114 MB → 143 MB
⚠️ +25.8%	en_tdnn_pyt_15M _{RSS @ ready · pulse_120ms}	`beaglev-ahead`	61.6 MB → 77.5 MB
⚠️ +25.7%	mobilenet_v2_1 _{RSS @ load · pass}	`cortex-a9`	42.6 MB → 53.5 MB
⚠️ +25.7%	en_tdnn_lstm_bn_q7 _{RSS @ ready · 2600ms}	`cortex-a9`	24.1 MB → 30.3 MB
⚠️ +25.6%	en_tdnn_pyt_15M _{RSS @ ready · pulse_120ms}	`cortex-a55`	81.3 MB → 102 MB
⚠️ +25.5%	openelm_270M_q40ef16_541 _{RSS @ load · cuda}	`jetson-orin-nx`	200 MB → 251 MB
⚠️ +25.5%	inceptionv3 _{RSS @ load · pass}	`orangepi-rv2`	113 MB → 142 MB
⚠️ +25.5%	hey_snips_v1 _{RSS @ load · 400ms}	`i9-11900kb_rtx-4060`	22.6 MB → 28.3 MB
⚠️ +25.4%	inceptionv3 _{RSS @ ready · pass}	`cortex-a9`	127 MB → 159 MB
⚠️ +25.4%	en_tdnn_pyt_15M _{RSS @ ready · pulse_120ms}	`cortex-a7`	65.3 MB → 82 MB
⚠️ +25.4%	voicecom_fake_quant _{RSS @ load · 2sec}	`cortex-a53`	22.8 MB → 28.5 MB
⚠️ +25.4%	mdl_en_2019_Q3_librispeech_onnx _{RSS @ ready · 2600ms}	`cortex-a9`	31.3 MB → 39.2 MB
⚠️ +25.3%	en_tdnn_lstm_bn_q7 _{RSS @ load · 2600ms}	`i9-11900kb_rtx-4060`	28.1 MB → 35.2 MB
⚠️ +25.3%	nemotron_3_5_asr_streaming_0_6b_f32f32_joint_pass _{RSS @ ready · cpu}	`i9-11900kb_rtx-4060`	64.5 MB → 80.9 MB
⚠️ +25.2%	en_tdnn_lstm_bn_q7 _{RSS @ ready · 2600ms}	`cortex-a7`	24 MB → 30.1 MB
⚠️ +25.1%	en_tdnn_8M _{RSS @ load · pulse_120ms}	`beaglev-ahead`	49.1 MB → 61.4 MB
⚠️ +25.0%	trunet _{RSS @ load · pulse1_f32}	`beaglev-ahead`	19.2 MB → 24 MB
⚠️ +25.0%	en_tdnn_8M _{RSS @ load · pulse_240ms}	`beaglev-ahead`	49 MB → 61.3 MB
⚠️ +24.8%	parakeet_tdt_600m_v3_f32f32_preprocessor_1s _{RSS @ load · cpu}	`i9-11900kb_rtx-4060`	23.7 MB → 29.6 MB
⚠️ +24.7%	en_tdnn_8M _{RSS @ ready · pulse_180ms}	`orangepi-rv2`	49.8 MB → 62.1 MB
⚠️ +24.6%	en_tdnn_8M _{RSS @ ready · pulse_240ms}	`orangepi-rv2`	49.8 MB → 62 MB
⚠️ +24.6%	en_tdnn_lstm_bn_q7 _{RSS @ load · pulse_240ms}	`i9-11900kb_rtx-4060`	29 MB → 36.1 MB
⚠️ +24.6%	hey_snips_v4_model17_nnef _{RSS @ load · pulse8}	`orangepi-rv2`	14.5 MB → 18.1 MB
⚠️ +24.5%	arm_ml_kws_cnn_m _{RSS @ ready · pass}	`beaglev-ahead`	15.9 MB → 19.8 MB
⚠️ +24.4%	voicecom_fake_quant _{RSS @ load · 2sec}	`cortex-a55`	22.7 MB → 28.3 MB
⚠️ +24.3%	en_tdnn_8M _{RSS @ load · 2600ms}	`cortex-a7`	51.3 MB → 63.8 MB
⚠️ +24.2%	inceptionv1q _{RSS @ load · pass}	`cortex-a55`	27.2 MB → 33.8 MB
⚠️ +24.2%	en_tdnn_pyt_15M _{RSS @ ready · pulse_120ms}	`cortex-a9`	65.8 MB → 81.7 MB
⚠️ +24.1%	nemotron_3_5_asr_streaming_0_6b_f32f32_preprocessor_1s _{RSS @ load · cuda}	`i9-11900kb_rtx-4060`	23.8 MB → 29.5 MB
⚠️ +24.0%	nemotron_3_5_asr_streaming_0_6b_f32f32_preprocessor_1s _{RSS @ load · cpu}	`i9-11900kb_rtx-4060`	23.5 MB → 29.1 MB
⚠️ +23.9%	en_tdnn_8M _{RSS @ ready · pulse_180ms}	`beaglev-ahead`	50.1 MB → 62.1 MB
⚠️ +23.9%	en_tdnn_8M _{RSS @ load · pulse_240ms}	`cortex-a9`	51.9 MB → 64.2 MB
⚠️ +23.8%	parakeet_tdt_600m_v3_f32f32_preprocessor_1s _{RSS @ load · cuda}	`i9-11900kb_rtx-4060`	23.8 MB → 29.4 MB
⚠️ +23.8%	hey_snips_v1 _{RSS @ ready · 400ms}	`cortex-a7`	18.6 MB → 23 MB
⚠️ +23.8%	en_tdnn_8M _{RSS @ load · pulse_180ms}	`cortex-a9`	52.1 MB → 64.4 MB
⚠️ +23.7%	speaker_id _{RSS @ load · pulse8}	`i9-11900kb_rtx-4060`	24.4 MB → 30.2 MB
⚠️ +23.6%	mdl_en_2019_Q3_librispeech_onnx _{RSS @ ready · pulse_240ms}	`cortex-a9`	32.3 MB → 39.9 MB
⚠️ +23.6%	dummy_conmer_12M _{RSS @ ready · pulse_120ms}	`orangepi-rv2`	68.7 MB → 84.9 MB
⚠️ +23.6%	en_tdnn_15M _{RSS @ load · pulse_240ms}	`i9-11900kb_rtx-4060`	111 MB → 137 MB
⚠️ +23.4%	en_tdnn_8M _{RSS @ load · pulse_120ms}	`cortex-a7`	51.8 MB → 63.9 MB
⚠️ +23.4%	en_tdnn_8M _{RSS @ load · pulse_240ms}	`cortex-a7`	52 MB → 64.1 MB
⚠️ +23.4%	arm_ml_kws_cnn_m _{RSS @ ready · pass}	`orangepi-rv2`	16.5 MB → 20.3 MB
⚠️ +23.3%	inceptionv1q _{RSS @ load · pass}	`cortex-a53`	27.6 MB → 34 MB
⚠️ +23.0%	hey_snips_v4_model17_nnef _{RSS @ load · pulse8}	`beaglev-ahead`	14.6 MB → 17.9 MB
⚠️ +22.9%	en_tdnn_15M _{RSS @ load · pulse_120ms}	`i9-11900kb_rtx-4060`	111 MB → 136 MB
⚠️ +22.8%	nemotron_3_5_asr_streaming_0_6b_f32f32_preprocessor_pulse100ms _{RSS @ load · cpu}	`i9-11900kb_rtx-4060`	25.2 MB → 30.9 MB
⚠️ +22.8%	en_tdnn_lstm_bn_q7 _{RSS @ ready · pulse_240ms}	`cortex-a7`	24.9 MB → 30.5 MB
⚠️ +22.7%	en_tdnn_8M _{RSS @ load · 2600ms}	`cortex-a9`	51.6 MB → 63.3 MB
⚠️ +22.6%	en_tdnn_15M _{RSS @ load · 2600ms}	`i9-11900kb_rtx-4060`	111 MB → 136 MB
⚠️ +22.6%	qwen3_1_7B_q40ef16_541 _{RSS @ ready · cuda}	`i9-11900kb_rtx-4060`	575 MB → 705 MB
⚠️ +22.5%	speaker_id _{RSS @ load · pulse8}	`cortex-a7`	18.6 MB → 22.8 MB
⚠️ +22.4%	hey_snips_v4_model17 _{RSS @ load · pulse8}	`cortex-a9`	24.3 MB → 29.8 MB
⚠️ +22.2%	en_tdnn_lstm_bn_q7 _{RSS @ ready · pulse_240ms}	`cortex-a9`	25.2 MB → 30.8 MB
⚠️ +22.1%	hey_snips_v31 _{RSS @ load · 400ms}	`orangepi-rv2`	14.6 MB → 17.8 MB
⚠️ +22.0%	en_tdnn_15M _{RSS @ ready · 2600ms}	`cortex-a9`	91 MB → 111 MB
⚠️ +22.0%	en_tdnn_8M _{RSS @ load · pulse_240ms}	`orangepi-rv2`	49 MB → 59.8 MB
⚠️ +22.0%	trunet _{RSS @ load · pulse1_f32}	`orangepi-rv2`	19.2 MB → 23.5 MB
⚠️ +22.0%	en_tdnn_8M _{RSS @ load · pulse_180ms}	`orangepi-rv2`	49 MB → 59.8 MB
⚠️ +22.0%	inceptionv3 _{RSS @ load · pass_mt}	`orangepi-rv2`	113 MB → 138 MB
⚠️ +21.9%	trunet _{RSS @ ready · pulse1_f32}	`i9-11900kb_rtx-4060`	32.5 MB → 39.6 MB
⚠️ +21.9%	hey_snips_v4_model17 _{RSS @ load · 2sec}	`cortex-a9`	24.1 MB → 29.4 MB
⚠️ +21.8%	en_tdnn_8M _{RSS @ ready · pulse_120ms}	`cortex-a53`	63.9 MB → 77.8 MB
⚠️ +21.7%	nemotron_3_5_asr_streaming_0_6b_f32f32_preprocessor_1s _{RSS @ ready · cpu}	`i9-11900kb_rtx-4060`	25.3 MB → 30.8 MB
⚠️ +21.6%	en_tdnn_8M _{RSS @ ready · pulse_180ms}	`cortex-a53`	63.7 MB → 77.5 MB
⚠️ +21.6%	parakeet_tdt_600m_v3_f32f32_preprocessor_1s _{RSS @ ready · cpu}	`i9-11900kb_rtx-4060`	25.8 MB → 31.4 MB
⚠️ +21.6%	nemotron_3_5_asr_streaming_0_6b_f32f32_preprocessor_pulse100ms _{RSS @ ready · cpu}	`i9-11900kb_rtx-4060`	26.4 MB → 32.1 MB
⚠️ +21.6%	hey_snips_v31 _{RSS @ load · 400ms}	`beaglev-ahead`	14.6 MB → 17.8 MB
⚠️ +21.5%	hey_snips_v31 _{RSS @ ready · 400ms}	`i9-11900kb_rtx-4060`	26.1 MB → 31.7 MB
⚠️ +21.3%	speaker_id _{RSS @ ready · pulse8}	`i9-11900kb_rtx-4060`	26.5 MB → 32.1 MB
⚠️ +21.2%	voicecom_float _{RSS @ load · 2sec}	`cortex-a53`	23.5 MB → 28.5 MB
⚠️ +21.2%	trunet _{RSS @ ready · pulse1_f16}	`i9-11900kb_rtx-4060`	30.4 MB → 36.9 MB
⚠️ +21.2%	inceptionv3 _{RSS @ load · pass_mt}	`cortex-a7`	116 MB → 141 MB
⚠️ +21.0%	en_tdnn_8M _{RSS @ load · 2600ms}	`orangepi-rv2`	48.2 MB → 58.4 MB
⚠️ +21.0%	hey_snips_v31 _{RSS @ ready · 400ms}	`orangepi-rv2`	15.8 MB → 19.1 MB
⚠️ +20.9%	nemotron_3_5_asr_streaming_0_6b_f32f32_preprocessor_pulse100ms _{RSS @ load · cuda}	`i9-11900kb_rtx-4060`	25.5 MB → 30.8 MB
⚠️ +20.9%	en_tdnn_8M _{RSS @ load · 2600ms}	`beaglev-ahead`	48.2 MB → 58.3 MB
⚠️ +20.8%	hey_snips_v4_model17_nnef _{RSS @ load · pulse8}	`i9-11900kb_rtx-4060`	24.5 MB → 29.6 MB
⚠️ +20.7%	inceptionv3 _{RSS @ ready · pass_mt}	`orangepi-rv2`	118 MB → 143 MB
⚠️ +20.7%	hey_snips_v4_model17_nnef _{RSS @ ready · pulse8}	`cortex-a9`	19.8 MB → 23.9 MB
⚠️ +20.6%	en_tdnn_8M _{RSS @ ready · pulse_240ms}	`cortex-a53`	64.1 MB → 77.4 MB
⚠️ +20.6%	dummy_conmer_12M _{RSS @ load · pulse_120ms}	`orangepi-rv2`	67.7 MB → 81.6 MB
⚠️ +20.5%	dummy_conmer_12M _{RSS @ load · pulse_120ms}	`beaglev-ahead`	67.8 MB → 81.8 MB
⚠️ +20.4%	openelm_270M_q40ef16_541 _{RSS @ load · cuda}	`i9-11900kb_rtx-4060`	206 MB → 248 MB
⚠️ +20.4%	hey_snips_v4_model17 _{RSS @ load · 2sec}	`cortex-a7`	24.1 MB → 29 MB
⚠️ +20.1%	openelm_270M_q40ef16_541 _{RSS @ load · cpu}	`i9-11900kb_rtx-4060`	206 MB → 248 MB
⚠️ +20.1%	voicecom_float _{RSS @ load · 2sec}	`cortex-a55`	23.8 MB → 28.5 MB
⚠️ +20.1%	speaker_id _{RSS @ load · pulse8}	`orangepi-rv2`	15.9 MB → 19.1 MB
⚠️ +19.9%	hey_snips_v1 _{RSS @ load · 400ms}	`orangepi-rv2`	13.7 MB → 16.4 MB
⚠️ +19.8%	hey_snips_v1 _{RSS @ load · 400ms}	`beaglev-ahead`	13.7 MB → 16.4 MB
⚠️ +19.8%	llama_3_2_1B_instruct_q40ef16_541 _{RSS @ ready · cuda}	`jetson-orin-nx`	357 MB → 428 MB
⚠️ +19.8%	hey_snips_v1 _{RSS @ load · 400ms}	`cortex-a9`	16.7 MB → 20 MB
⚠️ +19.6%	llama_3_2_1B_q40ef32_516 _{RSS @ ready · cuda}	`jetson-orin-nx`	356 MB → 426 MB
⚠️ +19.5%	en_tdnn_lstm_bn_q7 _{RSS @ ready · pulse_240ms}	`i9-11900kb_rtx-4060`	31.9 MB → 38.1 MB
⚠️ +19.5%	en_tdnn_15M _{RSS @ load · pulse_120ms}	`orangepi-rv2`	87.3 MB → 104 MB
⚠️ +19.4%	en_tdnn_15M _{RSS @ load · 2600ms}	`beaglev-ahead`	86.3 MB → 103 MB
⚠️ +19.4%	trunet _{RSS @ load · pulse1_f32}	`i9-11900kb_rtx-4060`	29.4 MB → 35.2 MB
⚠️ +19.4%	mobilenet_v1_1 _{RSS @ load · pass}	`cortex-a53`	48.1 MB → 57.5 MB
⚠️ +19.3%	en_tdnn_15M _{RSS @ load · pulse_240ms}	`orangepi-rv2`	87.3 MB → 104 MB
⚠️ +19.3%	hey_snips_v1 _{RSS @ load · 400ms}	`cortex-a7`	16.7 MB → 19.9 MB
⚠️ +19.2%	llama_3_2_3B_instruct_q40ef16_541 _{RSS @ ready · cuda}	`jetson-orin-nx`	433 MB → 516 MB
⚠️ +19.1%	llama_3_2_3B_q40ef32_516 _{RSS @ ready · cuda}	`jetson-orin-nx`	436 MB → 520 MB
⚠️ +19.1%	en_tdnn_15M _{RSS @ ready · pulse_240ms}	`i9-11900kb_rtx-4060`	127 MB → 151 MB
⚠️ +19.1%	en_tdnn_8M _{RSS @ load · pulse_180ms}	`beaglev-ahead`	49 MB → 58.4 MB
⚠️ +19.0%	en_tdnn_15M _{RSS @ load · 2600ms}	`orangepi-rv2`	86.4 MB → 103 MB
⚠️ +18.9%	en_tdnn_15M_nnef _{RSS @ ready · pulse_240ms}	`orangepi-rv2`	85.7 MB → 102 MB
⚠️ +18.6%	en_tdnn_15M_nnef _{RSS @ ready · pulse_240ms}	`beaglev-ahead`	86.1 MB → 102 MB
⚠️ +18.6%	inceptionv3 _{RSS @ load · pass_mt}	`i9-11900kb_rtx-4060`	151 MB → 179 MB
⚠️ +18.6%	inceptionv3 _{RSS @ ready · pass_mt}	`cortex-a9`	126 MB → 150 MB
⚠️ +18.5%	en_tdnn_15M _{RSS @ load · pulse_120ms}	`beaglev-ahead`	87.3 MB → 103 MB
⚠️ +18.5%	en_tdnn_15M _{RSS @ load · pulse_240ms}	`beaglev-ahead`	87.3 MB → 103 MB
⚠️ +18.5%	speaker_id _{RSS @ ready · pulse8}	`orangepi-rv2`	17 MB → 20.2 MB
⚠️ +18.4%	en_tdnn_8M _{RSS @ ready · 2600ms}	`cortex-a53`	64.7 MB → 76.6 MB
⚠️ +18.2%	mdl_en_2019_Q3_librispeech_onnx _{RSS @ load · 2600ms}	`cortex-a53`	32.5 MB → 38.4 MB
⚠️ +18.0%	inceptionv3 _{RSS @ load · pass_mt}	`cortex-a9`	116 MB → 137 MB
⚠️ +17.9%	en_tdnn_pyt_15M _{RSS @ load · pulse_120ms}	`cortex-a53`	74.4 MB → 87.8 MB
⚠️ +17.8%	inceptionv3 _{RSS @ load · pass}	`cortex-a7`	116 MB → 137 MB
⚠️ +17.8%	en_tdnn_lstm_bn_q7 _{RSS @ ready · 2600ms}	`i9-11900kb_rtx-4060`	31.7 MB → 37.3 MB
⚠️ +17.7%	hey_snips_v4_model17 _{RSS @ ready · 2sec}	`cortex-a7`	25.6 MB → 30.2 MB
⚠️ +17.6%	trunet _{RSS @ load · pulse1_f16}	`cortex-a7`	21.7 MB → 25.6 MB
⚠️ +17.5%	en_tdnn_15M _{RSS @ load · pulse_240ms}	`cortex-a9`	90.2 MB → 106 MB
⚠️ +17.5%	en_tdnn_15M _{RSS @ ready · pulse_120ms}	`i9-11900kb_rtx-4060`	128 MB → 151 MB
⚠️ +17.5%	en_tdnn_15M _{RSS @ load · pulse_240ms}	`cortex-a7`	90.3 MB → 106 MB
⚠️ +17.5%	en_tdnn_15M _{RSS @ load · pulse_120ms}	`cortex-a9`	89.5 MB → 105 MB
⚠️ +17.4%	trunet _{RSS @ load · pulse1_f16}	`cortex-a9`	21.8 MB → 25.6 MB
⚠️ +17.4%	dummy_conmer_12M _{RSS @ load · pulse_120ms}	`i9-11900kb_rtx-4060`	85.9 MB → 101 MB
⚠️ +16.9%	en_tdnn_15M _{RSS @ ready · pulse_240ms}	`cortex-a9`	92 MB → 108 MB
⚠️ +16.8%	arm_ml_kws_cnn_m _{RSS @ load · pass}	`orangepi-rv2`	14.6 MB → 17 MB
⚠️ +16.7%	llama_3_2_3B_q40ef32_516 _{RSS @ ready · cuda}	`i9-11900kb_rtx-4060`	617 MB → 720 MB
⚠️ +16.7%	en_tdnn_15M _{RSS @ ready · pulse_120ms}	`cortex-a9`	91.7 MB → 107 MB
⚠️ +16.4%	hey_snips_v4_model17 _{RSS @ ready · 2sec}	`cortex-a9`	25.7 MB → 29.9 MB
⚠️ +16.3%	hey_snips_v31 _{RSS @ load · 400ms}	`cortex-a7`	18.3 MB → 21.3 MB
⚠️ +16.3%	en_tdnn_15M _{RSS @ load · 2600ms}	`cortex-a9`	89.8 MB → 104 MB
⚠️ +16.1%	speaker_id _{RSS @ load · pulse8}	`beaglev-ahead`	16 MB → 18.6 MB
⚠️ +16.0%	en_tdnn_15M _{RSS @ ready · pulse_240ms}	`cortex-a7`	92.1 MB → 107 MB
⚠️ +16.0%	en_tdnn_15M_nnef _{RSS @ ready · pulse_240ms}	`cortex-a53`	91.6 MB → 106 MB
⚠️ +16.0%	hey_snips_v4_model17_nnef _{RSS @ ready · pulse8}	`cortex-a7`	19.7 MB → 22.9 MB
⚠️ +16.0%	trunet _{RSS @ load · pulse1_f32}	`cortex-a7`	21.7 MB → 25.2 MB
⚠️ +15.9%	en_tdnn_lstm_bn_q7 _{RSS @ load · pulse_240ms}	`cortex-a53`	24.5 MB → 28.5 MB
⚠️ +15.9%	en_tdnn_15M _{RSS @ load · pulse_120ms}	`cortex-a7`	90.4 MB → 105 MB
⚠️ +15.8%	en_tdnn_8M_nnef _{RSS @ load · pulse_240ms}	`i9-11900kb_rtx-4060`	54.4 MB → 63 MB
⚠️ +15.8%	en_tdnn_15M_nnef _{RSS @ load · pulse_240ms}	`i9-11900kb_rtx-4060`	92.3 MB → 107 MB
⚠️ +15.6%	en_tdnn_pyt_15M _{RSS @ ready · pulse_120ms}	`i9-11900kb_rtx-4060`	90.5 MB → 105 MB
⚠️ +15.5%	trunet _{RSS @ load · pulse1_f32}	`cortex-a9`	22.5 MB → 26 MB
⚠️ +15.3%	en_tdnn_15M _{RSS @ load · 2600ms}	`cortex-a7`	90.2 MB → 104 MB
⚠️ +15.2%	hey_snips_v1 _{RSS @ ready · 400ms}	`cortex-a55`	22.1 MB → 25.4 MB
⚠️ +15.1%	hey_snips_v31 _{RSS @ load · 400ms}	`cortex-a9`	18.3 MB → 21 MB
⚠️ +14.9%	inceptionv3 _{RSS @ load · pass}	`i9-11900kb_rtx-4060`	150 MB → 172 MB
⚠️ +14.9%	en_tdnn_15M _{RSS @ ready · pulse_120ms}	`cortex-a7`	92 MB → 106 MB
⚠️ +14.7%	en_tdnn_15M _{RSS @ ready · 2600ms}	`cortex-a7`	91.4 MB → 105 MB
⚠️ +14.5%	trunet _{RSS @ ready · pulse1_f16}	`cortex-a7`	23 MB → 26.4 MB
⚠️ +14.4%	en_tdnn_8M _{RSS @ load · pulse_180ms}	`cortex-a53`	54.3 MB → 62.1 MB
⚠️ +14.3%	dummy_conmer_12M _{RSS @ ready · pulse_120ms}	`cortex-a7`	76.6 MB → 87.6 MB
⚠️ +14.0%	en_tdnn_15M _{RSS @ load · pulse_120ms}	`cortex-a53`	95 MB → 108 MB
⚠️ +14.0%	trunet _{RSS @ load · pulse1_f32}	`cortex-a55`	24.6 MB → 28.1 MB
⚠️ +13.9%	llama_3_2_3B_instruct_q40ef16_541 _{RSS @ ready · cuda}	`i9-11900kb_rtx-4060`	618 MB → 704 MB
⚠️ +13.8%	mdl_en_2019_Q3_librispeech_onnx _{RSS @ ready · 2600ms}	`cortex-a53`	35.2 MB → 40.1 MB
⚠️ +13.7%	hey_snips_v4_model17 _{RSS @ ready · pulse8}	`cortex-a9`	26.9 MB → 30.6 MB
⚠️ +13.7%	en_tdnn_8M _{RSS @ load · pulse_240ms}	`cortex-a53`	54.5 MB → 61.9 MB
⚠️ +13.7%	en_tdnn_8M _{RSS @ load · pulse_120ms}	`cortex-a53`	54.7 MB → 62.2 MB
⚠️ +13.6%	hey_snips_v4_model17 _{RSS @ load · pulse8}	`cortex-a7`	24.3 MB → 27.6 MB
⚠️ +13.5%	inceptionv3 _{RSS @ load · pass}	`cortex-a9`	116 MB → 132 MB
⚠️ +13.5%	dummy_conmer_12M _{RSS @ ready · pulse_120ms}	`cortex-a9`	77.1 MB → 87.5 MB
⚠️ +13.4%	nemotron_3_5_asr_streaming_0_6b_f32f32_decoder_pass _{RSS @ ready · cuda}	`jetson-orin-nx`	215 MB → 244 MB
⚠️ +13.2%	parakeet_tdt_600m_v3_f32f32_decoder_pass _{RSS @ ready · cpu}	`i9-11900kb_rtx-4060`	87.2 MB → 98.7 MB
⚠️ +13.2%	hey_snips_v4_model17_nnef _{RSS @ load · pulse8}	`cortex-a9`	17.9 MB → 20.2 MB
⚠️ +13.1%	trunet _{RSS @ ready · pulse1_f16}	`cortex-a9`	23.3 MB → 26.4 MB
⚠️ +13.0%	en_tdnn_8M _{RSS @ load · 2600ms}	`cortex-a53`	54.6 MB → 61.7 MB
⚠️ +12.9%	mdl_en_2019_Q3_librispeech_onnx _{RSS @ load · pulse_240ms}	`cortex-a53`	32.5 MB → 36.6 MB
⚠️ +12.9%	en_tdnn_15M _{RSS @ load · pulse_240ms}	`cortex-a53`	94.8 MB → 107 MB
⚠️ +12.9%	en_tdnn_15M _{RSS @ load · 2600ms}	`cortex-a53`	94.8 MB → 107 MB
⚠️ +12.7%	llama_3_2_1B_instruct_q40ef16_541 _{RSS @ ready · cuda}	`i9-11900kb_rtx-4060`	538 MB → 607 MB
⚠️ +12.5%	llama_3_2_1B_q40ef32_516 _{RSS @ ready · cpu}	`i9-11900kb_rtx-4060`	728 MB → 819 MB
⚠️ +12.5%	arm_ml_kws_cnn_m _{RSS @ ready · pass}	`cortex-a7`	20.6 MB → 23.1 MB
⚠️ +12.4%	llama_3_2_1B_instruct_q40ef16_541 _{RSS @ ready · cpu}	`i9-11900kb_rtx-4060`	728 MB → 818 MB
⚠️ +12.4%	inceptionv3 _{RSS @ ready · pass}	`i9-11900kb_rtx-4060`	192 MB → 216 MB
⚠️ +12.3%	hey_snips_v1 _{RSS @ ready · 400ms}	`cortex-a53`	21.9 MB → 24.6 MB
⚠️ +12.2%	llama_3_2_1B_instruct_q40ef16_541 _{RSS @ ready · cpu_mt}	`i9-11900kb_rtx-4060`	727 MB → 816 MB
⚠️ +12.0%	en_tdnn_lstm_bn_q7 _{RSS @ load · 2600ms}	`cortex-a53`	24.3 MB → 27.2 MB
⚠️ +12.0%	trunet _{RSS @ ready · pulse1_f32}	`cortex-a9`	25.1 MB → 28.1 MB
⚠️ +11.9%	hey_snips_v4_model17_nnef _{RSS @ ready · pulse8}	`cortex-a55`	24.6 MB → 27.6 MB
⚠️ +11.8%	llama_3_2_1B_q40ef32_516 _{RSS @ ready · cuda}	`i9-11900kb_rtx-4060`	539 MB → 603 MB
⚠️ +11.7%	llama_3_1_8B_instruct_q40ef16_541 _{RSS @ ready · cuda}	`i9-11900kb_rtx-4060`	690 MB → 771 MB
⚠️ +11.4%	speaker_id _{RSS @ load · pulse8}	`cortex-a53`	19.5 MB → 21.8 MB
⚠️ +11.4%	arm_ml_kws_cnn_m _{RSS @ load · pass}	`cortex-a7`	18.1 MB → 20.2 MB
⚠️ +11.3%	arm_ml_kws_cnn_m _{RSS @ ready · pass}	`cortex-a9`	20.4 MB → 22.7 MB
⚠️ +11.2%	trunet _{RSS @ load · pulse1_f16}	`cortex-a53`	25.8 MB → 28.7 MB
⚠️ +11.2%	voicecom_fake_quant _{RSS @ ready · 2sec}	`cortex-a53`	33 MB → 36.7 MB
⚠️ +11.2%	arm_ml_kws_cnn_m _{RSS @ load · pass}	`cortex-a9`	18 MB → 20 MB
⚠️ +11.1%	dummy_conmer_12M _{RSS @ load · pulse_120ms}	`cortex-a9`	73.6 MB → 81.7 MB
⚠️ +11.0%	llama_3_2_1B_q40ef32_516 _{RSS @ load · cuda}	`jetson-orin-nx`	730 MB → 810 MB
⚠️ +10.9%	speaker_id _{RSS @ load · pulse8}	`cortex-a55`	19.7 MB → 21.8 MB
⚠️ +10.9%	inceptionv1q _{RSS @ ready · pass}	`cortex-a9`	30.2 MB → 33.5 MB
⚠️ +10.9%	parakeet_tdt_600m_v3_f32f32_decoder_pass _{RSS @ ready · cuda}	`jetson-orin-nx`	219 MB → 243 MB
⚠️ +10.6%	hey_snips_v4_model17_nnef _{RSS @ ready · pulse8}	`cortex-a53`	24.5 MB → 27.1 MB
⚠️ +10.5%	trunet _{RSS @ ready · pulse1_f16}	`cortex-a55`	26 MB → 28.7 MB
⚠️ +10.4%	voicecom_fake_quant _{RSS @ ready · 2sec}	`cortex-a55`	33 MB → 36.5 MB
⚠️ +10.4%	llama_3_2_1B_instruct_q40ef16_541 _{RSS @ load · cuda}	`jetson-orin-nx`	728 MB → 804 MB
⚠️ +10.2%	llama_3_2_1B_q40ef32_516 _{RSS @ load · cpu}	`i9-11900kb_rtx-4060`	732 MB → 806 MB
⚠️ +10.1%	trunet _{RSS @ ready · pulse1_f16}	`cortex-a53`	26.2 MB → 28.8 MB
⚠️ +9.9%	en_tdnn_15M _{load · 2600ms}	`cortex-a53`	1.46 s → 1.61 s
⚠️ +9.9%	llama_3_2_1B_instruct_q40ef16_541 _{RSS @ load · cpu_mt}	`i9-11900kb_rtx-4060`	731 MB → 803 MB
⚠️ +9.8%	llama_3_2_1B_instruct_q40ef16_541 _{RSS @ load · cpu}	`i9-11900kb_rtx-4060`	732 MB → 804 MB
⚠️ +9.7%	llama_3_2_1B_q40ef32_516 _{RSS @ load · cuda}	`i9-11900kb_rtx-4060`	732 MB → 804 MB
⚠️ +9.6%	trunet _{RSS @ ready · pulse1_f32}	`cortex-a55`	27.5 MB → 30.1 MB
⚠️ +9.6%	llama_3_2_1B_instruct_q40ef16_541 _{RSS @ load · cuda}	`i9-11900kb_rtx-4060`	731 MB → 801 MB
⚠️ +9.5%	en_tdnn_15M _{RSS @ ready · pulse_120ms}	`cortex-a53`	111 MB → 122 MB
⚠️ +9.4%	llama_3_1_8B_instruct_q40ef16_541 _{RSS @ ready · cuda}	`jetson-orin-nx`	508 MB → 556 MB
⚠️ +9.4%	en_tdnn_15M _{RSS @ ready · 2600ms}	`cortex-a53`	110 MB → 120 MB
⚠️ +9.3%	speaker_id _{RSS @ ready · pulse8}	`beaglev-ahead`	17 MB → 18.6 MB
⚠️ +9.3%	nemotron_3_5_asr_streaming_0_6b_f32f32_decoder_pass _{RSS @ ready · cpu}	`i9-11900kb_rtx-4060`	100 MB → 109 MB
⚠️ +9.2%	dummy_conmer_12M _{RSS @ load · pulse_120ms}	`cortex-a55`	79.4 MB → 86.7 MB
⚠️ +8.9%	dummy_conmer_12M _{RSS @ load · pulse_120ms}	`cortex-a53`	79.7 MB → 86.7 MB
⚠️ +8.5%	dummy_conmer_12M _{RSS @ load · pulse_120ms}	`cortex-a7`	73.7 MB → 80 MB
⚠️ +8.1%	hey_snips_v31 _{RSS @ ready · 400ms}	`cortex-a53`	21.1 MB → 22.8 MB
⚠️ +8.0%	en_tdnn_15M _{load · 2600ms}	`cortex-a9`	2.5 s → 2.7 s
⚠️ +7.8%	en_tdnn_15M_nnef _{RSS @ ready · pulse_240ms}	`cortex-a7`	89 MB → 96 MB
⚠️ +7.5%	arm_ml_kws_cnn_m _{RSS @ ready · pass}	`cortex-a55`	23.4 MB → 25.2 MB
⚠️ +7.1%	trunet _{RSS @ ready · pulse1_f32}	`cortex-a53`	27.4 MB → 29.4 MB
⚠️ +7.1%	en_tdnn_15M_nnef _{RSS @ ready · pulse_240ms}	`cortex-a9`	89.4 MB → 95.7 MB
⚠️ +7.1%	en_tdnn_15M _{load · pulse_240ms}	`cortex-a53`	1.71 s → 1.83 s
⚠️ +6.9%	arm_ml_kws_cnn_m _{RSS @ ready · pass}	`cortex-a53`	23.6 MB → 25.3 MB
⚠️ +6.8%	hey_snips_v31 _{RSS @ ready · 400ms}	`cortex-a55`	21.3 MB → 22.7 MB
⚠️ +6.7%	llama_3_2_3B_instruct_q40ef16_541 _{RSS @ load · cuda}	`jetson-orin-nx`	1.85 GB → 1.97 GB
⚠️ +6.7%	llama_3_2_3B_instruct_q40ef16_541 _{RSS @ load · cuda}	`i9-11900kb_rtx-4060`	1.85 GB → 1.98 GB
⚠️ +6.7%	mobilenet_v2_1 _{RSS @ load · pass}	`cortex-a53`	67.5 MB → 72 MB
⚠️ +6.5%	llama_3_2_3B_q40ef32_516 _{RSS @ load · cuda}	`i9-11900kb_rtx-4060`	1.85 GB → 1.97 GB
⚠️ +6.4%	en_tdnn_15M _{load+optimize · 2600ms}	`cortex-a53`	1.78 s → 1.89 s
⚠️ +6.4%	voicecom_float _{RSS @ ready · 2sec}	`cortex-a53`	31.5 MB → 33.5 MB
⚠️ +6.3%	en_tdnn_15M _{load · pulse_120ms}	`cortex-a53`	1.72 s → 1.83 s
⚠️ +6.3%	voicecom_float _{RSS @ ready · 2sec}	`cortex-a55`	31.5 MB → 33.5 MB
⚠️ +6.2%	mobilenet_v2_1 _{RSS @ load · pass_mt}	`cortex-a53`	67.5 MB → 71.7 MB
⚠️ +6.0%	qwen3_1_7B_q40ef16_541 _{RSS @ load · cuda}	`i9-11900kb_rtx-4060`	1.02 GB → 1.08 GB
⚠️ +5.8%	qwen3_1_7B_q40ef16_541 _{RSS @ load · cuda}	`jetson-orin-nx`	1.01 GB → 1.07 GB
⚠️ +5.3%	en_tdnn_8M_nnef _{RSS @ load · pulse_240ms}	`orangepi-rv2`	46.4 MB → 48.9 MB
⚠️ +5.3%	en_tdnn_15M _{load · pulse_240ms}	`cortex-a9`	2.87 s → 3.02 s
⚠️ +5.1%	en_tdnn_15M _{load · pulse_120ms}	`cortex-a9`	2.87 s → 3.02 s

github-actions · 2026-09-07T15:51:54Z

🔴 Bench vs main — 4 speed regression(s) · ⚠️ 402 secondary

Reference: 2026-09-07 morning nightly run (0d old) · full report → run

Speed — evaltime · prefill · decode

Δ	metric	device	main → PR
🔴 +7.4%	inceptionv3 _{evaltime · pass_mt}	`apple-m1-max`	45.9 ms → 49.3 ms
🔴 +7.3%	inceptionv3 _{evaltime · pass}	`apple-m1-max`	59.1 ms → 63.5 ms
🔴 +7.3%	mobilenet_v1_1 _{evaltime · pass}	`apple-m1-max`	18 ms → 19.3 ms
🔴 +6.8%	llama_3_2_1B_instruct_q40ef16_541 _{prefill · cpu_mt}	`apple-m1-max`	1.88 ms/tok _{533.2 tok/s} → 2 ms/tok _{499.5 tok/s}

Improvements

Δ	metric	device	main → PR
🟢 -21.0%	speaker_id _{evaltime · pulse8}	`beaglev-ahead`	3.16 ms/pulse _{0.0394 RTF} → 2.49 ms/pulse _{0.0312 RTF}
🟢 -19.6%	hey_snips_v4_model17 _{evaltime · pulse8}	`beaglev-ahead`	5.88 ms/pulse _{0.0735 RTF} → 4.73 ms/pulse _{0.0591 RTF}
🟢 -16.4%	mobilenet_v2_1 _{evaltime · pass_mt}	`beaglev-ahead`	504 ms → 421 ms
🟢 -15.0%	hey_snips_v4_model17 _{evaltime · 2sec}	`beaglev-ahead`	48.8 ms → 41.5 ms
🟢 -14.7%	trunet _{evaltime · pulse1_f32}	`beaglev-ahead`	5.69 ms/pulse _{0.569 RTF} → 4.85 ms/pulse _{0.485 RTF}

+19 more improvement(s)

Δ	metric	device	main → PR
🟢 -10.4%	inceptionv3 _{evaltime · pass_mt}	`beaglev-ahead`	2.27 s → 2.03 s
🟢 -9.3%	hey_snips_v4_model17_nnef _{evaltime · pulse8}	`beaglev-ahead`	5.24 ms/pulse _{0.0655 RTF} → 4.76 ms/pulse _{0.0595 RTF}
🟢 -9.2%	hey_snips_v4_model17_nnef _{evaltime · pulse8}	`cortex-a55`	2.1 ms/pulse _{0.0262 RTF} → 1.91 ms/pulse _{0.0238 RTF}
🟢 -8.6%	trunet _{evaltime · pulse1_f32}	`cortex-a7`	8.07 ms/pulse _{0.807 RTF} → 7.38 ms/pulse _{0.738 RTF}
🟢 -8.6%	hey_snips_v4_model17 _{evaltime · pulse8}	`cortex-a9`	6.32 ms/pulse _{0.079 RTF} → 5.78 ms/pulse _{0.0723 RTF}
🟢 -8.4%	hey_snips_v31 _{evaltime · 400ms}	`beaglev-ahead`	6.17 ms → 5.66 ms
🟢 -8.3%	hey_snips_v4_model17_nnef _{evaltime · pulse8}	`cortex-a9`	6.3 ms/pulse _{0.0787 RTF} → 5.77 ms/pulse _{0.0722 RTF}
🟢 -8.2%	hey_snips_v4_model17 _{evaltime · pulse8}	`cortex-a55`	2.1 ms/pulse _{0.0262 RTF} → 1.93 ms/pulse _{0.0241 RTF}
🟢 -7.6%	mobilenet_v2_1 _{evaltime · pass}	`beaglev-ahead`	1.15 s → 1.06 s
🟢 -7.6%	hey_snips_v4_model17 _{evaltime · pulse8}	`cortex-a7`	6.96 ms/pulse _{0.0869 RTF} → 6.43 ms/pulse _{0.0804 RTF}
🟢 -7.5%	hey_snips_v4_model17_nnef _{evaltime · pulse8}	`cortex-a7`	6.92 ms/pulse _{0.0865 RTF} → 6.4 ms/pulse _{0.08 RTF}
🟢 -6.7%	hey_snips_v4_model17_nnef _{evaltime · pulse8}	`apple-m1-max`	0.116 ms/pulse _{0.00145 RTF} → 0.108 ms/pulse _{0.00135 RTF}
🟢 -6.7%	trunet _{evaltime · pulse1_f32}	`cortex-a9`	8.1 ms/pulse _{0.81 RTF} → 7.56 ms/pulse _{0.756 RTF}
🟢 -6.4%	hey_snips_v4_model17 _{evaltime · pulse8}	`apple-m1-max`	0.116 ms/pulse _{0.00144 RTF} → 0.108 ms/pulse _{0.00135 RTF}
🟢 -6.3%	mobilenet_v2_1 _{evaltime · pass_mt}	`cortex-a9`	953 ms → 893 ms
🟢 -5.8%	hey_snips_v4_model17 _{evaltime · pulse8}	`cortex-a53`	2.73 ms/pulse _{0.0342 RTF} → 2.57 ms/pulse _{0.0322 RTF}
🟢 -5.3%	hey_snips_v4_model17_nnef _{evaltime · pulse8}	`cortex-a53`	2.72 ms/pulse _{0.034 RTF} → 2.57 ms/pulse _{0.0322 RTF}
🟢 -5.3%	en_tdnn_8M _{evaltime · pulse_240ms}	`beaglev-ahead`	107 ms/pulse _{0.445 RTF} → 101 ms/pulse _{0.422 RTF}
🟢 -5.1%	inceptionv1q _{evaltime · pass}	`apple-m1-max`	41.6 ms → 39.4 ms

⚠️ 402 secondary regression(s)

Δ	metric	device	main → PR
⚠️ +100.5%	hey_snips_v4_model17 _{RSS @ load · 2sec}	`orangepi-rv2`	16.9 MB → 33.9 MB
⚠️ +100.2%	hey_snips_v4_model17 _{RSS @ load · 2sec}	`beaglev-ahead`	16.9 MB → 33.8 MB
⚠️ +91.2%	hey_snips_v4_model17 _{RSS @ ready · 2sec}	`orangepi-rv2`	18.2 MB → 34.8 MB
⚠️ +85.4%	hey_snips_v4_model17 _{RSS @ ready · 2sec}	`beaglev-ahead`	18.5 MB → 34.3 MB
⚠️ +75.1%	voicecom_fake_quant _{RSS @ ready · 2sec}	`beaglev-ahead`	18.3 MB → 32 MB
⚠️ +74.5%	voicecom_fake_quant _{RSS @ ready · 2sec}	`orangepi-rv2`	18.6 MB → 32.4 MB
⚠️ +67.3%	hey_snips_v4_model17 _{RSS @ load · pulse8}	`beaglev-ahead`	17.8 MB → 29.7 MB
⚠️ +67.2%	hey_snips_v4_model17 _{RSS @ load · pulse8}	`orangepi-rv2`	17.7 MB → 29.6 MB
⚠️ +64.3%	mobilenet_v2_1 _{RSS @ load · pass_mt}	`cortex-a7`	42.6 MB → 70 MB
⚠️ +64.2%	inceptionv1q _{RSS @ ready · pass}	`orangepi-rv2`	26.2 MB → 43 MB
⚠️ +60.8%	hey_snips_v4_model17 _{RSS @ ready · pulse8}	`orangepi-rv2`	18.9 MB → 30.4 MB
⚠️ +58.5%	voicecom_float _{RSS @ ready · 2sec}	`cortex-a9`	20.8 MB → 33 MB
⚠️ +58.1%	voicecom_float _{RSS @ ready · 2sec}	`beaglev-ahead`	17.7 MB → 28 MB
⚠️ +57.2%	hey_snips_v4_model17 _{RSS @ ready · pulse8}	`beaglev-ahead`	19 MB → 29.8 MB
⚠️ +56.2%	en_tdnn_8M _{RSS @ ready · pulse_240ms}	`orangepi-rv2`	49.8 MB → 77.7 MB
⚠️ +55.1%	voicecom_float _{RSS @ ready · 2sec}	`orangepi-rv2`	18 MB → 27.9 MB
⚠️ +54.9%	mobilenet_v1_1 _{RSS @ ready · pass}	`cortex-a9`	36.6 MB → 56.8 MB
⚠️ +52.6%	voicecom_fake_quant _{RSS @ ready · 2sec}	`cortex-a7`	22.2 MB → 33.9 MB
⚠️ +52.5%	en_tdnn_8M _{RSS @ ready · 2600ms}	`cortex-a7`	52.7 MB → 80.3 MB
⚠️ +52.3%	mobilenet_v2_1 _{RSS @ load · pass_mt}	`orangepi-rv2`	38.8 MB → 59 MB
⚠️ +52.2%	en_tdnn_8M _{RSS @ ready · 2600ms}	`cortex-a9`	52.8 MB → 80.4 MB
⚠️ +52.1%	voicecom_fake_quant _{RSS @ ready · 2sec}	`cortex-a9`	22.3 MB → 34 MB
⚠️ +52.1%	inceptionv1q _{RSS @ load · pass}	`orangepi-rv2`	20.2 MB → 30.8 MB
⚠️ +52.1%	mobilenet_v2_1 _{RSS @ ready · pass}	`beaglev-ahead`	41 MB → 62.3 MB
⚠️ +52.0%	mobilenet_v2_1 _{RSS @ load · pass_mt}	`beaglev-ahead`	38.8 MB → 59 MB
⚠️ +50.9%	en_tdnn_8M _{RSS @ ready · pulse_180ms}	`orangepi-rv2`	49.8 MB → 75.1 MB
⚠️ +50.9%	mobilenet_v2_1 _{RSS @ ready · pass_mt}	`cortex-a7`	45.7 MB → 69 MB
⚠️ +50.7%	mobilenet_v2_1 _{RSS @ ready · pass}	`orangepi-rv2`	40.9 MB → 61.7 MB
⚠️ +50.5%	en_tdnn_8M _{RSS @ ready · 2600ms}	`beaglev-ahead`	49.3 MB → 74.2 MB
⚠️ +50.2%	en_tdnn_8M _{RSS @ ready · pulse_240ms}	`cortex-a7`	54.2 MB → 81.3 MB
⚠️ +50.1%	en_tdnn_8M _{RSS @ ready · 2600ms}	`orangepi-rv2`	49 MB → 73.5 MB
⚠️ +49.7%	en_tdnn_8M _{RSS @ ready · pulse_120ms}	`orangepi-rv2`	49.8 MB → 74.5 MB
⚠️ +49.1%	voicecom_float _{RSS @ ready · 2sec}	`cortex-a7`	20.9 MB → 31.1 MB
⚠️ +48.8%	mobilenet_v2_1 _{RSS @ ready · pass}	`cortex-a9`	45.9 MB → 68.2 MB
⚠️ +48.0%	en_tdnn_8M _{RSS @ ready · pulse_240ms}	`beaglev-ahead`	50.1 MB → 74.2 MB
⚠️ +47.5%	mobilenet_v1_1 _{RSS @ load · pass}	`cortex-a9`	35 MB → 51.6 MB
⚠️ +46.7%	en_tdnn_8M _{RSS @ ready · pulse_180ms}	`beaglev-ahead`	50.1 MB → 73.5 MB
⚠️ +44.7%	en_tdnn_8M _{RSS @ ready · pulse_180ms}	`cortex-a9`	54.1 MB → 78.3 MB
⚠️ +44.4%	en_tdnn_8M _{RSS @ ready · pulse_120ms}	`beaglev-ahead`	50.1 MB → 72.3 MB
⚠️ +44.2%	mobilenet_v1_1 _{RSS @ ready · pass}	`beaglev-ahead`	33 MB → 47.5 MB
⚠️ +43.7%	mobilenet_v2_1 _{RSS @ ready · pass_mt}	`beaglev-ahead`	41 MB → 59 MB
⚠️ +43.6%	en_tdnn_pyt_15M _{RSS @ load · pulse_120ms}	`beaglev-ahead`	60.3 MB → 86.7 MB
⚠️ +43.4%	hey_snips_v4_model17 _{RSS @ load · 2sec}	`i9-11900kb_rtx-4060`	33.8 MB → 48.5 MB
⚠️ +43.2%	hey_snips_v4_model17_nnef _{RSS @ ready · pulse8}	`orangepi-rv2`	16.4 MB → 23.5 MB
⚠️ +43.1%	en_tdnn_pyt_15M _{RSS @ load · pulse_120ms}	`orangepi-rv2`	60.3 MB → 86.2 MB
⚠️ +42.6%	inceptionv1q _{RSS @ load · pass}	`beaglev-ahead`	20.2 MB → 28.8 MB
⚠️ +42.5%	mobilenet_v2_1 _{RSS @ load · pass_mt}	`cortex-a9`	42.5 MB → 60.6 MB
⚠️ +41.7%	voicecom_float _{RSS @ load · 2sec}	`beaglev-ahead`	16.3 MB → 23.2 MB
⚠️ +41.6%	mobilenet_v2_1 _{RSS @ ready · pass}	`cortex-a7`	46 MB → 65.1 MB
⚠️ +41.4%	hey_snips_v4_model17_nnef _{RSS @ ready · pulse8}	`beaglev-ahead`	16.7 MB → 23.6 MB
⚠️ +41.3%	voicecom_float _{RSS @ load · 2sec}	`orangepi-rv2`	16.3 MB → 23.1 MB
⚠️ +41.1%	voicecom_fake_quant _{RSS @ load · 2sec}	`cortex-a9`	19.3 MB → 27.3 MB
⚠️ +40.9%	mobilenet_v2_1 _{RSS @ ready · pass_mt}	`cortex-a55`	53.1 MB → 74.8 MB
⚠️ +40.5%	en_tdnn_8M _{RSS @ ready · pulse_120ms}	`cortex-a7`	53.8 MB → 75.6 MB
⚠️ +40.5%	llama_3_2_3B_instruct_q40ef16_541 _{RSS @ ready · cuda}	`jetson-orin-nx`	433 MB → 608 MB
⚠️ +40.1%	en_tdnn_8M _{RSS @ ready · pulse_120ms}	`cortex-a9`	53.7 MB → 75.2 MB
⚠️ +40.1%	mobilenet_v2_1 _{RSS @ load · pass}	`cortex-a9`	42.6 MB → 59.6 MB
⚠️ +39.8%	voicecom_fake_quant _{RSS @ load · 2sec}	`cortex-a7`	19.2 MB → 26.9 MB
⚠️ +39.6%	llama_3_2_3B_q40ef32_516 _{RSS @ ready · cuda}	`jetson-orin-nx`	436 MB → 609 MB
⚠️ +39.4%	hey_snips_v4_model17 _{RSS @ load · 2sec}	`cortex-a55`	28.1 MB → 39.2 MB
⚠️ +38.5%	en_tdnn_8M _{RSS @ ready · pulse_180ms}	`cortex-a7`	54 MB → 74.9 MB
⚠️ +38.5%	voicecom_fake_quant _{RSS @ load · 2sec}	`orangepi-rv2`	17.1 MB → 23.7 MB
⚠️ +38.4%	mobilenet_v2_1 _{RSS @ load · pass}	`cortex-a7`	43 MB → 59.5 MB
⚠️ +38.4%	mobilenet_v2_1 _{RSS @ load · pass}	`beaglev-ahead`	38.8 MB → 53.7 MB
⚠️ +38.3%	inceptionv3 _{RSS @ ready · pass}	`orangepi-rv2`	118 MB → 164 MB
⚠️ +38.1%	voicecom_fake_quant _{RSS @ load · 2sec}	`beaglev-ahead`	17.2 MB → 23.7 MB
⚠️ +38.0%	en_tdnn_8M _{RSS @ ready · pulse_240ms}	`cortex-a9`	53.8 MB → 74.3 MB
⚠️ +37.1%	mobilenet_v2_1 _{RSS @ load · pass}	`orangepi-rv2`	38.7 MB → 53.1 MB
⚠️ +36.9%	en_tdnn_8M_nnef _{RSS @ ready · pulse_240ms}	`beaglev-ahead`	47.8 MB → 65.4 MB
⚠️ +36.6%	hey_snips_v1 _{RSS @ ready · 400ms}	`beaglev-ahead`	14.9 MB → 20.4 MB
⚠️ +36.1%	hey_snips_v4_model17 _{RSS @ load · 2sec}	`cortex-a53`	28.4 MB → 38.6 MB
⚠️ +35.6%	voicecom_float _{RSS @ load · 2sec}	`cortex-a9`	19.3 MB → 26.2 MB
⚠️ +35.5%	en_tdnn_8M_nnef _{RSS @ ready · pulse_240ms}	`cortex-a53`	55.4 MB → 75.1 MB
⚠️ +35.2%	en_tdnn_pyt_15M _{RSS @ load · pulse_120ms}	`cortex-a7`	63.6 MB → 85.9 MB
⚠️ +35.2%	trunet _{RSS @ ready · pulse1_f16}	`orangepi-rv2`	18.5 MB → 25 MB
⚠️ +34.7%	mobilenet_v2_1 _{RSS @ ready · pass_mt}	`cortex-a53`	53 MB → 71.4 MB
⚠️ +34.7%	mobilenet_v1_1 _{RSS @ load · pass}	`beaglev-ahead`	31.2 MB → 42 MB
⚠️ +34.6%	mobilenet_v1_1 _{RSS @ load · pass}	`orangepi-rv2`	31.1 MB → 41.9 MB
⚠️ +34.6%	voicecom_float _{RSS @ load · 2sec}	`cortex-a7`	19.5 MB → 26.2 MB
⚠️ +34.5%	trunet _{RSS @ ready · pulse1_f16}	`beaglev-ahead`	18.6 MB → 25 MB
⚠️ +34.4%	en_tdnn_pyt_15M _{RSS @ ready · pulse_120ms}	`beaglev-ahead`	61.6 MB → 82.8 MB
⚠️ +34.4%	mdl_en_2019_Q3_librispeech_onnx _{RSS @ load · pulse_240ms}	`cortex-a9`	29.4 MB → 39.6 MB
⚠️ +33.7%	en_tdnn_lstm_bn_q7 _{RSS @ load · pulse_240ms}	`cortex-a9`	22.4 MB → 30 MB
⚠️ +33.4%	en_tdnn_pyt_15M _{RSS @ load · pulse_120ms}	`cortex-a9`	64.2 MB → 85.6 MB
⚠️ +33.4%	mdl_en_2019_Q3_librispeech_onnx _{RSS @ load · pulse_240ms}	`cortex-a7`	29.4 MB → 39.2 MB
⚠️ +33.1%	en_tdnn_pyt_15M _{RSS @ ready · pulse_120ms}	`orangepi-rv2`	61.4 MB → 81.7 MB
⚠️ +32.9%	trunet _{RSS @ load · pulse1_f16}	`orangepi-rv2`	17.9 MB → 23.9 MB
⚠️ +32.9%	mdl_en_2019_Q3_librispeech_onnx _{RSS @ load · 2600ms}	`cortex-a7`	29.2 MB → 38.8 MB
⚠️ +32.7%	en_tdnn_lstm_bn_q7 _{RSS @ load · pulse_240ms}	`cortex-a7`	22.3 MB → 29.7 MB
⚠️ +32.5%	hey_snips_v4_model17_nnef _{RSS @ load · pulse8}	`beaglev-ahead`	14.6 MB → 19.3 MB
⚠️ +32.5%	hey_snips_v4_model17 _{RSS @ ready · 2sec}	`i9-11900kb_rtx-4060`	36.9 MB → 48.9 MB
⚠️ +32.3%	en_tdnn_8M _{RSS @ load · pulse_120ms}	`cortex-a9`	51.4 MB → 68 MB
⚠️ +32.2%	trunet _{RSS @ load · pulse1_f16}	`beaglev-ahead`	18.1 MB → 23.9 MB
⚠️ +31.8%	hey_snips_v1 _{RSS @ ready · 400ms}	`orangepi-rv2`	14.9 MB → 19.7 MB
⚠️ +31.7%	hey_snips_v4_model17_nnef _{RSS @ load · pulse8}	`orangepi-rv2`	14.5 MB → 19.1 MB
⚠️ +31.3%	mobilenet_v1_1 _{RSS @ ready · pass}	`orangepi-rv2`	32.9 MB → 43.2 MB
⚠️ +31.3%	en_tdnn_lstm_bn_q7 _{RSS @ load · 2600ms}	`cortex-a9`	22.1 MB → 29.1 MB
⚠️ +31.2%	inceptionv3 _{RSS @ ready · pass_mt}	`orangepi-rv2`	118 MB → 155 MB
⚠️ +31.1%	mobilenet_v2_1 _{RSS @ ready · pass_mt}	`orangepi-rv2`	40.9 MB → 53.6 MB
⚠️ +30.9%	en_tdnn_8M _{RSS @ load · 2600ms}	`orangepi-rv2`	48.2 MB → 63.1 MB
⚠️ +30.9%	en_tdnn_8M_nnef _{RSS @ ready · pulse_240ms}	`cortex-a7`	51 MB → 66.8 MB
⚠️ +30.8%	llama_3_2_3B_q40ef32_516 _{RSS @ ready · cuda}	`i9-11900kb_rtx-4060`	617 MB → 807 MB
⚠️ +30.8%	en_tdnn_8M_nnef _{RSS @ ready · pulse_240ms}	`orangepi-rv2`	47.5 MB → 62.1 MB
⚠️ +30.8%	en_tdnn_8M _{RSS @ load · pulse_120ms}	`orangepi-rv2`	49.1 MB → 64.1 MB
⚠️ +30.5%	inceptionv1q _{RSS @ load · pass}	`cortex-a7`	23.9 MB → 31.2 MB
⚠️ +30.3%	mobilenet_v1_1 _{RSS @ load · pass}	`cortex-a7`	35.2 MB → 45.9 MB
⚠️ +30.3%	en_tdnn_8M _{RSS @ load · pulse_240ms}	`beaglev-ahead`	49 MB → 63.9 MB
⚠️ +30.2%	en_tdnn_8M _{RSS @ load · pulse_180ms}	`cortex-a9`	52.1 MB → 67.8 MB
⚠️ +30.1%	mdl_en_2019_Q3_librispeech_onnx _{RSS @ load · 2600ms}	`cortex-a9`	29.4 MB → 38.2 MB
⚠️ +30.1%	en_tdnn_8M _{RSS @ ready · 2600ms}	`cortex-a55`	62.2 MB → 80.9 MB
⚠️ +30.0%	dummy_conmer_12M _{RSS @ ready · pulse_120ms}	`orangepi-rv2`	68.7 MB → 89.3 MB
⚠️ +30.0%	inceptionv1q _{RSS @ ready · pass}	`beaglev-ahead`	26.2 MB → 34 MB
⚠️ +29.9%	trunet _{RSS @ ready · pulse1_f32}	`orangepi-rv2`	20.3 MB → 26.3 MB
⚠️ +29.7%	voicecom_fake_quant _{RSS @ load · 2sec}	`cortex-a55`	22.7 MB → 29.5 MB
⚠️ +29.7%	en_tdnn_8M _{RSS @ load · pulse_240ms}	`cortex-a9`	51.9 MB → 67.2 MB
⚠️ +29.1%	en_tdnn_lstm_bn_q7 _{RSS @ load · 2600ms}	`cortex-a7`	21.9 MB → 28.3 MB
⚠️ +28.8%	llama_3_2_1B_instruct_q40ef16_541 _{RSS @ ready · cuda}	`jetson-orin-nx`	357 MB → 460 MB
⚠️ +28.8%	hey_snips_v1 _{RSS @ ready · 400ms}	`cortex-a7`	18.6 MB → 23.9 MB
⚠️ +28.4%	parakeet_tdt_600m_v3_f32f32_decoder_pass _{RSS @ load · cuda}	`jetson-orin-nx`	67.2 MB → 86.3 MB
⚠️ +28.4%	llama_3_2_3B_instruct_q40ef16_541 _{RSS @ ready · cuda}	`i9-11900kb_rtx-4060`	618 MB → 794 MB
⚠️ +28.4%	hey_snips_v1 _{RSS @ ready · 400ms}	`cortex-a9`	18.5 MB → 23.8 MB
⚠️ +28.4%	inceptionv1q _{RSS @ load · pass}	`cortex-a9`	24.1 MB → 30.9 MB
⚠️ +28.2%	parakeet_tdt_600m_v3_f32f32_preprocessor_1s _{RSS @ load · cuda}	`jetson-orin-nx`	21.3 MB → 27.3 MB
⚠️ +28.1%	voicecom_fake_quant _{RSS @ load · 2sec}	`i9-11900kb_rtx-4060`	27.1 MB → 34.6 MB
⚠️ +27.9%	mobilenet_v1_1 _{RSS @ ready · pass}	`cortex-a7`	36.9 MB → 47.2 MB
⚠️ +27.8%	dummy_conmer_12M _{RSS @ ready · pulse_120ms}	`beaglev-ahead`	69.2 MB → 88.4 MB
⚠️ +27.8%	en_tdnn_8M_nnef _{RSS @ ready · pulse_240ms}	`cortex-a9`	50.8 MB → 65 MB
⚠️ +27.7%	trunet _{RSS @ ready · pulse1_f32}	`beaglev-ahead`	20.4 MB → 26 MB
⚠️ +27.5%	nemotron_3_5_asr_streaming_0_6b_f32f32_preprocessor_1s _{RSS @ load · cuda}	`jetson-orin-nx`	21 MB → 26.8 MB
⚠️ +26.9%	openelm_270M_q40ef16_541 _{RSS @ ready · cpu}	`i9-11900kb_rtx-4060`	192 MB → 244 MB
⚠️ +26.9%	en_tdnn_8M _{RSS @ load · pulse_180ms}	`cortex-a55`	54.5 MB → 69.2 MB
⚠️ +26.6%	mdl_en_2019_Q3_librispeech_onnx _{RSS @ ready · 2600ms}	`cortex-a7`	31.4 MB → 39.7 MB
⚠️ +26.6%	mdl_en_2019_Q3_librispeech_onnx _{RSS @ load · 2600ms}	`i9-11900kb_rtx-4060`	36 MB → 45.5 MB
⚠️ +26.4%	en_tdnn_lstm_bn_q7 _{RSS @ ready · 2600ms}	`cortex-a9`	24.1 MB → 30.5 MB
⚠️ +26.3%	llama_3_2_1B_q40ef32_516 _{RSS @ ready · cuda}	`jetson-orin-nx`	356 MB → 450 MB
⚠️ +26.3%	en_tdnn_8M _{RSS @ load · pulse_180ms}	`cortex-a53`	54.3 MB → 68.6 MB
⚠️ +26.0%	arm_ml_kws_cnn_m _{RSS @ ready · pass}	`beaglev-ahead`	15.9 MB → 20 MB
⚠️ +25.6%	en_tdnn_8M_nnef _{RSS @ ready · pulse_240ms}	`cortex-a55`	55.5 MB → 69.7 MB
⚠️ +25.4%	trunet _{RSS @ load · pulse1_f32}	`orangepi-rv2`	19.2 MB → 24.1 MB
⚠️ +25.3%	en_tdnn_lstm_bn_q7 _{RSS @ ready · pulse_240ms}	`cortex-a7`	24.9 MB → 31.1 MB
⚠️ +25.3%	en_tdnn_8M _{RSS @ ready · pulse_120ms}	`cortex-a55`	63.3 MB → 79.3 MB
⚠️ +25.2%	en_tdnn_lstm_bn_q7 _{RSS @ ready · 2600ms}	`cortex-a7`	24 MB → 30.1 MB
⚠️ +25.1%	en_tdnn_pyt_15M _{RSS @ ready · pulse_120ms}	`cortex-a7`	65.3 MB → 81.7 MB
⚠️ +25.1%	en_tdnn_8M _{RSS @ ready · pulse_240ms}	`cortex-a53`	64.1 MB → 80.2 MB
⚠️ +25.0%	en_tdnn_8M _{RSS @ load · pulse_120ms}	`cortex-a7`	51.8 MB → 64.7 MB
⚠️ +25.0%	mdl_en_2019_Q3_librispeech_onnx _{RSS @ ready · pulse_240ms}	`cortex-a9`	32.3 MB → 40.4 MB
⚠️ +25.0%	inceptionv1q _{RSS @ ready · pass}	`i9-11900kb_rtx-4060`	43.4 MB → 54.2 MB
⚠️ +25.0%	mdl_en_2019_Q3_librispeech_onnx _{RSS @ ready · 2600ms}	`cortex-a9`	31.3 MB → 39.1 MB
⚠️ +24.8%	inceptionv3 _{RSS @ load · pass}	`beaglev-ahead`	114 MB → 142 MB
⚠️ +24.8%	trunet _{RSS @ load · pulse1_f32}	`beaglev-ahead`	19.2 MB → 24 MB
⚠️ +24.8%	mdl_en_2019_Q3_librispeech_onnx _{RSS @ ready · pulse_240ms}	`cortex-a7`	32.2 MB → 40.2 MB
⚠️ +24.8%	llama_3_1_8B_instruct_q40ef16_541 _{RSS @ ready · cuda}	`jetson-orin-nx`	508 MB → 634 MB
⚠️ +24.3%	nemotron_3_5_asr_streaming_0_6b_f32f32_decoder_pass _{RSS @ load · cuda}	`jetson-orin-nx`	75.7 MB → 94.1 MB
⚠️ +24.3%	en_tdnn_8M _{RSS @ load · pulse_180ms}	`cortex-a7`	51.6 MB → 64.2 MB
⚠️ +24.3%	llama_3_1_8B_instruct_q40ef16_541 _{RSS @ ready · cuda}	`i9-11900kb_rtx-4060`	690 MB → 857 MB
⚠️ +24.1%	inceptionv1q _{RSS @ load · pass}	`i9-11900kb_rtx-4060`	30.7 MB → 38.1 MB
⚠️ +24.1%	hey_snips_v31 _{RSS @ load · 400ms}	`beaglev-ahead`	14.6 MB → 18.1 MB
⚠️ +23.9%	en_tdnn_lstm_bn_q7 _{RSS @ load · pulse_240ms}	`cortex-a55`	24.6 MB → 30.5 MB
⚠️ +23.8%	en_tdnn_8M _{RSS @ load · pulse_240ms}	`orangepi-rv2`	49 MB → 60.7 MB
⚠️ +23.6%	mdl_en_2019_Q3_librispeech_onnx _{RSS @ load · pulse_240ms}	`i9-11900kb_rtx-4060`	37.6 MB → 46.5 MB
⚠️ +23.6%	en_tdnn_8M _{RSS @ load · pulse_240ms}	`cortex-a7`	52 MB → 64.2 MB
⚠️ +23.5%	en_tdnn_lstm_bn_q7 _{RSS @ ready · pulse_240ms}	`cortex-a9`	25.2 MB → 31.2 MB
⚠️ +23.5%	inceptionv3 _{RSS @ load · pass_mt}	`beaglev-ahead`	114 MB → 140 MB
⚠️ +23.4%	arm_ml_kws_cnn_m _{RSS @ ready · pass}	`orangepi-rv2`	16.5 MB → 20.3 MB
⚠️ +23.3%	en_tdnn_8M _{RSS @ load · 2600ms}	`cortex-a7`	51.3 MB → 63.3 MB
⚠️ +23.3%	speaker_id _{RSS @ load · pulse8}	`cortex-a7`	18.6 MB → 22.9 MB
⚠️ +23.2%	voicecom_float _{RSS @ load · 2sec}	`i9-11900kb_rtx-4060`	26.8 MB → 33 MB
⚠️ +23.0%	inceptionv3 _{RSS @ load · pass_mt}	`cortex-a9`	116 MB → 143 MB
⚠️ +23.0%	hey_snips_v31 _{RSS @ load · 400ms}	`orangepi-rv2`	14.6 MB → 18 MB
⚠️ +22.8%	qwen3_1_7B_q40ef16_541 _{RSS @ ready · cuda}	`jetson-orin-nx`	393 MB → 482 MB
⚠️ +22.8%	hey_snips_v1 _{RSS @ load · 400ms}	`orangepi-rv2`	13.7 MB → 16.8 MB
⚠️ +22.8%	inceptionv1q _{RSS @ load · pass}	`cortex-a53`	27.6 MB → 33.8 MB
⚠️ +22.6%	dummy_conmer_12M _{RSS @ load · pulse_120ms}	`beaglev-ahead`	67.8 MB → 83.2 MB
⚠️ +22.5%	hey_snips_v1 _{RSS @ load · 400ms}	`beaglev-ahead`	13.7 MB → 16.8 MB
⚠️ +22.4%	inceptionv3 _{RSS @ load · pass}	`orangepi-rv2`	113 MB → 139 MB
⚠️ +22.4%	en_tdnn_15M _{RSS @ ready · pulse_120ms}	`cortex-a9`	91.7 MB → 112 MB
⚠️ +22.4%	en_tdnn_15M _{RSS @ ready · 2600ms}	`cortex-a9`	91 MB → 111 MB
⚠️ +22.3%	mdl_en_2019_Q3_librispeech_onnx _{RSS @ load · pulse_240ms}	`cortex-a55`	32.5 MB → 39.8 MB
⚠️ +22.2%	en_tdnn_pyt_15M _{RSS @ load · pulse_120ms}	`cortex-a55`	74.2 MB → 90.7 MB
⚠️ +22.2%	en_tdnn_8M _{RSS @ load · pulse_240ms}	`cortex-a53`	54.5 MB → 66.6 MB
⚠️ +22.1%	en_tdnn_8M _{RSS @ load · 2600ms}	`cortex-a9`	51.6 MB → 63 MB
⚠️ +22.1%	inceptionv3 _{RSS @ ready · pass_mt}	`cortex-a7`	127 MB → 155 MB
⚠️ +22.1%	en_tdnn_8M_nnef _{RSS @ ready · pulse_240ms}	`i9-11900kb_rtx-4060`	59.5 MB → 72.6 MB
⚠️ +22.1%	en_tdnn_lstm_bn_q7 _{RSS @ load · pulse_240ms}	`cortex-a53`	24.5 MB → 30 MB
⚠️ +21.9%	hey_snips_v31 _{RSS @ ready · 400ms}	`beaglev-ahead`	15.8 MB → 19.3 MB
⚠️ +21.9%	inceptionv1q _{RSS @ load · pass}	`cortex-a55`	27.2 MB → 33.2 MB
⚠️ +21.8%	inceptionv3 _{RSS @ load · pass_mt}	`cortex-a7`	116 MB → 142 MB
⚠️ +21.7%	voicecom_fake_quant _{RSS @ load · 2sec}	`cortex-a53`	22.8 MB → 27.7 MB
⚠️ +21.7%	speaker_id _{RSS @ load · pulse8}	`cortex-a9`	18.4 MB → 22.3 MB
⚠️ +21.7%	en_tdnn_8M _{RSS @ load · pulse_120ms}	`cortex-a53`	54.7 MB → 66.5 MB
⚠️ +21.7%	en_tdnn_15M _{RSS @ ready · pulse_240ms}	`cortex-a9`	92 MB → 112 MB
⚠️ +21.6%	en_tdnn_pyt_15M _{RSS @ load · pulse_120ms}	`cortex-a53`	74.4 MB → 90.5 MB
⚠️ +21.6%	parakeet_tdt_600m_v3_f32f32_decoder_pass _{RSS @ load · cpu}	`i9-11900kb_rtx-4060`	69.4 MB → 84.4 MB
⚠️ +21.5%	inceptionv3 _{RSS @ ready · pass}	`cortex-a7`	126 MB → 153 MB
⚠️ +21.3%	dummy_conmer_12M _{RSS @ load · pulse_120ms}	`orangepi-rv2`	67.7 MB → 82.2 MB
⚠️ +21.0%	en_tdnn_8M _{RSS @ ready · 2600ms}	`cortex-a53`	64.7 MB → 78.3 MB
⚠️ +21.0%	hey_snips_v31 _{RSS @ ready · 400ms}	`orangepi-rv2`	15.8 MB → 19.1 MB
⚠️ +20.9%	inceptionv3 _{RSS @ load · pass}	`cortex-a7`	116 MB → 141 MB
⚠️ +20.9%	en_tdnn_lstm_bn_q7 _{RSS @ load · pulse_240ms}	`i9-11900kb_rtx-4060`	29 MB → 35.1 MB
⚠️ +20.8%	en_tdnn_8M _{RSS @ ready · pulse_120ms}	`cortex-a53`	63.9 MB → 77.2 MB
⚠️ +20.7%	hey_snips_v4_model17_nnef _{RSS @ ready · pulse8}	`i9-11900kb_rtx-4060`	27.9 MB → 33.7 MB
⚠️ +20.6%	hey_snips_v4_model17 _{RSS @ load · 2sec}	`cortex-a9`	24.1 MB → 29.1 MB
⚠️ +20.6%	en_tdnn_lstm_bn_q7 _{RSS @ load · 2600ms}	`cortex-a55`	24.3 MB → 29.3 MB
⚠️ +20.5%	parakeet_tdt_600m_v3_f32f32_decoder_pass _{RSS @ load · cuda}	`i9-11900kb_rtx-4060`	69.2 MB → 83.5 MB
⚠️ +20.5%	mobilenet_v2_1 _{RSS @ ready · pass_mt}	`cortex-a9`	45.7 MB → 55.1 MB
⚠️ +20.4%	inceptionv1q _{RSS @ ready · pass}	`cortex-a9`	30.2 MB → 36.3 MB
⚠️ +20.4%	inceptionv3 _{RSS @ ready · pass}	`cortex-a9`	127 MB → 153 MB
⚠️ +20.4%	trunet _{RSS @ load · pulse1_f16}	`i9-11900kb_rtx-4060`	27.6 MB → 33.2 MB
⚠️ +20.3%	en_tdnn_pyt_15M _{RSS @ ready · pulse_120ms}	`cortex-a53`	80.3 MB → 96.6 MB
⚠️ +20.3%	hey_snips_v4_model17 _{RSS @ ready · 2sec}	`cortex-a55`	32.6 MB → 39.2 MB
⚠️ +20.2%	en_tdnn_lstm_bn_q7 _{RSS @ load · 2600ms}	`i9-11900kb_rtx-4060`	28.1 MB → 33.8 MB
⚠️ +20.2%	en_tdnn_lstm_bn_q7 _{RSS @ load · 2600ms}	`cortex-a53`	24.3 MB → 29.2 MB
⚠️ +20.1%	hey_snips_v4_model17 _{RSS @ load · 2sec}	`cortex-a7`	24.1 MB → 28.9 MB
⚠️ +20.0%	mdl_en_2019_Q3_librispeech_onnx _{RSS @ load · 2600ms}	`cortex-a55`	32.3 MB → 38.7 MB
⚠️ +20.0%	hey_snips_v1 _{RSS @ load · 400ms}	`cortex-a7`	16.7 MB → 20 MB
⚠️ +20.0%	en_tdnn_8M _{RSS @ load · 2600ms}	`cortex-a53`	54.6 MB → 65.5 MB
⚠️ +19.7%	en_tdnn_8M _{RSS @ load · pulse_120ms}	`cortex-a55`	54.4 MB → 65.1 MB
⚠️ +19.7%	inceptionv3 _{RSS @ ready · pass}	`beaglev-ahead`	119 MB → 142 MB
⚠️ +19.5%	hey_snips_v4_model17 _{RSS @ load · pulse8}	`cortex-a9`	24.3 MB → 29 MB
⚠️ +19.5%	hey_snips_v4_model17_nnef _{RSS @ load · pulse8}	`i9-11900kb_rtx-4060`	24.5 MB → 29.3 MB
⚠️ +19.4%	en_tdnn_8M _{RSS @ ready · pulse_180ms}	`cortex-a55`	64.1 MB → 76.6 MB
⚠️ +19.3%	inceptionv3 _{RSS @ ready · pass_mt}	`cortex-a9`	126 MB → 151 MB
⚠️ +19.3%	inceptionv3 _{RSS @ load · pass_mt}	`orangepi-rv2`	113 MB → 135 MB
⚠️ +19.2%	speaker_id _{RSS @ load · pulse8}	`orangepi-rv2`	15.9 MB → 19 MB
⚠️ +19.2%	qwen3_1_7B_q40ef16_541 _{RSS @ ready · cuda}	`i9-11900kb_rtx-4060`	575 MB → 685 MB
⚠️ +19.0%	hey_snips_v1 _{RSS @ load · 400ms}	`cortex-a9`	16.7 MB → 19.8 MB
⚠️ +19.0%	en_tdnn_8M _{RSS @ load · pulse_180ms}	`beaglev-ahead`	49 MB → 58.3 MB
⚠️ +18.9%	en_tdnn_pyt_15M _{RSS @ ready · pulse_120ms}	`cortex-a9`	65.8 MB → 78.2 MB
⚠️ +18.7%	en_tdnn_8M _{RSS @ load · pulse_120ms}	`beaglev-ahead`	49.1 MB → 58.3 MB
⚠️ +18.7%	en_tdnn_8M _{RSS @ ready · pulse_240ms}	`cortex-a55`	64 MB → 75.9 MB
⚠️ +18.6%	inceptionv3 _{RSS @ load · pass}	`cortex-a9`	116 MB → 138 MB
⚠️ +18.5%	speaker_id _{RSS @ load · pulse8}	`beaglev-ahead`	16 MB → 19 MB
⚠️ +18.5%	en_tdnn_8M _{RSS @ load · 2600ms}	`beaglev-ahead`	48.2 MB → 57.1 MB
⚠️ +18.4%	hey_snips_v31 _{RSS @ load · 400ms}	`i9-11900kb_rtx-4060`	23.6 MB → 27.9 MB
⚠️ +18.3%	en_tdnn_15M _{RSS @ load · pulse_120ms}	`beaglev-ahead`	87.3 MB → 103 MB
⚠️ +18.3%	voicecom_float _{RSS @ load · 2sec}	`cortex-a55`	23.8 MB → 28.1 MB
⚠️ +18.3%	trunet _{RSS @ load · pulse1_f16}	`cortex-a7`	21.7 MB → 25.7 MB
⚠️ +18.3%	trunet _{RSS @ load · pulse1_f16}	`cortex-a9`	21.8 MB → 25.8 MB
⚠️ +18.2%	en_tdnn_8M _{RSS @ load · pulse_180ms}	`orangepi-rv2`	49 MB → 57.9 MB
⚠️ +18.1%	en_tdnn_8M _{RSS @ load · 2600ms}	`cortex-a55`	53.8 MB → 63.5 MB
⚠️ +18.1%	en_tdnn_15M _{RSS @ ready · pulse_120ms}	`beaglev-ahead`	88.4 MB → 104 MB
⚠️ +18.0%	hey_snips_v4_model17_nnef _{RSS @ ready · pulse8}	`cortex-a9`	19.8 MB → 23.3 MB
⚠️ +17.9%	en_tdnn_15M _{RSS @ load · pulse_240ms}	`cortex-a9`	90.2 MB → 106 MB
⚠️ +17.8%	llama_3_2_1B_instruct_q40ef16_541 _{RSS @ ready · cuda}	`i9-11900kb_rtx-4060`	538 MB → 634 MB
⚠️ +17.7%	en_tdnn_15M _{RSS @ load · pulse_120ms}	`cortex-a9`	89.5 MB → 105 MB
⚠️ +17.7%	speaker_id _{RSS @ ready · pulse8}	`beaglev-ahead`	17 MB → 20 MB
⚠️ +17.7%	speaker_id _{RSS @ ready · pulse8}	`orangepi-rv2`	17 MB → 20.1 MB
⚠️ +17.7%	inceptionv3 _{RSS @ ready · pass_mt}	`beaglev-ahead`	119 MB → 140 MB
⚠️ +17.6%	en_tdnn_15M _{RSS @ ready · pulse_120ms}	`orangepi-rv2`	88 MB → 103 MB
⚠️ +17.5%	hey_snips_v4_model17 _{RSS @ ready · 2sec}	`cortex-a7`	25.6 MB → 30.1 MB
⚠️ +17.4%	trunet _{RSS @ load · pulse1_f32}	`cortex-a7`	21.7 MB → 25.5 MB
⚠️ +17.4%	hey_snips_v4_model17 _{RSS @ ready · 2sec}	`cortex-a9`	25.7 MB → 30.1 MB
⚠️ +17.3%	en_tdnn_15M _{RSS @ load · pulse_120ms}	`orangepi-rv2`	87.3 MB → 102 MB
⚠️ +17.2%	en_tdnn_15M _{RSS @ load · 2600ms}	`beaglev-ahead`	86.3 MB → 101 MB
⚠️ +17.1%	en_tdnn_15M _{RSS @ load · 2600ms}	`orangepi-rv2`	86.4 MB → 101 MB
⚠️ +17.1%	voicecom_float _{RSS @ load · 2sec}	`cortex-a53`	23.5 MB → 27.5 MB
⚠️ +17.1%	en_tdnn_15M _{RSS @ load · pulse_240ms}	`beaglev-ahead`	87.3 MB → 102 MB
⚠️ +16.9%	nemotron_3_5_asr_streaming_0_6b_f32f32_decoder_pass _{RSS @ load · cpu}	`i9-11900kb_rtx-4060`	81.6 MB → 95.4 MB
⚠️ +16.8%	hey_snips_v31 _{RSS @ ready · 400ms}	`cortex-a53`	21.1 MB → 24.6 MB
⚠️ +16.6%	en_tdnn_15M_nnef _{RSS @ ready · pulse_240ms}	`cortex-a53`	91.6 MB → 107 MB
⚠️ +16.6%	en_tdnn_15M _{RSS @ load · 2600ms}	`cortex-a9`	89.8 MB → 105 MB
⚠️ +16.5%	en_tdnn_lstm_bn_q7 _{RSS @ ready · pulse_240ms}	`i9-11900kb_rtx-4060`	31.9 MB → 37.2 MB
⚠️ +16.5%	nemotron_3_5_asr_streaming_0_6b_f32f32_decoder_pass _{RSS @ load · cuda}	`i9-11900kb_rtx-4060`	81.9 MB → 95.4 MB
⚠️ +16.5%	en_tdnn_8M _{RSS @ ready · pulse_180ms}	`cortex-a53`	63.7 MB → 74.2 MB
⚠️ +16.4%	en_tdnn_15M _{RSS @ load · pulse_240ms}	`cortex-a7`	90.3 MB → 105 MB
⚠️ +16.1%	arm_ml_kws_cnn_m _{RSS @ load · pass}	`i9-11900kb_rtx-4060`	23.4 MB → 27.2 MB
⚠️ +16.1%	speaker_id _{RSS @ load · pulse8}	`cortex-a53`	19.5 MB → 22.7 MB
⚠️ +16.1%	mobilenet_v2_1 _{RSS @ ready · pass}	`cortex-a55`	52.9 MB → 61.4 MB
⚠️ +15.9%	trunet _{RSS @ ready · pulse1_f16}	`cortex-a55`	26 MB → 30.1 MB
⚠️ +15.8%	en_tdnn_15M _{RSS @ load · pulse_240ms}	`orangepi-rv2`	87.3 MB → 101 MB
⚠️ +15.8%	speaker_id _{RSS @ load · pulse8}	`cortex-a55`	19.7 MB → 22.8 MB
⚠️ +15.7%	arm_ml_kws_cnn_m _{RSS @ ready · pass}	`i9-11900kb_rtx-4060`	26.5 MB → 30.7 MB
⚠️ +15.6%	voicecom_fake_quant _{RSS @ ready · 2sec}	`cortex-a55`	33 MB → 38.2 MB
⚠️ +15.6%	en_tdnn_15M _{RSS @ ready · pulse_240ms}	`orangepi-rv2`	88 MB → 102 MB
⚠️ +15.5%	en_tdnn_15M _{RSS @ load · 2600ms}	`cortex-a7`	90.2 MB → 104 MB
⚠️ +15.5%	mobilenet_v2_1 _{RSS @ ready · pass}	`cortex-a53`	52.8 MB → 61 MB
⚠️ +15.4%	hey_snips_v4_model17 _{RSS @ load · pulse8}	`cortex-a7`	24.3 MB → 28 MB
⚠️ +15.4%	en_tdnn_8M _{RSS @ ready · 2600ms}	`i9-11900kb_rtx-4060`	73.3 MB → 84.6 MB
⚠️ +15.4%	openelm_270M_q40ef16_541 _{RSS @ load · cuda}	`jetson-orin-nx`	200 MB → 231 MB
⚠️ +15.3%	en_tdnn_15M _{RSS @ ready · pulse_240ms}	`cortex-a7`	92.1 MB → 106 MB
⚠️ +15.2%	hey_snips_v1 _{RSS @ ready · 400ms}	`i9-11900kb_rtx-4060`	25.7 MB → 29.6 MB
⚠️ +15.2%	hey_snips_v1 _{load · 400ms}	`cortex-a7`	66 ms → 76 ms
⚠️ +15.1%	en_tdnn_15M _{RSS @ load · pulse_120ms}	`cortex-a7`	90.4 MB → 104 MB
⚠️ +15.1%	hey_snips_v31 _{RSS @ load · 400ms}	`cortex-a9`	18.3 MB → 21 MB
⚠️ +15.0%	trunet _{RSS @ load · pulse1_f32}	`cortex-a55`	24.6 MB → 28.3 MB
⚠️ +15.0%	trunet _{RSS @ ready · pulse1_f16}	`cortex-a7`	23 MB → 26.5 MB
⚠️ +14.9%	parakeet_tdt_600m_v3_f32f32_preprocessor_1s _{RSS @ load · cuda}	`i9-11900kb_rtx-4060`	23.8 MB → 27.3 MB
⚠️ +14.9%	hey_snips_v4_model17_nnef _{RSS @ ready · pulse8}	`cortex-a7`	19.7 MB → 22.7 MB
⚠️ +14.8%	parakeet_tdt_600m_v3_f32f32_preprocessor_1s _{RSS @ load · cpu}	`i9-11900kb_rtx-4060`	23.7 MB → 27.2 MB
⚠️ +14.8%	llama_3_2_1B_q40ef32_516 _{RSS @ ready · cuda}	`i9-11900kb_rtx-4060`	539 MB → 619 MB
⚠️ +14.7%	hey_snips_v31 _{RSS @ ready · 400ms}	`cortex-a55`	21.3 MB → 24.4 MB
⚠️ +14.7%	inceptionv1q _{RSS @ ready · pass}	`cortex-a55`	40.6 MB → 46.6 MB
⚠️ +14.6%	mdl_en_2019_Q3_librispeech_onnx _{RSS @ ready · 2600ms}	`i9-11900kb_rtx-4060`	41.4 MB → 47.4 MB
⚠️ +14.5%	hey_snips_v4_model17_nnef _{RSS @ ready · pulse8}	`cortex-a55`	24.6 MB → 28.2 MB
⚠️ +14.5%	nemotron_3_5_asr_streaming_0_6b_f32f32_preprocessor_pulse100ms _{RSS @ ready · cpu}	`i9-11900kb_rtx-4060`	26.4 MB → 30.2 MB
⚠️ +14.5%	trunet _{RSS @ ready · pulse1_f16}	`cortex-a53`	26.2 MB → 30 MB
⚠️ +14.5%	en_tdnn_lstm_bn_q7 _{RSS @ ready · 2600ms}	`i9-11900kb_rtx-4060`	31.7 MB → 36.3 MB
⚠️ +14.4%	mdl_en_2019_Q3_librispeech_onnx _{RSS @ load · 2600ms}	`cortex-a53`	32.5 MB → 37.2 MB
⚠️ +14.4%	trunet _{RSS @ ready · pulse1_f16}	`i9-11900kb_rtx-4060`	30.4 MB → 34.8 MB
⚠️ +14.4%	nemotron_3_5_asr_streaming_0_6b_f32f32_preprocessor_pulse100ms _{RSS @ load · cpu}	`i9-11900kb_rtx-4060`	25.2 MB → 28.8 MB
⚠️ +14.2%	parakeet_tdt_600m_v3_f32f32_joint_pass _{RSS @ ready · cpu}	`i9-11900kb_rtx-4060`	51.3 MB → 58.6 MB
⚠️ +14.1%	arm_ml_kws_cnn_m _{RSS @ ready · pass}	`cortex-a9`	20.4 MB → 23.3 MB
⚠️ +14.1%	hey_snips_v1 _{RSS @ load · 400ms}	`i9-11900kb_rtx-4060`	22.6 MB → 25.7 MB
⚠️ +14.1%	arm_ml_kws_cnn_m _{RSS @ load · pass}	`orangepi-rv2`	14.6 MB → 16.6 MB
⚠️ +14.1%	hey_snips_v4_model17 _{RSS @ load · pulse8}	`i9-11900kb_rtx-4060`	36.1 MB → 41.2 MB
⚠️ +14.1%	en_tdnn_8M _{RSS @ load · pulse_240ms}	`i9-11900kb_rtx-4060`	65.6 MB → 74.8 MB
⚠️ +14.0%	en_tdnn_15M _{RSS @ ready · pulse_120ms}	`cortex-a7`	92 MB → 105 MB
⚠️ +14.0%	dummy_conmer_12M _{RSS @ load · pulse_120ms}	`cortex-a55`	79.4 MB → 90.5 MB
⚠️ +14.0%	en_tdnn_15M _{RSS @ ready · 2600ms}	`orangepi-rv2`	87.2 MB → 99.4 MB
⚠️ +14.0%	speaker_id _{RSS @ load · pulse8}	`i9-11900kb_rtx-4060`	24.4 MB → 27.8 MB
⚠️ +13.9%	nemotron_3_5_asr_streaming_0_6b_f32f32_preprocessor_pulse100ms _{load · cpu}	`i9-11900kb_rtx-4060`	36 ms → 41 ms
⚠️ +13.9%	hey_snips_v1 _{RSS @ ready · 400ms}	`cortex-a55`	22.1 MB → 25.1 MB
⚠️ +13.8%	hey_snips_v31 _{RSS @ ready · 400ms}	`i9-11900kb_rtx-4060`	26.1 MB → 29.7 MB
⚠️ +13.8%	trunet _{RSS @ ready · pulse1_f16}	`cortex-a9`	23.3 MB → 26.6 MB
⚠️ +13.7%	mdl_en_2019_Q3_librispeech_onnx _{RSS @ ready · 2600ms}	`cortex-a55`	35.1 MB → 40 MB
⚠️ +13.6%	hey_snips_v4_model17_nnef _{RSS @ ready · pulse8}	`cortex-a53`	24.5 MB → 27.9 MB
⚠️ +13.5%	en_tdnn_8M _{RSS @ load · pulse_180ms}	`i9-11900kb_rtx-4060`	64.4 MB → 73.1 MB
⚠️ +13.5%	nemotron_3_5_asr_streaming_0_6b_f32f32_preprocessor_pulse100ms _{load+optimize · cpu}	`i9-11900kb_rtx-4060`	37 ms → 42 ms
⚠️ +13.5%	nemotron_3_5_asr_streaming_0_6b_f32f32_preprocessor_1s _{RSS @ load · cpu}	`i9-11900kb_rtx-4060`	23.5 MB → 26.7 MB
⚠️ +13.4%	dummy_conmer_12M _{RSS @ ready · pulse_120ms}	`cortex-a7`	76.6 MB → 86.9 MB
⚠️ +13.4%	en_tdnn_15M _{RSS @ ready · 2600ms}	`beaglev-ahead`	87.6 MB → 99.4 MB
⚠️ +13.4%	en_tdnn_15M _{RSS @ ready · pulse_240ms}	`beaglev-ahead`	88.4 MB → 100 MB
⚠️ +13.2%	trunet _{RSS @ load · pulse1_f32}	`cortex-a9`	22.5 MB → 25.5 MB
⚠️ +13.2%	en_tdnn_8M _{RSS @ load · pulse_240ms}	`cortex-a55`	54.3 MB → 61.5 MB
⚠️ +13.1%	trunet _{RSS @ load · pulse1_f32}	`cortex-a53`	25.2 MB → 28.5 MB
⚠️ +13.0%	trunet _{RSS @ load · pulse1_f32}	`i9-11900kb_rtx-4060`	29.4 MB → 33.3 MB
⚠️ +13.0%	en_tdnn_lstm_bn_q7 _{RSS @ ready · 2600ms}	`cortex-a53`	26.7 MB → 30.2 MB
⚠️ +12.8%	mdl_en_2019_Q3_librispeech_onnx _{RSS @ ready · pulse_240ms}	`cortex-a55`	36.1 MB → 40.7 MB
⚠️ +12.8%	en_tdnn_8M _{RSS @ ready · pulse_180ms}	`i9-11900kb_rtx-4060`	72.6 MB → 81.9 MB
⚠️ +12.8%	hey_snips_v4_model17 _{RSS @ ready · pulse8}	`cortex-a7`	26.6 MB → 30 MB
⚠️ +12.8%	en_tdnn_lstm_bn_q7 _{RSS @ ready · pulse_240ms}	`cortex-a55`	27.9 MB → 31.4 MB
⚠️ +12.8%	mdl_en_2019_Q3_librispeech_onnx _{RSS @ ready · pulse_240ms}	`i9-11900kb_rtx-4060`	42.4 MB → 47.8 MB
⚠️ +12.7%	nemotron_3_5_asr_streaming_0_6b_f32f32_preprocessor_1s _{RSS @ load · cuda}	`i9-11900kb_rtx-4060`	23.8 MB → 26.8 MB
⚠️ +12.7%	arm_ml_kws_cnn_m _{RSS @ ready · pass}	`cortex-a7`	20.6 MB → 23.2 MB
⚠️ +12.6%	hey_snips_v1 _{RSS @ ready · 400ms}	`cortex-a53`	21.9 MB → 24.7 MB
⚠️ +12.6%	en_tdnn_15M _{RSS @ load · pulse_120ms}	`cortex-a55`	94.8 MB → 107 MB
⚠️ +12.6%	nemotron_3_5_asr_streaming_0_6b_f32f32_joint_pass _{RSS @ ready · cpu}	`i9-11900kb_rtx-4060`	64.5 MB → 72.7 MB
⚠️ +12.5%	en_tdnn_lstm_bn_q7 _{RSS @ ready · 2600ms}	`cortex-a55`	27 MB → 30.3 MB
⚠️ +12.5%	en_tdnn_15M _{RSS @ load · pulse_120ms}	`cortex-a53`	95 MB → 107 MB
⚠️ +12.2%	trunet _{RSS @ ready · pulse1_f32}	`cortex-a7`	25.1 MB → 28.2 MB
⚠️ +12.1%	parakeet_tdt_600m_v3_f32f32_preprocessor_1s _{RSS @ ready · cpu}	`i9-11900kb_rtx-4060`	25.8 MB → 29 MB
⚠️ +12.1%	en_tdnn_15M _{load · 2600ms}	`cortex-a53`	1.46 s → 1.64 s
⚠️ +12.1%	en_tdnn_15M _{RSS @ load · 2600ms}	`cortex-a55`	94.9 MB → 106 MB
⚠️ +12.1%	en_tdnn_15M _{RSS @ ready · 2600ms}	`cortex-a7`	91.4 MB → 102 MB
⚠️ +12.0%	en_tdnn_8M _{RSS @ ready · pulse_120ms}	`i9-11900kb_rtx-4060`	75.5 MB → 84.6 MB
⚠️ +12.0%	en_tdnn_8M _{RSS @ ready · pulse_240ms}	`i9-11900kb_rtx-4060`	73.8 MB → 82.7 MB
⚠️ +12.0%	speaker_id _{RSS @ ready · pulse8}	`i9-11900kb_rtx-4060`	26.5 MB → 29.7 MB
⚠️ +11.9%	hey_snips_v4_model17_nnef _{RSS @ load · pulse8}	`cortex-a55`	22.3 MB → 25 MB
⚠️ +11.9%	en_tdnn_15M _{RSS @ load · pulse_240ms}	`cortex-a53`	94.8 MB → 106 MB
⚠️ +11.9%	hey_snips_v4_model17 _{RSS @ ready · 2sec}	`cortex-a53`	32.7 MB → 36.6 MB
⚠️ +11.9%	arm_ml_kws_cnn_m _{RSS @ load · pass}	`cortex-a9`	18 MB → 20.1 MB
⚠️ +11.8%	arm_ml_kws_cnn_m _{RSS @ load · pass}	`cortex-a7`	18.1 MB → 20.2 MB
⚠️ +11.5%	openelm_270M_q40ef16_541 _{RSS @ load · cuda}	`i9-11900kb_rtx-4060`	206 MB → 230 MB
⚠️ +11.5%	en_tdnn_15M _{RSS @ load · pulse_240ms}	`cortex-a55`	95.1 MB → 106 MB
⚠️ +11.5%	openelm_270M_q40ef16_541 _{RSS @ load · cpu}	`i9-11900kb_rtx-4060`	206 MB → 230 MB
⚠️ +11.5%	nemotron_3_5_asr_streaming_0_6b_f32f32_preprocessor_1s _{RSS @ ready · cpu}	`i9-11900kb_rtx-4060`	25.3 MB → 28.2 MB
⚠️ +11.3%	voicecom_float _{RSS @ ready · 2sec}	`cortex-a55`	31.5 MB → 35.1 MB
⚠️ +11.3%	trunet _{RSS @ ready · pulse1_f32}	`cortex-a55`	27.5 MB → 30.6 MB
⚠️ +11.2%	en_tdnn_15M_nnef _{RSS @ ready · pulse_240ms}	`orangepi-rv2`	85.7 MB → 95.3 MB
⚠️ +11.2%	mdl_en_2019_Q3_librispeech_onnx _{RSS @ ready · 2600ms}	`cortex-a53`	35.2 MB → 39.1 MB
⚠️ +11.0%	en_tdnn_15M _{RSS @ load · 2600ms}	`cortex-a53`	94.8 MB → 105 MB
⚠️ +11.0%	nemotron_3_5_asr_streaming_0_6b_f32f32_preprocessor_pulse100ms _{RSS @ load · cuda}	`i9-11900kb_rtx-4060`	25.5 MB → 28.3 MB
⚠️ +10.9%	trunet _{RSS @ load · pulse1_f16}	`cortex-a53`	25.8 MB → 28.6 MB
⚠️ +10.6%	hey_snips_v1 _{RSS @ load · 400ms}	`cortex-a55`	19.8 MB → 21.9 MB
⚠️ +10.5%	hey_snips_v1 _{RSS @ load · 400ms}	`cortex-a53`	19.5 MB → 21.5 MB
⚠️ +10.4%	hey_snips_v4_model17_nnef _{RSS @ load · pulse8}	`cortex-a53`	22.2 MB → 24.5 MB
⚠️ +10.3%	dummy_conmer_12M _{RSS @ ready · pulse_120ms}	`cortex-a9`	77.1 MB → 85 MB
⚠️ +10.3%	mobilenet_v2_1 _{RSS @ load · pass_mt}	`cortex-a53`	67.5 MB → 74.4 MB
⚠️ +10.3%	dummy_conmer_12M _{RSS @ load · pulse_120ms}	`cortex-a53`	79.7 MB → 87.8 MB
⚠️ +9.9%	voicecom_float _{RSS @ ready · 2sec}	`cortex-a53`	31.5 MB → 34.6 MB
⚠️ +9.8%	hey_snips_v4_model17 _{RSS @ ready · pulse8}	`cortex-a9`	26.9 MB → 29.5 MB
⚠️ +9.8%	en_tdnn_15M _{load · pulse_120ms}	`cortex-a53`	1.72 s → 1.89 s
⚠️ +9.7%	en_tdnn_15M _{load · pulse_240ms}	`cortex-a53`	1.71 s → 1.88 s
⚠️ +9.7%	en_tdnn_15M_nnef _{RSS @ ready · pulse_240ms}	`i9-11900kb_rtx-4060`	96 MB → 105 MB
⚠️ +9.6%	inceptionv1q _{RSS @ ready · pass}	`cortex-a7`	30.2 MB → 33.1 MB
⚠️ +9.5%	trunet _{RSS @ ready · pulse1_f32}	`cortex-a53`	27.4 MB → 30 MB
⚠️ +9.4%	en_tdnn_15M _{load+optimize · 2600ms}	`cortex-a53`	1.78 s → 1.94 s
⚠️ +9.3%	dummy_conmer_12M _{RSS @ load · pulse_120ms}	`cortex-a9`	73.6 MB → 80.4 MB
⚠️ +9.3%	en_tdnn_15M _{load · 2600ms}	`cortex-a55`	1.06 s → 1.15 s
⚠️ +9.2%	en_tdnn_15M_nnef _{RSS @ ready · pulse_240ms}	`cortex-a55`	91.5 MB → 99.9 MB
⚠️ +9.1%	en_tdnn_15M_nnef _{RSS @ ready · pulse_240ms}	`beaglev-ahead`	86.1 MB → 94 MB
⚠️ +8.8%	en_tdnn_15M _{load · pulse_120ms}	`cortex-a55`	1.25 s → 1.36 s
⚠️ +8.6%	mdl_en_2019_Q3_librispeech_onnx _{RSS @ load · pulse_240ms}	`cortex-a53`	32.5 MB → 35.2 MB
⚠️ +8.5%	dummy_conmer_12M _{RSS @ load · pulse_120ms}	`i9-11900kb_rtx-4060`	85.9 MB → 93.2 MB
⚠️ +8.3%	dummy_conmer_12M _{RSS @ load · pulse_120ms}	`cortex-a7`	73.7 MB → 79.8 MB
⚠️ +7.6%	en_tdnn_15M _{load+optimize · pulse_120ms}	`cortex-a53`	2.04 s → 2.2 s
⚠️ +7.6%	en_tdnn_15M _{load · 2600ms}	`cortex-a9`	2.5 s → 2.69 s
⚠️ +7.5%	en_tdnn_15M _{load+optimize · pulse_240ms}	`cortex-a53`	2.03 s → 2.18 s
⚠️ +7.4%	en_tdnn_15M_nnef _{RSS @ ready · pulse_240ms}	`cortex-a7`	89 MB → 95.6 MB
⚠️ +7.2%	en_tdnn_8M_nnef _{RSS @ load · pulse_240ms}	`i9-11900kb_rtx-4060`	54.4 MB → 58.3 MB
⚠️ +7.1%	en_tdnn_15M_nnef _{RSS @ ready · pulse_240ms}	`cortex-a9`	89.4 MB → 95.8 MB
⚠️ +7.0%	arm_ml_kws_cnn_m _{RSS @ load · pass}	`cortex-a55`	20.3 MB → 21.7 MB
⚠️ +6.6%	arm_ml_kws_cnn_m _{RSS @ ready · pass}	`cortex-a55`	23.4 MB → 25 MB
⚠️ +6.5%	en_tdnn_15M _{load+optimize · 2600ms}	`cortex-a55`	1.26 s → 1.34 s
⚠️ +5.8%	llama_3_2_1B_q40ef32_516 _{RSS @ ready · cpu}	`i9-11900kb_rtx-4060`	728 MB → 770 MB
⚠️ +5.6%	en_tdnn_8M_nnef _{RSS @ load · pulse_240ms}	`orangepi-rv2`	46.4 MB → 49 MB
⚠️ +5.5%	en_tdnn_8M_nnef _{RSS @ load · pulse_240ms}	`beaglev-ahead`	46.5 MB → 49.1 MB
⚠️ +5.3%	en_tdnn_15M _{load · pulse_240ms}	`cortex-a9`	2.87 s → 3.02 s
⚠️ +5.3%	llama_3_2_1B_instruct_q40ef16_541 _{RSS @ ready · cpu_mt}	`i9-11900kb_rtx-4060`	727 MB → 765 MB
⚠️ +5.2%	arm_ml_kws_cnn_m _{RSS @ load · pass}	`apple-m1-max`	18.6 MB → 19.5 MB
⚠️ +5.1%	en_tdnn_15M _{load · pulse_120ms}	`cortex-a9`	2.87 s → 3.02 s
⚠️ +5.0%	mobilenet_v1_1 _{RSS @ load · pass}	`cortex-a53`	48.1 MB → 50.6 MB

kali · 2026-09-08T07:35:43Z

⚠️⚠️⚠️ Just rebased! ⚠️⚠️⚠️

github-actions · 2026-09-08T08:42:57Z

🔴 Bench vs main — 8 speed regression(s) · ⚠️ 390 secondary

Reference: 2026-09-08 morning nightly run (0d old) · full report → run

Speed — evaltime · prefill · decode

Δ	metric	device	main → PR
🔴 +8.1%	hey_snips_v31 _{evaltime · 400ms}	`cortex-a55`	1.32 ms → 1.43 ms
🔴 +8.0%	hey_snips_v1 _{evaltime · 400ms}	`cortex-a55`	2.18 ms → 2.35 ms
🔴 +7.7%	inceptionv3 _{evaltime · pass_mt}	`apple-m1-max`	45.9 ms → 49.5 ms
🔴 +7.3%	mobilenet_v1_1 _{evaltime · pass}	`apple-m1-max`	18.1 ms → 19.4 ms
🔴 +7.1%	inceptionv3 _{evaltime · pass}	`apple-m1-max`	59.3 ms → 63.5 ms

+3 more regression(s)

Δ	metric	device	main → PR
🔴 +5.4%	hey_snips_v4_model17 _{evaltime · 2sec}	`cortex-a53`	16.1 ms → 17 ms
🔴 +5.4%	hey_snips_v31 _{evaltime · 400ms}	`cortex-a53`	2.08 ms → 2.2 ms
🔴 +5.1%	llama_3_2_1B_instruct_q40ef16_541 _{prefill · cpu_mt}	`apple-m1-max`	1.87 ms/tok _{534.4 tok/s} → 1.97 ms/tok _{508.4 tok/s}

Improvements

Δ	metric	device	main → PR
🟢 -22.2%	speaker_id _{evaltime · pulse8}	`beaglev-ahead`	3.17 ms/pulse _{0.0396 RTF} → 2.46 ms/pulse _{0.0308 RTF}
🟢 -20.9%	hey_snips_v4_model17 _{evaltime · pulse8}	`beaglev-ahead`	5.9 ms/pulse _{0.0738 RTF} → 4.67 ms/pulse _{0.0584 RTF}
🟢 -15.4%	hey_snips_v4_model17 _{evaltime · 2sec}	`beaglev-ahead`	49 ms → 41.4 ms
🟢 -14.3%	trunet _{evaltime · pulse1_f32}	`beaglev-ahead`	5.66 ms/pulse _{0.566 RTF} → 4.85 ms/pulse _{0.485 RTF}
🟢 -13.0%	mobilenet_v2_1 _{evaltime · pass_mt}	`beaglev-ahead`	505 ms → 439 ms

+29 more improvement(s)

Δ	metric	device	main → PR
🟢 -11.3%	hey_snips_v4_model17_nnef _{evaltime · pulse8}	`beaglev-ahead`	5.23 ms/pulse _{0.0654 RTF} → 4.64 ms/pulse _{0.058 RTF}
🟢 -9.3%	hey_snips_v4_model17_nnef _{evaltime · pulse8}	`cortex-a9`	6.33 ms/pulse _{0.0791 RTF} → 5.74 ms/pulse _{0.0717 RTF}
🟢 -8.8%	inceptionv3 _{evaltime · pass_mt}	`beaglev-ahead`	2.24 s → 2.04 s
🟢 -8.6%	hey_snips_v4_model17 _{evaltime · pulse8}	`cortex-a9`	6.32 ms/pulse _{0.079 RTF} → 5.77 ms/pulse _{0.0722 RTF}
🟢 -8.2%	hey_snips_v4_model17 _{evaltime · pulse8}	`cortex-a7`	6.94 ms/pulse _{0.0867 RTF} → 6.37 ms/pulse _{0.0796 RTF}
🟢 -8.1%	hey_snips_v4_model17_nnef _{evaltime · pulse8}	`cortex-a7`	6.94 ms/pulse _{0.0868 RTF} → 6.38 ms/pulse _{0.0797 RTF}
🟢 -8.1%	hey_snips_v4_model17 _{evaltime · pulse8}	`cortex-a55`	2.11 ms/pulse _{0.0264 RTF} → 1.94 ms/pulse _{0.0242 RTF}
🟢 -7.4%	openelm_270M_q40ef16_541 _{decode · cuda}	`i9-11900kb_rtx-4060`	2.14 ms/tok _{466.3 tok/s} → 1.98 ms/tok _{503.8 tok/s}
🟢 -7.2%	openelm_270M_q40ef16_541 _{decode · metal}	`apple-m1-max`	4.54 ms/tok _{220.3 tok/s} → 4.21 ms/tok _{237.3 tok/s}
🟢 -6.9%	trunet _{evaltime · pulse1_f32}	`cortex-a9`	8.1 ms/pulse _{0.81 RTF} → 7.54 ms/pulse _{0.754 RTF}
🟢 -6.8%	hey_snips_v31 _{evaltime · 400ms}	`beaglev-ahead`	6.17 ms → 5.75 ms
🟢 -6.5%	mobilenet_v2_1 _{evaltime · pass_mt}	`cortex-a9`	965 ms → 902 ms
🟢 -6.1%	en_tdnn_8M _{evaltime · pulse_240ms}	`beaglev-ahead`	108 ms/pulse _{0.449 RTF} → 101 ms/pulse _{0.422 RTF}
🟢 -5.9%	hey_snips_v4_model17_nnef _{evaltime · pulse8}	`cortex-a53`	2.75 ms/pulse _{0.0344 RTF} → 2.59 ms/pulse _{0.0324 RTF}
🟢 -5.7%	mobilenet_v1_1 _{evaltime · pass}	`beaglev-ahead`	914 ms → 862 ms
🟢 -5.5%	hey_snips_v4_model17 _{evaltime · pulse8}	`cortex-a53`	2.75 ms/pulse _{0.0344 RTF} → 2.6 ms/pulse _{0.0325 RTF}
🟢 -5.4%	mobilenet_v2_1 _{evaltime · pass}	`beaglev-ahead`	1.14 s → 1.08 s
🟢 -5.2%	en_tdnn_8M _{evaltime · pulse_120ms}	`beaglev-ahead`	54.3 ms/pulse _{0.453 RTF} → 51.5 ms/pulse _{0.429 RTF}
🟢 -5.2%	hey_snips_v4_model17 _{evaltime · pulse8}	`apple-m1-max`	0.115 ms/pulse _{0.00143 RTF} → 0.109 ms/pulse _{0.00136 RTF}
🟢 -5.2%	hey_snips_v4_model17_nnef _{evaltime · pulse8}	`apple-m1-max`	0.115 ms/pulse _{0.00143 RTF} → 0.109 ms/pulse _{0.00136 RTF}
🟢 -5.1%	en_tdnn_pyt_15M _{evaltime · pulse_120ms}	`beaglev-ahead`	76.7 ms/pulse _{0.639 RTF} → 72.8 ms/pulse _{0.606 RTF}
🟢 -4.8%	llama_3_1_8B_instruct_q40ef16_541 _{decode · cuda}	`i9-11900kb_rtx-4060`	20.7 ms/tok _{48.27 tok/s} → 19.7 ms/tok _{50.72 tok/s}
🟢 -4.8%	qwen3_1_7B_q40ef16_541 _{decode · metal}	`apple-m1-max`	9.5 ms/tok _{105.3 tok/s} → 9.04 ms/tok _{110.6 tok/s}
🟢 -4.2%	llama_3_1_8B_instruct_q40ef16_541 _{decode · cuda}	`jetson-orin-nx`	83.1 ms/tok _{12.04 tok/s} → 79.6 ms/tok _{12.57 tok/s}
🟢 -3.6%	llama_3_2_1B_instruct_q40ef16_541 _{decode · metal}	`apple-m1-max`	5.75 ms/tok _{173.9 tok/s} → 5.54 ms/tok _{180.5 tok/s}
🟢 -3.2%	qwen3_1_7B_q40ef16_541 _{decode · cuda}	`i9-11900kb_rtx-4060`	6.2 ms/tok _{161.3 tok/s} → 6 ms/tok _{166.6 tok/s}
🟢 -3.1%	openelm_270M_q40ef16_541 _{decode · cuda}	`jetson-orin-nx`	11 ms/tok _{90.81 tok/s} → 10.7 ms/tok _{93.75 tok/s}
🟢 -3.0%	qwen3_1_7B_q40ef16_541 _{prefill · cuda}	`i9-11900kb_rtx-4060`	0.11 ms/tok _{9105 tok/s} → 0.106 ms/tok _{9391 tok/s}
🟢 -3.0%	openelm_270M_q40ef16_541 _{prefill · cuda}	`i9-11900kb_rtx-4060`	0.0328 ms/tok _{3.048e4 tok/s} → 0.0318 ms/tok _{3.142e4 tok/s}

⚠️ 390 secondary regression(s)

Δ	metric	device	main → PR
⚠️ +100.2%	hey_snips_v4_model17 _{RSS @ load · 2sec}	`beaglev-ahead`	16.4 MB → 32.9 MB
⚠️ +91.0%	hey_snips_v4_model17 _{RSS @ ready · 2sec}	`orangepi-rv2`	17.7 MB → 33.8 MB
⚠️ +87.3%	hey_snips_v4_model17 _{RSS @ ready · 2sec}	`beaglev-ahead`	17.8 MB → 33.4 MB
⚠️ +84.5%	hey_snips_v4_model17 _{RSS @ load · 2sec}	`orangepi-rv2`	16.4 MB → 30.2 MB
⚠️ +79.1%	voicecom_fake_quant _{RSS @ ready · 2sec}	`beaglev-ahead`	17.9 MB → 32 MB
⚠️ +79.0%	mobilenet_v2_1 _{RSS @ load · pass_mt}	`orangepi-rv2`	38 MB → 68 MB
⚠️ +77.6%	voicecom_fake_quant _{RSS @ ready · 2sec}	`orangepi-rv2`	18.2 MB → 32.2 MB
⚠️ +73.1%	mobilenet_v2_1 _{RSS @ load · pass_mt}	`cortex-a9`	42.6 MB → 73.7 MB
⚠️ +72.8%	mobilenet_v2_1 _{RSS @ ready · pass_mt}	`orangepi-rv2`	40.1 MB → 69.3 MB
⚠️ +72.1%	voicecom_float _{RSS @ ready · 2sec}	`beaglev-ahead`	17 MB → 29.3 MB
⚠️ +69.8%	hey_snips_v4_model17 _{RSS @ load · pulse8}	`orangepi-rv2`	17.6 MB → 29.9 MB
⚠️ +68.2%	voicecom_float _{RSS @ ready · 2sec}	`orangepi-rv2`	17.5 MB → 29.4 MB
⚠️ +67.3%	hey_snips_v4_model17 _{RSS @ load · pulse8}	`beaglev-ahead`	17.6 MB → 29.5 MB
⚠️ +66.1%	hey_snips_v4_model17 _{RSS @ ready · pulse8}	`orangepi-rv2`	18.6 MB → 30.9 MB
⚠️ +65.3%	inceptionv1q _{RSS @ ready · pass}	`beaglev-ahead`	25.7 MB → 42.4 MB
⚠️ +62.5%	hey_snips_v4_model17 _{RSS @ ready · pulse8}	`beaglev-ahead`	18.7 MB → 30.4 MB
⚠️ +58.2%	en_tdnn_8M _{RSS @ ready · 2600ms}	`orangepi-rv2`	48.3 MB → 76.4 MB
⚠️ +57.1%	en_tdnn_8M _{RSS @ ready · pulse_120ms}	`orangepi-rv2`	49.4 MB → 77.6 MB
⚠️ +57.1%	mobilenet_v1_1 _{RSS @ ready · pass}	`cortex-a9`	36.1 MB → 56.8 MB
⚠️ +56.9%	en_tdnn_8M _{RSS @ ready · pulse_180ms}	`orangepi-rv2`	49.4 MB → 77.5 MB
⚠️ +56.8%	en_tdnn_8M _{RSS @ ready · pulse_240ms}	`orangepi-rv2`	49.4 MB → 77.5 MB
⚠️ +56.1%	en_tdnn_8M _{RSS @ ready · pulse_180ms}	`beaglev-ahead`	49.7 MB → 77.5 MB
⚠️ +55.4%	voicecom_fake_quant _{RSS @ ready · 2sec}	`cortex-a7`	21.9 MB → 34 MB
⚠️ +54.6%	mobilenet_v2_1 _{RSS @ ready · pass}	`orangepi-rv2`	40.1 MB → 62 MB
⚠️ +53.1%	en_tdnn_8M _{RSS @ ready · pulse_120ms}	`cortex-a7`	53.2 MB → 81.4 MB
⚠️ +52.8%	en_tdnn_8M _{RSS @ ready · pulse_120ms}	`cortex-a9`	53.3 MB → 81.5 MB
⚠️ +52.6%	inceptionv1q _{RSS @ ready · pass}	`orangepi-rv2`	25.8 MB → 39.3 MB
⚠️ +52.5%	inceptionv1q _{RSS @ load · pass}	`orangepi-rv2`	19.9 MB → 30.3 MB
⚠️ +51.6%	en_tdnn_8M _{RSS @ ready · 2600ms}	`cortex-a9`	52.6 MB → 79.7 MB
⚠️ +51.4%	mobilenet_v1_1 _{RSS @ load · pass}	`orangepi-rv2`	30.4 MB → 45.9 MB
⚠️ +51.2%	en_tdnn_8M _{RSS @ ready · pulse_240ms}	`cortex-a7`	53.3 MB → 80.6 MB
⚠️ +51.0%	en_tdnn_8M _{RSS @ ready · pulse_240ms}	`beaglev-ahead`	49.6 MB → 74.9 MB
⚠️ +50.6%	en_tdnn_8M _{RSS @ ready · 2600ms}	`beaglev-ahead`	48.5 MB → 73.1 MB
⚠️ +50.4%	en_tdnn_8M_nnef _{RSS @ ready · pulse_240ms}	`orangepi-rv2`	47.3 MB → 71.2 MB
⚠️ +48.3%	mobilenet_v2_1 _{RSS @ ready · pass_mt}	`cortex-a53`	52.8 MB → 78.2 MB
⚠️ +47.5%	voicecom_float _{RSS @ ready · 2sec}	`cortex-a7`	20.6 MB → 30.4 MB
⚠️ +47.4%	en_tdnn_8M _{RSS @ ready · pulse_120ms}	`beaglev-ahead`	49.6 MB → 73.2 MB
⚠️ +47.3%	mobilenet_v1_1 _{RSS @ load · pass}	`cortex-a9`	34.8 MB → 51.2 MB
⚠️ +47.2%	hey_snips_v4_model17_nnef _{RSS @ ready · pulse8}	`orangepi-rv2`	16.1 MB → 23.7 MB
⚠️ +46.3%	en_tdnn_8M _{RSS @ ready · 2600ms}	`cortex-a7`	52.5 MB → 76.8 MB
⚠️ +46.2%	en_tdnn_8M _{RSS @ ready · pulse_240ms}	`cortex-a9`	53.4 MB → 78.1 MB
⚠️ +46.1%	hey_snips_v4_model17 _{RSS @ load · 2sec}	`i9-11900kb_rtx-4060`	32.6 MB → 47.6 MB
⚠️ +46.1%	inceptionv1q _{RSS @ load · pass}	`beaglev-ahead`	19.9 MB → 29 MB
⚠️ +46.0%	mobilenet_v1_1 _{RSS @ ready · pass}	`beaglev-ahead`	32 MB → 46.7 MB
⚠️ +45.8%	voicecom_float _{RSS @ ready · 2sec}	`cortex-a9`	20.9 MB → 30.5 MB
⚠️ +45.2%	en_tdnn_8M _{RSS @ ready · pulse_180ms}	`cortex-a7`	53.4 MB → 77.6 MB
⚠️ +43.8%	hey_snips_v4_model17_nnef _{RSS @ ready · pulse8}	`beaglev-ahead`	16.4 MB → 23.5 MB
⚠️ +43.5%	mobilenet_v2_1 _{RSS @ ready · pass}	`cortex-a7`	45.1 MB → 64.7 MB
⚠️ +43.1%	voicecom_float _{RSS @ load · 2sec}	`beaglev-ahead`	15.9 MB → 22.7 MB
⚠️ +43.1%	voicecom_fake_quant _{RSS @ load · 2sec}	`cortex-a7`	19 MB → 27.2 MB
⚠️ +43.1%	voicecom_float _{RSS @ load · 2sec}	`orangepi-rv2`	15.9 MB → 22.8 MB
⚠️ +42.9%	mobilenet_v2_1 _{RSS @ ready · pass_mt}	`cortex-a9`	45.4 MB → 64.8 MB
⚠️ +41.7%	llama_3_2_3B_instruct_q40ef16_541 _{RSS @ ready · cuda}	`jetson-orin-nx`	430 MB → 610 MB
⚠️ +41.5%	hey_snips_v1 _{RSS @ ready · 400ms}	`orangepi-rv2`	14.3 MB → 20.2 MB
⚠️ +41.2%	mobilenet_v2_1 _{RSS @ load · pass_mt}	`cortex-a7`	42.3 MB → 59.8 MB
⚠️ +41.1%	en_tdnn_pyt_15M _{RSS @ load · pulse_120ms}	`orangepi-rv2`	60.4 MB → 85.2 MB
⚠️ +41.0%	en_tdnn_pyt_15M _{RSS @ load · pulse_120ms}	`beaglev-ahead`	60.4 MB → 85.1 MB
⚠️ +40.8%	hey_snips_v1 _{RSS @ ready · 400ms}	`beaglev-ahead`	14.2 MB → 20 MB
⚠️ +40.6%	voicecom_fake_quant _{RSS @ load · 2sec}	`orangepi-rv2`	16.9 MB → 23.7 MB
⚠️ +40.6%	mobilenet_v2_1 _{RSS @ load · pass}	`beaglev-ahead`	38 MB → 53.5 MB
⚠️ +40.4%	mobilenet_v2_1 _{RSS @ load · pass}	`cortex-a9`	42.5 MB → 59.6 MB
⚠️ +40.3%	mobilenet_v2_1 _{RSS @ load · pass}	`orangepi-rv2`	38 MB → 53.3 MB
⚠️ +40.2%	voicecom_fake_quant _{RSS @ load · 2sec}	`cortex-a9`	19.2 MB → 26.9 MB
⚠️ +40.1%	mobilenet_v2_1 _{RSS @ load · pass}	`cortex-a7`	42.3 MB → 59.3 MB
⚠️ +40.1%	mobilenet_v2_1 _{RSS @ ready · pass}	`beaglev-ahead`	40.1 MB → 56.2 MB
⚠️ +39.8%	voicecom_fake_quant _{RSS @ load · 2sec}	`beaglev-ahead`	16.9 MB → 23.6 MB
⚠️ +39.4%	hey_snips_v4_model17 _{RSS @ load · 2sec}	`cortex-a55`	28.2 MB → 39.3 MB
⚠️ +39.2%	inceptionv3 _{RSS @ ready · pass_mt}	`beaglev-ahead`	118 MB → 164 MB
⚠️ +39.0%	parakeet_tdt_600m_v3_f32f32_preprocessor_1s _{RSS @ load · cuda}	`jetson-orin-nx`	21.3 MB → 29.6 MB
⚠️ +38.8%	en_tdnn_8M _{RSS @ ready · pulse_180ms}	`cortex-a9`	53.4 MB → 74.1 MB
⚠️ +37.5%	en_tdnn_pyt_15M _{RSS @ load · pulse_120ms}	`cortex-a9`	63.7 MB → 87.7 MB
⚠️ +36.7%	en_tdnn_pyt_15M _{RSS @ load · pulse_120ms}	`cortex-a7`	63.8 MB → 87.2 MB
⚠️ +36.4%	en_tdnn_8M_nnef _{RSS @ ready · pulse_240ms}	`cortex-a55`	55.7 MB → 75.9 MB
⚠️ +36.3%	en_tdnn_8M_nnef _{RSS @ ready · pulse_240ms}	`beaglev-ahead`	47.4 MB → 64.7 MB
⚠️ +36.2%	inceptionv3 _{RSS @ ready · pass}	`orangepi-rv2`	118 MB → 160 MB
⚠️ +36.2%	hey_snips_v4_model17 _{RSS @ load · 2sec}	`cortex-a53`	28.6 MB → 38.9 MB
⚠️ +36.0%	en_tdnn_8M_nnef _{RSS @ ready · pulse_240ms}	`cortex-a53`	55.8 MB → 75.9 MB
⚠️ +35.9%	trunet _{RSS @ ready · pulse1_f16}	`orangepi-rv2`	18.6 MB → 25.3 MB
⚠️ +35.8%	mobilenet_v1_1 _{RSS @ load · pass}	`beaglev-ahead`	30.4 MB → 41.3 MB
⚠️ +35.7%	hey_snips_v4_model17_nnef _{RSS @ load · pulse8}	`orangepi-rv2`	14.4 MB → 19.5 MB
⚠️ +35.3%	mobilenet_v1_1 _{RSS @ ready · pass}	`orangepi-rv2`	32 MB → 43.3 MB
⚠️ +34.9%	trunet _{RSS @ ready · pulse1_f16}	`beaglev-ahead`	18.7 MB → 25.2 MB
⚠️ +34.5%	inceptionv3 _{RSS @ ready · pass}	`beaglev-ahead`	118 MB → 159 MB
⚠️ +34.2%	en_tdnn_pyt_15M _{RSS @ ready · pulse_120ms}	`beaglev-ahead`	61.4 MB → 82.4 MB
⚠️ +33.8%	trunet _{RSS @ load · pulse1_f16}	`orangepi-rv2`	18.2 MB → 24.4 MB
⚠️ +33.4%	trunet _{RSS @ load · pulse1_f16}	`beaglev-ahead`	18.2 MB → 24.3 MB
⚠️ +33.3%	mobilenet_v2_1 _{RSS @ ready · pass}	`cortex-a53`	53.2 MB → 70.9 MB
⚠️ +33.1%	trunet _{RSS @ ready · pulse1_f32}	`orangepi-rv2`	20.3 MB → 27 MB
⚠️ +33.0%	en_tdnn_8M_nnef _{RSS @ ready · pulse_240ms}	`cortex-a9`	50.6 MB → 67.3 MB
⚠️ +32.6%	inceptionv1q _{RSS @ load · pass}	`cortex-a9`	23.5 MB → 31.2 MB
⚠️ +32.6%	voicecom_float _{RSS @ load · 2sec}	`cortex-a7`	19.2 MB → 25.5 MB
⚠️ +32.6%	mdl_en_2019_Q3_librispeech_onnx _{RSS @ load · pulse_240ms}	`cortex-a9`	29.1 MB → 38.6 MB
⚠️ +32.4%	voicecom_float _{RSS @ load · 2sec}	`cortex-a9`	19.4 MB → 25.7 MB
⚠️ +32.2%	voicecom_fake_quant _{RSS @ ready · 2sec}	`cortex-a9`	22 MB → 29.1 MB
⚠️ +32.2%	hey_snips_v4_model17_nnef _{RSS @ load · pulse8}	`beaglev-ahead`	14.4 MB → 19.1 MB
⚠️ +32.1%	en_tdnn_lstm_bn_q7 _{RSS @ load · pulse_240ms}	`cortex-a7`	21.8 MB → 28.8 MB
⚠️ +31.7%	en_tdnn_8M _{RSS @ load · 2600ms}	`beaglev-ahead`	47.4 MB → 62.4 MB
⚠️ +31.6%	inceptionv3 _{RSS @ ready · pass_mt}	`cortex-a7`	126 MB → 166 MB
⚠️ +31.2%	voicecom_fake_quant _{RSS @ load · 2sec}	`cortex-a55`	22.6 MB → 29.6 MB
⚠️ +31.2%	en_tdnn_pyt_15M _{RSS @ ready · pulse_120ms}	`orangepi-rv2`	61.2 MB → 80.3 MB
⚠️ +30.9%	en_tdnn_8M _{RSS @ ready · 2600ms}	`cortex-a55`	62.7 MB → 82.1 MB
⚠️ +30.7%	llama_3_2_3B_instruct_q40ef16_541 _{RSS @ ready · cuda}	`i9-11900kb_rtx-4060`	618 MB → 808 MB
⚠️ +30.6%	inceptionv1q _{RSS @ load · pass}	`cortex-a7`	23.5 MB → 30.7 MB
⚠️ +30.6%	hey_snips_v1 _{RSS @ ready · 400ms}	`cortex-a7`	18.1 MB → 23.6 MB
⚠️ +30.6%	voicecom_fake_quant _{RSS @ load · 2sec}	`i9-11900kb_rtx-4060`	26.2 MB → 34.2 MB
⚠️ +30.6%	openelm_270M_q40ef16_541 _{RSS @ ready · cpu}	`i9-11900kb_rtx-4060`	193 MB → 252 MB
⚠️ +30.3%	mobilenet_v1_1 _{RSS @ load · pass}	`cortex-a7`	34.8 MB → 45.4 MB
⚠️ +30.0%	dummy_conmer_12M _{RSS @ ready · pulse_120ms}	`orangepi-rv2`	68.7 MB → 89.3 MB
⚠️ +29.7%	mobilenet_v2_1 _{RSS @ load · pass_mt}	`beaglev-ahead`	38 MB → 49.3 MB
⚠️ +29.4%	mdl_en_2019_Q3_librispeech_onnx _{RSS @ load · 2600ms}	`cortex-a9`	29 MB → 37.5 MB
⚠️ +29.4%	arm_ml_kws_cnn_m _{RSS @ ready · pass}	`beaglev-ahead`	15.1 MB → 19.6 MB
⚠️ +29.4%	mdl_en_2019_Q3_librispeech_onnx _{RSS @ load · 2600ms}	`cortex-a7`	29.1 MB → 37.7 MB
⚠️ +29.2%	en_tdnn_pyt_15M _{RSS @ ready · pulse_120ms}	`cortex-a7`	65.2 MB → 84.3 MB
⚠️ +29.2%	voicecom_fake_quant _{RSS @ load · 2sec}	`cortex-a53`	22.4 MB → 29 MB
⚠️ +29.1%	dummy_conmer_12M _{RSS @ ready · pulse_120ms}	`beaglev-ahead`	69.1 MB → 89.2 MB
⚠️ +28.8%	mobilenet_v1_1 _{RSS @ ready · pass}	`cortex-a7`	36.2 MB → 46.7 MB
⚠️ +28.7%	en_tdnn_8M_nnef _{RSS @ ready · pulse_240ms}	`cortex-a7`	50.8 MB → 65.4 MB
⚠️ +28.1%	nemotron_3_5_asr_streaming_0_6b_f32f32_preprocessor_1s _{RSS @ load · cuda}	`jetson-orin-nx`	21 MB → 27 MB
⚠️ +28.0%	inceptionv3 _{RSS @ ready · pass_mt}	`orangepi-rv2`	118 MB → 151 MB
⚠️ +27.7%	en_tdnn_pyt_15M _{RSS @ ready · pulse_120ms}	`cortex-a9`	65.2 MB → 83.3 MB
⚠️ +27.5%	mdl_en_2019_Q3_librispeech_onnx _{RSS @ load · pulse_240ms}	`cortex-a7`	29.1 MB → 37.1 MB
⚠️ +27.4%	en_tdnn_8M _{RSS @ ready · pulse_180ms}	`cortex-a55`	64.2 MB → 81.8 MB
⚠️ +27.4%	hey_snips_v31 _{RSS @ load · 400ms}	`beaglev-ahead`	14.2 MB → 18 MB
⚠️ +27.3%	llama_3_2_1B_instruct_q40ef16_541 _{RSS @ ready · cuda}	`jetson-orin-nx`	356 MB → 454 MB
⚠️ +27.0%	qwen3_1_7B_q40ef16_541 _{RSS @ ready · cuda}	`jetson-orin-nx`	389 MB → 494 MB
⚠️ +26.8%	inceptionv3 _{RSS @ load · pass_mt}	`beaglev-ahead`	113 MB → 143 MB
⚠️ +26.4%	en_tdnn_8M _{RSS @ load · pulse_240ms}	`cortex-a55`	54.2 MB → 68.5 MB
⚠️ +26.3%	hey_snips_v31 _{RSS @ ready · 400ms}	`beaglev-ahead`	15.2 MB → 19.2 MB
⚠️ +26.1%	en_tdnn_lstm_bn_q7 _{RSS @ load · pulse_240ms}	`cortex-a9`	22.4 MB → 28.3 MB
⚠️ +26.0%	hey_snips_v1 _{RSS @ load · 400ms}	`orangepi-rv2`	13 MB → 16.4 MB
⚠️ +25.8%	hey_snips_v1 _{RSS @ ready · 400ms}	`cortex-a9`	18.2 MB → 22.9 MB
⚠️ +25.7%	mdl_en_2019_Q3_librispeech_onnx _{RSS @ ready · 2600ms}	`cortex-a7`	31 MB → 38.9 MB
⚠️ +25.5%	voicecom_float _{RSS @ load · 2sec}	`i9-11900kb_rtx-4060`	26.2 MB → 32.9 MB
⚠️ +25.5%	trunet _{RSS @ ready · pulse1_f32}	`beaglev-ahead`	20.4 MB → 25.6 MB
⚠️ +25.4%	hey_snips_v31 _{RSS @ load · 400ms}	`orangepi-rv2`	14.2 MB → 17.8 MB
⚠️ +25.4%	llama_3_1_8B_instruct_q40ef16_541 _{RSS @ ready · cuda}	`jetson-orin-nx`	505 MB → 633 MB
⚠️ +25.3%	en_tdnn_8M _{RSS @ load · pulse_120ms}	`cortex-a9`	51.3 MB → 64.3 MB
⚠️ +25.3%	mdl_en_2019_Q3_librispeech_onnx _{RSS @ ready · pulse_240ms}	`cortex-a9`	31.7 MB → 39.7 MB
⚠️ +25.3%	en_tdnn_lstm_bn_q7 _{RSS @ load · 2600ms}	`cortex-a9`	22.3 MB → 27.9 MB
⚠️ +25.2%	trunet _{RSS @ load · pulse1_f32}	`orangepi-rv2`	19.4 MB → 24.2 MB
⚠️ +25.0%	hey_snips_v1 _{RSS @ load · 400ms}	`beaglev-ahead`	13.1 MB → 16.4 MB
⚠️ +25.0%	en_tdnn_8M _{RSS @ load · pulse_120ms}	`cortex-a7`	51.3 MB → 64.1 MB
⚠️ +24.9%	mobilenet_v2_1 _{RSS @ ready · pass_mt}	`beaglev-ahead`	40.1 MB → 50 MB
⚠️ +24.6%	en_tdnn_8M _{RSS @ ready · pulse_240ms}	`cortex-a53`	63.8 MB → 79.4 MB
⚠️ +24.6%	en_tdnn_8M _{RSS @ load · 2600ms}	`orangepi-rv2`	47.3 MB → 59 MB
⚠️ +24.5%	en_tdnn_lstm_bn_q7 _{RSS @ load · 2600ms}	`cortex-a7`	22.1 MB → 27.6 MB
⚠️ +24.1%	trunet _{RSS @ load · pulse1_f32}	`beaglev-ahead`	19.4 MB → 24 MB
⚠️ +24.1%	arm_ml_kws_cnn_m _{RSS @ ready · pass}	`orangepi-rv2`	15.8 MB → 19.7 MB
⚠️ +24.1%	inceptionv1q _{RSS @ ready · pass}	`i9-11900kb_rtx-4060`	43.9 MB → 54.5 MB
⚠️ +24.0%	mdl_en_2019_Q3_librispeech_onnx _{RSS @ ready · 2600ms}	`cortex-a9`	31.2 MB → 38.7 MB
⚠️ +23.9%	en_tdnn_8M _{RSS @ load · pulse_120ms}	`orangepi-rv2`	48.6 MB → 60.3 MB
⚠️ +23.9%	en_tdnn_8M _{RSS @ load · pulse_120ms}	`cortex-a55`	54.6 MB → 67.6 MB
⚠️ +23.8%	en_tdnn_8M _{RSS @ load · pulse_180ms}	`orangepi-rv2`	48.6 MB → 60.2 MB
⚠️ +23.8%	en_tdnn_8M _{RSS @ load · pulse_180ms}	`beaglev-ahead`	48.7 MB → 60.2 MB
⚠️ +23.6%	en_tdnn_lstm_bn_q7 _{RSS @ ready · pulse_240ms}	`cortex-a7`	24.6 MB → 30.5 MB
⚠️ +23.6%	en_tdnn_8M _{RSS @ load · pulse_240ms}	`orangepi-rv2`	48.7 MB → 60.1 MB
⚠️ +23.6%	en_tdnn_8M _{RSS @ load · pulse_240ms}	`cortex-a7`	51.3 MB → 63.4 MB
⚠️ +23.5%	parakeet_tdt_600m_v3_f32f32_decoder_pass _{RSS @ load · cuda}	`jetson-orin-nx`	66.6 MB → 82.2 MB
⚠️ +23.4%	hey_snips_v31 _{RSS @ ready · 400ms}	`orangepi-rv2`	15.3 MB → 18.9 MB
⚠️ +23.4%	hey_snips_v4_model17_nnef _{RSS @ ready · pulse8}	`i9-11900kb_rtx-4060`	27.8 MB → 34.3 MB
⚠️ +23.4%	en_tdnn_8M _{RSS @ load · pulse_180ms}	`cortex-a7`	51.4 MB → 63.5 MB
⚠️ +23.4%	hey_snips_v4_model17 _{RSS @ ready · 2sec}	`cortex-a55`	32.8 MB → 40.5 MB
⚠️ +23.3%	en_tdnn_8M_nnef _{RSS @ ready · pulse_240ms}	`i9-11900kb_rtx-4060`	60 MB → 74.1 MB
⚠️ +23.1%	en_tdnn_8M _{RSS @ load · pulse_180ms}	`cortex-a9`	51.4 MB → 63.3 MB
⚠️ +23.1%	inceptionv3 _{RSS @ load · pass_mt}	`cortex-a7`	116 MB → 142 MB
⚠️ +23.1%	en_tdnn_lstm_bn_q7 _{RSS @ load · pulse_240ms}	`cortex-a55`	24.4 MB → 30 MB
⚠️ +23.0%	inceptionv1q _{RSS @ load · pass}	`cortex-a55`	27.9 MB → 34.4 MB
⚠️ +23.0%	llama_3_1_8B_instruct_q40ef16_541 _{RSS @ ready · cuda}	`i9-11900kb_rtx-4060`	691 MB → 850 MB
⚠️ +22.9%	en_tdnn_pyt_15M _{RSS @ load · pulse_120ms}	`cortex-a55`	74.6 MB → 91.7 MB
⚠️ +22.6%	en_tdnn_8M _{RSS @ ready · pulse_240ms}	`cortex-a55`	64.3 MB → 78.8 MB
⚠️ +22.6%	inceptionv1q _{RSS @ load · pass}	`i9-11900kb_rtx-4060`	31.7 MB → 38.8 MB
⚠️ +22.4%	hey_snips_v4_model17_nnef _{RSS @ load · pulse8}	`i9-11900kb_rtx-4060`	24.7 MB → 30.2 MB
⚠️ +22.3%	dummy_conmer_12M _{RSS @ load · pulse_120ms}	`beaglev-ahead`	67.9 MB → 83.1 MB
⚠️ +22.3%	speaker_id _{RSS @ load · pulse8}	`cortex-a9`	18.4 MB → 22.5 MB
⚠️ +22.2%	en_tdnn_lstm_bn_q7 _{RSS @ load · 2600ms}	`i9-11900kb_rtx-4060`	28.5 MB → 34.8 MB
⚠️ +22.0%	parakeet_tdt_600m_v3_f32f32_decoder_pass _{RSS @ load · cpu}	`i9-11900kb_rtx-4060`	69.6 MB → 84.9 MB
⚠️ +22.0%	en_tdnn_8M _{RSS @ load · 2600ms}	`cortex-a7`	51.4 MB → 62.7 MB
⚠️ +22.0%	en_tdnn_lstm_bn_q7 _{RSS @ ready · 2600ms}	`cortex-a9`	24.2 MB → 29.5 MB
⚠️ +21.8%	inceptionv3 _{RSS @ load · pass_mt}	`orangepi-rv2`	113 MB → 138 MB
⚠️ +21.7%	parakeet_tdt_600m_v3_f32f32_decoder_pass _{RSS @ load · cuda}	`i9-11900kb_rtx-4060`	69.5 MB → 84.6 MB
⚠️ +21.7%	hey_snips_v4_model17 _{RSS @ load · 2sec}	`cortex-a9`	23.6 MB → 28.8 MB
⚠️ +21.7%	en_tdnn_pyt_15M _{RSS @ load · pulse_120ms}	`cortex-a53`	74.6 MB → 90.7 MB
⚠️ +21.5%	inceptionv3 _{RSS @ load · pass}	`beaglev-ahead`	113 MB → 137 MB
⚠️ +21.3%	inceptionv3 _{RSS @ ready · pass}	`cortex-a7`	126 MB → 153 MB
⚠️ +21.3%	en_tdnn_15M _{RSS @ ready · pulse_240ms}	`cortex-a9`	91.2 MB → 111 MB
⚠️ +21.2%	en_tdnn_lstm_bn_q7 _{RSS @ load · pulse_240ms}	`cortex-a53`	24.5 MB → 29.7 MB
⚠️ +21.2%	en_tdnn_8M _{RSS @ load · pulse_240ms}	`cortex-a53`	54.1 MB → 65.6 MB
⚠️ +21.2%	en_tdnn_lstm_bn_q7 _{RSS @ ready · 2600ms}	`cortex-a7`	24.1 MB → 29.2 MB
⚠️ +21.2%	inceptionv3 _{RSS @ load · pass}	`cortex-a7`	116 MB → 141 MB
⚠️ +20.9%	en_tdnn_8M _{RSS @ load · pulse_180ms}	`cortex-a55`	54.7 MB → 66.1 MB
⚠️ +20.8%	inceptionv1q _{RSS @ load · pass}	`cortex-a53`	27.8 MB → 33.5 MB
⚠️ +20.8%	en_tdnn_lstm_bn_q7 _{RSS @ load · pulse_240ms}	`i9-11900kb_rtx-4060`	29.7 MB → 35.9 MB
⚠️ +20.8%	hey_snips_v4_model17 _{RSS @ ready · 2sec}	`cortex-a53`	33.2 MB → 40.1 MB
⚠️ +20.7%	mdl_en_2019_Q3_librispeech_onnx _{RSS @ load · pulse_240ms}	`cortex-a55`	32.7 MB → 39.5 MB
⚠️ +20.6%	hey_snips_v1 _{RSS @ load · 400ms}	`cortex-a7`	16.4 MB → 19.8 MB
⚠️ +20.4%	en_tdnn_lstm_bn_q7 _{RSS @ ready · pulse_240ms}	`cortex-a9`	24.9 MB → 30 MB
⚠️ +20.4%	hey_snips_v31 _{RSS @ load · 400ms}	`i9-11900kb_rtx-4060`	23.2 MB → 27.9 MB
⚠️ +20.2%	inceptionv3 _{RSS @ ready · pass}	`cortex-a9`	126 MB → 152 MB
⚠️ +20.1%	mdl_en_2019_Q3_librispeech_onnx _{RSS @ ready · pulse_240ms}	`cortex-a7`	31.9 MB → 38.3 MB
⚠️ +20.1%	mobilenet_v2_1 _{RSS @ ready · pass_mt}	`cortex-a7`	45.2 MB → 54.2 MB
⚠️ +20.0%	hey_snips_v1 _{RSS @ load · 400ms}	`i9-11900kb_rtx-4060`	22.3 MB → 26.7 MB
⚠️ +19.9%	en_tdnn_8M _{RSS @ load · pulse_120ms}	`cortex-a53`	54.8 MB → 65.6 MB
⚠️ +19.9%	arm_ml_kws_cnn_m _{RSS @ load · pass}	`beaglev-ahead`	14.1 MB → 16.9 MB
⚠️ +19.8%	en_tdnn_8M _{RSS @ load · pulse_180ms}	`cortex-a53`	54.6 MB → 65.4 MB
⚠️ +19.8%	dummy_conmer_12M _{RSS @ load · pulse_120ms}	`orangepi-rv2`	67.7 MB → 81.1 MB
⚠️ +19.7%	mdl_en_2019_Q3_librispeech_onnx _{RSS @ load · pulse_240ms}	`cortex-a53`	32.9 MB → 39.4 MB
⚠️ +19.7%	en_tdnn_8M _{RSS @ load · 2600ms}	`cortex-a55`	54.1 MB → 64.7 MB
⚠️ +19.5%	en_tdnn_lstm_bn_q7 _{RSS @ load · 2600ms}	`cortex-a55`	24.1 MB → 28.8 MB
⚠️ +19.4%	inceptionv3 _{RSS @ ready · pass_mt}	`cortex-a9`	126 MB → 150 MB
⚠️ +19.3%	mobilenet_v2_1 _{RSS @ ready · pass}	`cortex-a9`	45.2 MB → 54 MB
⚠️ +19.1%	trunet _{RSS @ load · pulse1_f16}	`cortex-a7`	21.4 MB → 25.5 MB
⚠️ +19.0%	hey_snips_v4_model17 _{RSS @ load · 2sec}	`cortex-a7`	23.9 MB → 28.4 MB
⚠️ +19.0%	nemotron_3_5_asr_streaming_0_6b_f32f32_decoder_pass _{RSS @ load · cuda}	`jetson-orin-nx`	79.1 MB → 94.2 MB
⚠️ +18.9%	en_tdnn_8M _{RSS @ load · pulse_120ms}	`beaglev-ahead`	48.6 MB → 57.8 MB
⚠️ +18.7%	trunet _{RSS @ load · pulse1_f32}	`cortex-a55`	24.6 MB → 29.2 MB
⚠️ +18.6%	hey_snips_v4_model17 _{RSS @ load · pulse8}	`cortex-a9`	24 MB → 28.4 MB
⚠️ +18.5%	speaker_id _{RSS @ load · pulse8}	`cortex-a53`	19.4 MB → 23 MB
⚠️ +18.5%	mdl_en_2019_Q3_librispeech_onnx _{RSS @ load · 2600ms}	`cortex-a53`	32.6 MB → 38.6 MB
⚠️ +18.5%	en_tdnn_15M _{RSS @ ready · pulse_240ms}	`beaglev-ahead`	88 MB → 104 MB
⚠️ +18.5%	en_tdnn_lstm_bn_q7 _{RSS @ load · 2600ms}	`cortex-a53`	24.3 MB → 28.8 MB
⚠️ +18.4%	en_tdnn_15M _{RSS @ load · pulse_240ms}	`beaglev-ahead`	86.8 MB → 103 MB
⚠️ +18.4%	en_tdnn_15M _{RSS @ load · pulse_120ms}	`beaglev-ahead`	86.9 MB → 103 MB
⚠️ +18.4%	en_tdnn_15M _{RSS @ load · 2600ms}	`beaglev-ahead`	85.6 MB → 101 MB
⚠️ +18.4%	hey_snips_v1 _{RSS @ ready · 400ms}	`i9-11900kb_rtx-4060`	25.6 MB → 30.3 MB
⚠️ +18.3%	en_tdnn_8M _{RSS @ load · pulse_240ms}	`beaglev-ahead`	48.6 MB → 57.5 MB
⚠️ +18.3%	mdl_en_2019_Q3_librispeech_onnx _{RSS @ load · 2600ms}	`cortex-a55`	32.6 MB → 38.5 MB
⚠️ +18.2%	inceptionv3 _{RSS @ load · pass}	`orangepi-rv2`	113 MB → 133 MB
⚠️ +18.2%	en_tdnn_15M _{RSS @ ready · 2600ms}	`beaglev-ahead`	86.8 MB → 103 MB
⚠️ +18.1%	speaker_id _{RSS @ load · pulse8}	`cortex-a7`	18.6 MB → 22 MB
⚠️ +18.1%	hey_snips_v31 _{RSS @ ready · 400ms}	`i9-11900kb_rtx-4060`	25.2 MB → 29.8 MB
⚠️ +18.0%	voicecom_float _{RSS @ load · 2sec}	`cortex-a53`	23.5 MB → 27.8 MB
⚠️ +18.0%	arm_ml_kws_cnn_m _{RSS @ ready · pass}	`i9-11900kb_rtx-4060`	25.9 MB → 30.6 MB
⚠️ +17.7%	en_tdnn_8M _{RSS @ ready · pulse_120ms}	`cortex-a55`	63.7 MB → 74.9 MB
⚠️ +17.7%	trunet _{RSS @ load · pulse1_f16}	`cortex-a9`	21.4 MB → 25.2 MB
⚠️ +17.6%	qwen3_1_7B_q40ef16_541 _{RSS @ ready · cuda}	`i9-11900kb_rtx-4060`	576 MB → 677 MB
⚠️ +17.5%	en_tdnn_8M _{RSS @ ready · pulse_120ms}	`cortex-a53`	65.1 MB → 76.5 MB
⚠️ +17.5%	en_tdnn_15M _{RSS @ ready · pulse_120ms}	`cortex-a9`	91.3 MB → 107 MB
⚠️ +17.4%	en_tdnn_15M _{RSS @ load · pulse_120ms}	`cortex-a9`	90.2 MB → 106 MB
⚠️ +17.4%	llama_3_2_1B_instruct_q40ef16_541 _{RSS @ ready · cuda}	`i9-11900kb_rtx-4060`	539 MB → 633 MB
⚠️ +17.4%	hey_snips_v4_model17 _{RSS @ ready · 2sec}	`cortex-a9`	25.4 MB → 29.8 MB
⚠️ +17.4%	en_tdnn_lstm_bn_q7 _{RSS @ ready · pulse_240ms}	`i9-11900kb_rtx-4060`	32.6 MB → 38.2 MB
⚠️ +17.3%	en_tdnn_15M _{RSS @ load · pulse_120ms}	`orangepi-rv2`	86.9 MB → 102 MB
⚠️ +17.3%	en_tdnn_15M _{RSS @ load · 2600ms}	`orangepi-rv2`	85.6 MB → 100 MB
⚠️ +17.3%	speaker_id _{RSS @ load · pulse8}	`cortex-a55`	19.4 MB → 22.8 MB
⚠️ +17.1%	en_tdnn_15M _{RSS @ load · pulse_240ms}	`orangepi-rv2`	86.9 MB → 102 MB
⚠️ +17.1%	speaker_id _{RSS @ ready · pulse8}	`orangepi-rv2`	16.9 MB → 19.8 MB
⚠️ +17.0%	nemotron_3_5_asr_streaming_0_6b_f32f32_decoder_pass _{RSS @ load · cpu}	`i9-11900kb_rtx-4060`	82.1 MB → 96 MB
⚠️ +17.0%	en_tdnn_15M _{RSS @ load · pulse_120ms}	`cortex-a7`	90.2 MB → 106 MB
⚠️ +17.0%	speaker_id _{RSS @ load · pulse8}	`i9-11900kb_rtx-4060`	25 MB → 29.3 MB
⚠️ +16.8%	mdl_en_2019_Q3_librispeech_onnx _{RSS @ load · 2600ms}	`i9-11900kb_rtx-4060`	36.5 MB → 42.6 MB
⚠️ +16.8%	nemotron_3_5_asr_streaming_0_6b_f32f32_decoder_pass _{RSS @ load · cuda}	`i9-11900kb_rtx-4060`	82.2 MB → 96 MB
⚠️ +16.7%	en_tdnn_15M _{RSS @ load · 2600ms}	`cortex-a7`	89.4 MB → 104 MB
⚠️ +16.7%	en_tdnn_15M _{RSS @ load · pulse_240ms}	`cortex-a9`	89.8 MB → 105 MB
⚠️ +16.7%	en_tdnn_8M _{RSS @ ready · pulse_120ms}	`i9-11900kb_rtx-4060`	74.9 MB → 87.4 MB
⚠️ +16.6%	parakeet_tdt_600m_v3_f32f32_preprocessor_1s _{RSS @ load · cpu}	`i9-11900kb_rtx-4060`	24.3 MB → 28.3 MB
⚠️ +16.6%	en_tdnn_15M _{RSS @ ready · pulse_120ms}	`cortex-a7`	91.3 MB → 106 MB
⚠️ +16.6%	speaker_id _{RSS @ ready · pulse8}	`beaglev-ahead`	16.9 MB → 19.7 MB
⚠️ +16.5%	arm_ml_kws_cnn_m _{RSS @ load · pass}	`orangepi-rv2`	14.1 MB → 16.4 MB
⚠️ +16.5%	inceptionv1q _{RSS @ ready · pass}	`cortex-a7`	29.4 MB → 34.3 MB
⚠️ +16.5%	arm_ml_kws_cnn_m _{RSS @ load · pass}	`i9-11900kb_rtx-4060`	22.9 MB → 26.7 MB
⚠️ +16.5%	en_tdnn_8M _{RSS @ load · pulse_240ms}	`cortex-a9`	51.4 MB → 59.9 MB
⚠️ +16.4%	mobilenet_v2_1 _{RSS @ ready · pass_mt}	`cortex-a55`	53.2 MB → 62 MB
⚠️ +16.4%	openelm_270M_q40ef16_541 _{RSS @ load · cuda}	`jetson-orin-nx`	199 MB → 232 MB
⚠️ +16.4%	en_tdnn_15M _{RSS @ ready · 2600ms}	`cortex-a7`	90.4 MB → 105 MB
⚠️ +16.3%	trunet _{RSS @ ready · pulse1_f16}	`cortex-a53`	26.5 MB → 30.9 MB
⚠️ +16.3%	voicecom_float _{RSS @ load · 2sec}	`cortex-a55`	24.1 MB → 28 MB
⚠️ +16.2%	parakeet_tdt_600m_v3_f32f32_preprocessor_1s _{RSS @ load · cuda}	`i9-11900kb_rtx-4060`	24.1 MB → 28 MB
⚠️ +16.1%	hey_snips_v4_model17_nnef _{RSS @ ready · pulse8}	`cortex-a9`	19.8 MB → 22.9 MB
⚠️ +16.1%	trunet _{RSS @ ready · pulse1_f16}	`cortex-a7`	22.8 MB → 26.5 MB
⚠️ +16.1%	inceptionv1q _{RSS @ ready · pass}	`cortex-a9`	29.8 MB → 34.6 MB
⚠️ +16.1%	en_tdnn_8M _{RSS @ load · 2600ms}	`cortex-a53`	54.4 MB → 63.1 MB
⚠️ +16.0%	hey_snips_v4_model17 _{RSS @ ready · 2sec}	`cortex-a7`	25.4 MB → 29.5 MB
⚠️ +16.0%	trunet _{RSS @ ready · pulse1_f16}	`i9-11900kb_rtx-4060`	31.3 MB → 36.3 MB
⚠️ +16.0%	speaker_id _{RSS @ load · pulse8}	`beaglev-ahead`	16.1 MB → 18.7 MB
⚠️ +16.0%	hey_snips_v4_model17_nnef _{RSS @ ready · pulse8}	`cortex-a55`	24.5 MB → 28.5 MB
⚠️ +16.0%	inceptionv3 _{RSS @ load · pass_mt}	`cortex-a9`	116 MB → 134 MB
⚠️ +15.9%	en_tdnn_15M _{RSS @ load · pulse_240ms}	`cortex-a7`	90.3 MB → 105 MB
⚠️ +15.8%	speaker_id _{RSS @ load · pulse8}	`orangepi-rv2`	16.1 MB → 18.6 MB
⚠️ +15.8%	en_tdnn_lstm_bn_q7 _{RSS @ ready · 2600ms}	`i9-11900kb_rtx-4060`	31.9 MB → 36.9 MB
⚠️ +15.4%	en_tdnn_15M _{RSS @ ready · pulse_240ms}	`cortex-a7`	91.4 MB → 106 MB
⚠️ +15.4%	hey_snips_v1 _{RSS @ load · 400ms}	`cortex-a9`	16.6 MB → 19.2 MB
⚠️ +15.3%	speaker_id _{RSS @ ready · pulse8}	`i9-11900kb_rtx-4060`	27.1 MB → 31.2 MB
⚠️ +15.3%	en_tdnn_15M _{RSS @ ready · 2600ms}	`cortex-a9`	90.6 MB → 104 MB
⚠️ +15.3%	hey_snips_v31 _{RSS @ ready · 400ms}	`cortex-a55`	21.1 MB → 24.3 MB
⚠️ +15.2%	en_tdnn_15M _{RSS @ load · 2600ms}	`cortex-a9`	89.6 MB → 103 MB
⚠️ +15.1%	hey_snips_v4_model17_nnef _{RSS @ ready · pulse8}	`cortex-a53`	24.8 MB → 28.5 MB
⚠️ +14.9%	mdl_en_2019_Q3_librispeech_onnx _{RSS @ ready · 2600ms}	`i9-11900kb_rtx-4060`	41 MB → 47.2 MB
⚠️ +14.9%	trunet _{RSS @ ready · pulse1_f16}	`cortex-a55`	26.5 MB → 30.4 MB
⚠️ +14.9%	en_tdnn_15M _{RSS @ ready · pulse_120ms}	`orangepi-rv2`	87.6 MB → 101 MB
⚠️ +14.8%	nemotron_3_5_asr_streaming_0_6b_f32f32_preprocessor_1s _{RSS @ load · cuda}	`i9-11900kb_rtx-4060`	24.1 MB → 27.6 MB
⚠️ +14.7%	hey_snips_v4_model17 _{RSS @ ready · pulse8}	`cortex-a9`	26.4 MB → 30.2 MB
⚠️ +14.7%	en_tdnn_15M _{RSS @ ready · pulse_240ms}	`orangepi-rv2`	87.6 MB → 100 MB
⚠️ +14.6%	trunet _{RSS @ ready · pulse1_f16}	`cortex-a9`	22.9 MB → 26.3 MB
⚠️ +14.5%	en_tdnn_15M _{RSS @ ready · 2600ms}	`orangepi-rv2`	86.5 MB → 99 MB
⚠️ +14.4%	hey_snips_v4_model17_nnef _{RSS @ ready · pulse8}	`cortex-a7`	19.8 MB → 22.7 MB
⚠️ +14.4%	en_tdnn_8M _{RSS @ ready · pulse_180ms}	`cortex-a53`	63.9 MB → 73.2 MB
⚠️ +14.2%	inceptionv3 _{RSS @ load · pass}	`cortex-a9`	116 MB → 132 MB
⚠️ +14.2%	en_tdnn_15M _{RSS @ load · pulse_240ms}	`cortex-a53`	95 MB → 108 MB
⚠️ +14.2%	en_tdnn_8M _{RSS @ load · 2600ms}	`cortex-a9`	51.4 MB → 58.6 MB
⚠️ +14.2%	parakeet_tdt_600m_v3_f32f32_preprocessor_1s _{RSS @ ready · cpu}	`i9-11900kb_rtx-4060`	26.1 MB → 29.7 MB
⚠️ +14.0%	hey_snips_v4_model17 _{RSS @ ready · 2sec}	`i9-11900kb_rtx-4060`	36.1 MB → 41.2 MB
⚠️ +14.0%	hey_snips_v4_model17_nnef _{RSS @ load · pulse8}	`cortex-a55`	21.9 MB → 25 MB
⚠️ +14.0%	dummy_conmer_12M _{RSS @ ready · pulse_120ms}	`cortex-a7`	76.4 MB → 87.1 MB
⚠️ +13.8%	nemotron_3_5_asr_streaming_0_6b_f32f32_preprocessor_1s _{RSS @ load · cpu}	`i9-11900kb_rtx-4060`	24 MB → 27.4 MB
⚠️ +13.8%	trunet _{RSS @ load · pulse1_f32}	`cortex-a53`	25.6 MB → 29.1 MB
⚠️ +13.6%	parakeet_tdt_600m_v3_f32f32_joint_pass _{RSS @ ready · cpu}	`i9-11900kb_rtx-4060`	51.9 MB → 58.9 MB
⚠️ +13.6%	trunet _{RSS @ ready · pulse1_f32}	`cortex-a7`	24.8 MB → 28.1 MB
⚠️ +13.5%	inceptionv1q _{RSS @ ready · pass}	`cortex-a55`	42 MB → 47.6 MB
⚠️ +13.5%	en_tdnn_8M _{RSS @ ready · pulse_180ms}	`i9-11900kb_rtx-4060`	74.9 MB → 85.1 MB
⚠️ +13.4%	nemotron_3_5_asr_streaming_0_6b_f32f32_preprocessor_pulse100ms _{RSS @ load · cpu}	`i9-11900kb_rtx-4060`	26.1 MB → 29.6 MB
⚠️ +13.2%	trunet _{RSS @ load · pulse1_f16}	`cortex-a53`	26.1 MB → 29.6 MB
⚠️ +13.2%	en_tdnn_8M _{RSS @ ready · pulse_240ms}	`i9-11900kb_rtx-4060`	76 MB → 86 MB
⚠️ +13.2%	dummy_conmer_12M _{RSS @ ready · pulse_120ms}	`cortex-a9`	76.7 MB → 86.8 MB
⚠️ +13.1%	en_tdnn_lstm_bn_q7 _{RSS @ ready · pulse_240ms}	`cortex-a55`	27.7 MB → 31.3 MB
⚠️ +13.1%	en_tdnn_8M _{RSS @ ready · 2600ms}	`cortex-a53`	63.2 MB → 71.5 MB
⚠️ +13.1%	nemotron_3_5_asr_streaming_0_6b_f32f32_joint_pass _{RSS @ ready · cpu}	`i9-11900kb_rtx-4060`	64.3 MB → 72.7 MB
⚠️ +13.1%	arm_ml_kws_cnn_m _{RSS @ ready · pass}	`cortex-a9`	20.2 MB → 22.8 MB
⚠️ +13.0%	en_tdnn_8M _{RSS @ load · pulse_120ms}	`i9-11900kb_rtx-4060`	66.2 MB → 74.8 MB
⚠️ +13.0%	mdl_en_2019_Q3_librispeech_onnx _{RSS @ load · pulse_240ms}	`i9-11900kb_rtx-4060`	38.2 MB → 43.2 MB
⚠️ +12.9%	hey_snips_v4_model17_nnef _{RSS @ load · pulse8}	`cortex-a53`	22 MB → 24.8 MB
⚠️ +12.8%	nemotron_3_5_asr_streaming_0_6b_f32f32_preprocessor_pulse100ms _{RSS @ load · cuda}	`i9-11900kb_rtx-4060`	26.4 MB → 29.8 MB
⚠️ +12.8%	trunet _{RSS @ ready · pulse1_f32}	`cortex-a55`	27.7 MB → 31.2 MB
⚠️ +12.8%	trunet _{RSS @ load · pulse1_f32}	`cortex-a7`	22.2 MB → 25.1 MB
⚠️ +12.8%	en_tdnn_pyt_15M _{RSS @ ready · pulse_120ms}	`cortex-a53`	81.8 MB → 92.2 MB
⚠️ +12.6%	hey_snips_v31 _{RSS @ ready · 400ms}	`cortex-a53`	21.1 MB → 23.7 MB
⚠️ +12.6%	en_tdnn_8M _{RSS @ load · 2600ms}	`i9-11900kb_rtx-4060`	64.8 MB → 72.9 MB
⚠️ +12.5%	trunet _{RSS @ ready · pulse1_f32}	`cortex-a53`	28 MB → 31.5 MB
⚠️ +12.5%	en_tdnn_15M _{load · 2600ms}	`cortex-a53`	1.46 s → 1.64 s
⚠️ +12.4%	mdl_en_2019_Q3_librispeech_onnx _{RSS @ ready · 2600ms}	`cortex-a55`	35.4 MB → 39.8 MB
⚠️ +12.4%	trunet _{RSS @ load · pulse1_f32}	`i9-11900kb_rtx-4060`	30.7 MB → 34.6 MB
⚠️ +12.4%	en_tdnn_15M _{RSS @ ready · pulse_120ms}	`beaglev-ahead`	88 MB → 98.9 MB
⚠️ +12.3%	en_tdnn_15M_nnef _{RSS @ ready · pulse_240ms}	`orangepi-rv2`	85.6 MB → 96.1 MB
⚠️ +12.2%	mdl_en_2019_Q3_librispeech_onnx _{RSS @ ready · pulse_240ms}	`cortex-a55`	36.5 MB → 40.9 MB
⚠️ +12.2%	en_tdnn_8M _{RSS @ load · pulse_180ms}	`i9-11900kb_rtx-4060`	65.2 MB → 73.1 MB
⚠️ +12.1%	nemotron_3_5_asr_streaming_0_6b_f32f32_preprocessor_1s _{RSS @ ready · cpu}	`i9-11900kb_rtx-4060`	25.7 MB → 28.8 MB
⚠️ +12.0%	hey_snips_v4_model17 _{RSS @ load · pulse8}	`cortex-a7`	23.9 MB → 26.8 MB
⚠️ +12.0%	hey_snips_v4_model17 _{RSS @ load · pulse8}	`i9-11900kb_rtx-4060`	36.6 MB → 41 MB
⚠️ +12.0%	en_tdnn_15M _{RSS @ load · 2600ms}	`cortex-a53`	94.6 MB → 106 MB
⚠️ +11.7%	mdl_en_2019_Q3_librispeech_onnx _{RSS @ ready · pulse_240ms}	`cortex-a53`	36.5 MB → 40.8 MB
⚠️ +11.6%	en_tdnn_15M _{RSS @ load · pulse_120ms}	`cortex-a53`	94.8 MB → 106 MB
⚠️ +11.6%	en_tdnn_15M _{RSS @ load · pulse_240ms}	`cortex-a55`	95.2 MB → 106 MB
⚠️ +11.6%	inceptionv1q _{RSS @ ready · pass}	`cortex-a53`	41.8 MB → 46.6 MB
⚠️ +11.5%	en_tdnn_15M_nnef _{RSS @ ready · pulse_240ms}	`beaglev-ahead`	85.8 MB → 95.6 MB
⚠️ +11.3%	mobilenet_v2_1 _{RSS @ ready · pass}	`cortex-a55`	53.1 MB → 59.2 MB
⚠️ +11.3%	openelm_270M_q40ef16_541 _{RSS @ load · cpu}	`i9-11900kb_rtx-4060`	207 MB → 231 MB
⚠️ +11.2%	hey_snips_v1 _{RSS @ ready · 400ms}	`cortex-a55`	22.1 MB → 24.6 MB
⚠️ +11.2%	hey_snips_v1 _{RSS @ ready · 400ms}	`cortex-a53`	22.1 MB → 24.5 MB
⚠️ +11.2%	mdl_en_2019_Q3_librispeech_onnx _{RSS @ ready · 2600ms}	`cortex-a53`	35.6 MB → 39.6 MB
⚠️ +10.9%	openelm_270M_q40ef16_541 _{RSS @ load · cuda}	`i9-11900kb_rtx-4060`	207 MB → 230 MB
⚠️ +10.8%	arm_ml_kws_cnn_m _{RSS @ ready · pass}	`cortex-a7`	20.1 MB → 22.2 MB
⚠️ +10.5%	en_tdnn_15M _{RSS @ load · pulse_120ms}	`cortex-a55`	94.7 MB → 105 MB
⚠️ +10.4%	en_tdnn_15M _{RSS @ load · 2600ms}	`cortex-a55`	94.8 MB → 105 MB
⚠️ +10.2%	dummy_conmer_12M _{RSS @ load · pulse_120ms}	`cortex-a53`	79.6 MB → 87.8 MB
⚠️ +10.1%	mdl_en_2019_Q3_librispeech_onnx _{RSS @ ready · pulse_240ms}	`i9-11900kb_rtx-4060`	42.7 MB → 47 MB
⚠️ +9.8%	en_tdnn_15M _{load · pulse_240ms}	`cortex-a53`	1.71 s → 1.87 s
⚠️ +9.7%	dummy_conmer_12M _{RSS @ load · pulse_120ms}	`i9-11900kb_rtx-4060`	86.8 MB → 95.2 MB
⚠️ +9.7%	dummy_conmer_12M _{RSS @ load · pulse_120ms}	`cortex-a55`	79.8 MB → 87.6 MB
⚠️ +9.7%	en_tdnn_15M _{load · pulse_120ms}	`cortex-a53`	1.71 s → 1.87 s
⚠️ +9.6%	arm_ml_kws_cnn_m _{RSS @ load · pass}	`cortex-a7`	17.9 MB → 19.6 MB
⚠️ +9.6%	voicecom_float _{RSS @ ready · 2sec}	`cortex-a53`	31.6 MB → 34.6 MB
⚠️ +9.5%	arm_ml_kws_cnn_m _{RSS @ load · pass}	`cortex-a9`	18 MB → 19.7 MB
⚠️ +9.5%	en_tdnn_15M_nnef _{RSS @ ready · pulse_240ms}	`i9-11900kb_rtx-4060`	96.9 MB → 106 MB
⚠️ +9.5%	en_tdnn_15M _{load+optimize · 2600ms}	`cortex-a53`	1.77 s → 1.94 s
⚠️ +9.3%	hey_snips_v4_model17 _{RSS @ ready · pulse8}	`cortex-a7`	26.5 MB → 28.9 MB
⚠️ +8.9%	en_tdnn_15M_nnef _{RSS @ ready · pulse_240ms}	`cortex-a53`	91.9 MB → 100 MB
⚠️ +8.7%	hey_snips_v1 _{RSS @ load · 400ms}	`cortex-a53`	19.6 MB → 21.4 MB
⚠️ +8.4%	voicecom_float _{RSS @ ready · 2sec}	`cortex-a55`	31.9 MB → 34.6 MB
⚠️ +8.3%	en_tdnn_15M _{load · 2600ms}	`cortex-a55`	1.05 s → 1.14 s
⚠️ +8.0%	en_tdnn_15M_nnef _{RSS @ ready · pulse_240ms}	`cortex-a55`	91.9 MB → 99.2 MB
⚠️ +8.0%	en_tdnn_8M _{RSS @ ready · 2600ms}	`i9-11900kb_rtx-4060`	73.8 MB → 79.7 MB
⚠️ +8.0%	en_tdnn_8M_nnef _{RSS @ load · pulse_240ms}	`i9-11900kb_rtx-4060`	55 MB → 59.4 MB
⚠️ +7.9%	dummy_conmer_12M _{RSS @ load · pulse_120ms}	`cortex-a7`	73.9 MB → 79.8 MB
⚠️ +7.6%	en_tdnn_15M _{load+optimize · pulse_240ms}	`cortex-a53`	2.02 s → 2.17 s
⚠️ +7.6%	en_tdnn_15M_nnef _{RSS @ ready · pulse_240ms}	`cortex-a7`	89.1 MB → 95.9 MB
⚠️ +7.5%	dummy_conmer_12M _{RSS @ load · pulse_120ms}	`cortex-a9`	74.2 MB → 79.7 MB
⚠️ +7.5%	en_tdnn_15M_nnef _{RSS @ ready · pulse_240ms}	`cortex-a9`	89.3 MB → 96 MB
⚠️ +7.5%	en_tdnn_15M _{load+optimize · pulse_120ms}	`cortex-a53`	2.02 s → 2.18 s
⚠️ +7.4%	en_tdnn_8M_nnef _{RSS @ load · pulse_240ms}	`orangepi-rv2`	46 MB → 49.4 MB
⚠️ +7.0%	en_tdnn_15M _{load · 2600ms}	`cortex-a9`	2.53 s → 2.71 s
⚠️ +7.0%	en_tdnn_15M _{load · pulse_240ms}	`cortex-a55`	1.26 s → 1.35 s
⚠️ +6.9%	en_tdnn_8M_nnef _{RSS @ load · pulse_240ms}	`beaglev-ahead`	46.1 MB → 49.2 MB
⚠️ +6.5%	en_tdnn_15M _{load · pulse_120ms}	`cortex-a55`	1.25 s → 1.33 s
⚠️ +6.3%	arm_ml_kws_cnn_m _{RSS @ load · pass}	`apple-m1-max`	18.7 MB → 19.8 MB
⚠️ +6.2%	en_tdnn_15M _{load+optimize · 2600ms}	`cortex-a55`	1.26 s → 1.34 s
⚠️ +5.9%	arm_ml_kws_cnn_m _{RSS @ load · pass}	`cortex-a55`	20.4 MB → 21.6 MB
⚠️ +5.8%	parakeet_tdt_600m_v3_f32f32_preprocessor_1s _{RSS @ ready · cuda}	`jetson-orin-nx`	191 MB → 202 MB
⚠️ +5.8%	en_tdnn_15M _{load · pulse_120ms}	`cortex-a9`	2.88 s → 3.05 s
⚠️ +5.5%	en_tdnn_15M _{load · pulse_240ms}	`cortex-a9`	2.88 s → 3.04 s
⚠️ +5.2%	en_tdnn_8M_nnef _{RSS @ load · pulse_240ms}	`cortex-a55`	51.7 MB → 54.3 MB
⚠️ +5.0%	llama_3_2_1B_q40ef32_516 _{RSS @ ready · cpu}	`i9-11900kb_rtx-4060`	729 MB → 766 MB

github-actions · 2026-10-13T12:27:55Z

🔴 Bench vs main — 1 speed regression(s) · ⚠️ 127 secondary

Reference: 2026-10-13 morning nightly run (0d old) · full report → run

Speed — evaltime · prefill · decode

Δ	metric	device	main → PR
🔴 +31.6%	inceptionv3 _{evaltime · pass}	`orangepi-rv2`	2.17 s → 2.86 s

Improvements

Δ	metric	device	main → PR
🟢 -41.3%	parakeet_tdt_600m_v3_f32f32_preprocessor_1s _{evaltime · cpu}	`i9-9900k`	0.581 ms → 0.341 ms
🟢 -40.9%	mobilenet_v2_1 _{evaltime · pass_mt}	`orangepi-rv2`	252 ms → 149 ms
🟢 -39.5%	nemotron_3_5_asr_streaming_0_6b_f32f32_preprocessor_1s _{evaltime · cpu}	`i9-9900k`	0.462 ms → 0.28 ms
🟢 -38.5%	trunet _{evaltime · pulse1_f32}	`orangepi-rv2`	6.01 ms/pulse _{0.601 RTF} → 3.69 ms/pulse _{0.369 RTF}
🟢 -36.8%	nemotron_3_5_asr_streaming_0_6b_f32f32_preprocessor_1s _{evaltime · cpu}	`i9-11900kb_rtx-4060`	0.401 ms → 0.253 ms

+53 more improvement(s)

Δ	metric	device	main → PR
🟢 -35.6%	speaker_id _{evaltime · pulse8}	`orangepi-rv2`	2.24 ms/pulse _{0.028 RTF} → 1.44 ms/pulse _{0.018 RTF}
🟢 -26.3%	speaker_id _{evaltime · pulse8}	`beaglev-ahead`	2.66 ms/pulse _{0.0333 RTF} → 1.96 ms/pulse _{0.0245 RTF}
🟢 -24.5%	mobilenet_v2_1 _{evaltime · pass_mt}	`beaglev-ahead`	328 ms → 247 ms
🟢 -22.0%	hey_snips_v4_model17_nnef _{evaltime · pulse8}	`beaglev-ahead`	5.19 ms/pulse _{0.0649 RTF} → 4.05 ms/pulse _{0.0506 RTF}
🟢 -21.4%	en_tdnn_8M _{evaltime · pulse_240ms}	`orangepi-rv2`	41.9 ms/pulse _{0.175 RTF} → 32.9 ms/pulse _{0.137 RTF}
🟢 -21.2%	hey_snips_v4_model17 _{evaltime · 2sec}	`orangepi-rv2`	39.5 ms → 31.2 ms
🟢 -19.7%	inceptionv3 _{evaltime · pass_mt}	`orangepi-rv2`	1.04 s → 833 ms
🟢 -18.6%	hey_snips_v31 _{evaltime · 400ms}	`orangepi-rv2`	2.29 ms → 1.86 ms
🟢 -17.9%	en_tdnn_8M _{evaltime · pulse_180ms}	`orangepi-rv2`	37.4 ms/pulse _{0.208 RTF} → 30.7 ms/pulse _{0.171 RTF}
🟢 -17.6%	hey_snips_v4_model17 _{evaltime · 2sec}	`beaglev-ahead`	41.7 ms → 34.4 ms
🟢 -17.2%	trunet _{evaltime · pulse1_f32}	`beaglev-ahead`	5.32 ms/pulse _{0.532 RTF} → 4.4 ms/pulse _{0.44 RTF}
🟢 -16.4%	mobilenet_v1_1 _{evaltime · pass}	`orangepi-rv2`	368 ms → 308 ms
🟢 -16.3%	en_tdnn_15M _{evaltime · pulse_240ms}	`orangepi-rv2`	72 ms/pulse _{0.3 RTF} → 60.3 ms/pulse _{0.251 RTF}
🟢 -16.0%	en_tdnn_8M _{evaltime · pulse_120ms}	`orangepi-rv2`	29.2 ms/pulse _{0.244 RTF} → 24.6 ms/pulse _{0.205 RTF}
🟢 -15.9%	hey_snips_v4_model17_nnef _{evaltime · pulse8}	`orangepi-rv2`	4.77 ms/pulse _{0.0596 RTF} → 4.01 ms/pulse _{0.0501 RTF}
🟢 -15.8%	en_tdnn_15M_nnef _{evaltime · pulse_240ms}	`orangepi-rv2`	71.4 ms/pulse _{0.297 RTF} → 60.1 ms/pulse _{0.25 RTF}
🟢 -15.7%	mobilenet_v2_1 _{evaltime · pass}	`orangepi-rv2`	556 ms → 468 ms
🟢 -15.7%	en_tdnn_8M_nnef _{evaltime · pulse_240ms}	`orangepi-rv2`	39.6 ms/pulse _{0.165 RTF} → 33.4 ms/pulse _{0.139 RTF}
🟢 -14.8%	en_tdnn_pyt_15M _{evaltime · pulse_120ms}	`orangepi-rv2`	37.7 ms/pulse _{0.314 RTF} → 32.1 ms/pulse _{0.268 RTF}
🟢 -12.9%	voicecom_fake_quant _{evaltime · 2sec}	`orangepi-rv2`	65.3 ms → 56.9 ms
🟢 -12.8%	inceptionv3 _{evaltime · pass_mt}	`beaglev-ahead`	1.78 s → 1.55 s
🟢 -12.5%	hey_snips_v4_model17 _{evaltime · pulse8}	`beaglev-ahead`	4.7 ms/pulse _{0.0587 RTF} → 4.11 ms/pulse _{0.0514 RTF}
🟢 -12.3%	hey_snips_v4_model17 _{evaltime · pulse8}	`orangepi-rv2`	4.6 ms/pulse _{0.0575 RTF} → 4.03 ms/pulse _{0.0504 RTF}
🟢 -12.3%	arm_ml_kws_cnn_m _{evaltime · pass}	`orangepi-rv2`	3.27 ms → 2.87 ms
🟢 -12.1%	mobilenet_v2_1 _{evaltime · pass}	`beaglev-ahead`	855 ms → 752 ms
🟢 -11.9%	en_tdnn_15M _{evaltime · 2600ms}	`orangepi-rv2`	752 ms → 662 ms
🟢 -10.3%	trunet _{evaltime · pulse1_f16}	`orangepi-rv2`	7.15 ms/pulse _{0.715 RTF} → 6.41 ms/pulse _{0.641 RTF}
🟢 -10.2%	mobilenet_v2_1 _{evaltime · pass_mt}	`cortex-a9`	552 ms → 496 ms
🟢 -9.5%	en_tdnn_15M _{evaltime · pulse_120ms}	`orangepi-rv2`	50.7 ms/pulse _{0.423 RTF} → 45.9 ms/pulse _{0.383 RTF}
🟢 -9.5%	hey_snips_v4_model17 _{evaltime · pulse8}	`cortex-a9`	6.38 ms/pulse _{0.0797 RTF} → 5.77 ms/pulse _{0.0721 RTF}
🟢 -9.4%	en_tdnn_8M _{evaltime · 2600ms}	`orangepi-rv2`	387 ms → 351 ms
🟢 -9.3%	hey_snips_v4_model17_nnef _{evaltime · pulse8}	`cortex-a9`	6.35 ms/pulse _{0.0794 RTF} → 5.76 ms/pulse _{0.072 RTF}
🟢 -9.0%	inceptionv1q _{evaltime · pass}	`orangepi-rv2`	654 ms → 596 ms
🟢 -8.6%	hey_snips_v4_model17_nnef _{evaltime · pulse8}	`cortex-a7`	7.16 ms/pulse _{0.0895 RTF} → 6.54 ms/pulse _{0.0817 RTF}
🟢 -8.6%	mobilenet_v1_1 _{evaltime · pass}	`beaglev-ahead`	663 ms → 607 ms
🟢 -8.4%	hey_snips_v4_model17_nnef _{evaltime · pulse8}	`cortex-a55`	2.08 ms/pulse _{0.026 RTF} → 1.9 ms/pulse _{0.0238 RTF}
🟢 -8.1%	mobilenet_v2_1 _{evaltime · pass_mt}	`cortex-a7`	534 ms → 491 ms
🟢 -8.0%	hey_snips_v31 _{evaltime · 400ms}	`beaglev-ahead`	4.67 ms → 4.3 ms
🟢 -7.8%	hey_snips_v4_model17 _{evaltime · pulse8}	`cortex-a7`	7.15 ms/pulse _{0.0893 RTF} → 6.59 ms/pulse _{0.0823 RTF}
🟢 -7.5%	voicecom_fake_quant _{evaltime · 2sec}	`i9-9900k`	5.38 ms → 4.98 ms
🟢 -7.0%	trunet _{evaltime · pulse1_f32}	`cortex-a7`	8.1 ms/pulse _{0.81 RTF} → 7.53 ms/pulse _{0.753 RTF}
🟢 -6.9%	hey_snips_v4_model17_nnef _{evaltime · pulse8}	`cortex-a53`	2.72 ms/pulse _{0.034 RTF} → 2.53 ms/pulse _{0.0316 RTF}
🟢 -6.4%	en_tdnn_8M _{evaltime · pulse_120ms}	`beaglev-ahead`	40.7 ms/pulse _{0.339 RTF} → 38.1 ms/pulse _{0.318 RTF}
🟢 -6.4%	hey_snips_v4_model17 _{evaltime · pulse8}	`cortex-a53`	2.71 ms/pulse _{0.0338 RTF} → 2.53 ms/pulse _{0.0317 RTF}
🟢 -6.4%	trunet _{evaltime · pulse1_f32}	`cortex-a9`	8.19 ms/pulse _{0.819 RTF} → 7.67 ms/pulse _{0.767 RTF}
🟢 -6.2%	en_tdnn_pyt_15M _{evaltime · pulse_120ms}	`beaglev-ahead`	60 ms/pulse _{0.5 RTF} → 56.2 ms/pulse _{0.469 RTF}
🟢 -6.2%	en_tdnn_8M _{evaltime · pulse_240ms}	`beaglev-ahead`	80.7 ms/pulse _{0.336 RTF} → 75.7 ms/pulse _{0.315 RTF}
🟢 -6.1%	en_tdnn_8M _{evaltime · pulse_180ms}	`beaglev-ahead`	72.6 ms/pulse _{0.403 RTF} → 68.1 ms/pulse _{0.379 RTF}
🟢 -6.0%	mobilenet_v2_1 _{evaltime · pass_mt}	`cortex-a53`	178 ms → 167 ms
🟢 -5.6%	mobilenet_v2_1 _{evaltime · pass}	`cortex-a9`	1.4 s → 1.32 s
🟢 -5.5%	en_tdnn_15M _{evaltime · 2600ms}	`beaglev-ahead`	1.69 s → 1.59 s
🟢 -5.5%	en_tdnn_lstm_bn_q7 _{evaltime · 2600ms}	`cortex-a7`	371 ms → 351 ms
🟢 -5.0%	en_tdnn_8M _{evaltime · 2600ms}	`beaglev-ahead`	820 ms → 779 ms

⚠️ 127 secondary regression(s)

Δ	metric	device	main → PR
⚠️ +33.8%	nemotron_3_5_asr_streaming_0_6b_f32f32_preprocessor_1s _{RSS @ load · cuda}	`jetson-orin-nx`	20.6 MB → 27.6 MB
⚠️ +28.6%	arm_ml_kws_cnn_m _{RSS @ load · pass}	`i9-9900k`	13.1 MB → 16.8 MB
⚠️ +28.5%	voicecom_float _{RSS @ load · 2sec}	`i9-9900k`	16.6 MB → 21.3 MB
⚠️ +26.9%	nemotron_3_5_asr_streaming_0_6b_f32f32_preprocessor_1s _{RSS @ load · cpu}	`i9-9900k`	15.4 MB → 19.6 MB
⚠️ +26.9%	hey_snips_v1 _{RSS @ load · 400ms}	`i9-9900k`	13.2 MB → 16.8 MB
⚠️ +25.8%	nemotron_3_5_asr_streaming_0_6b_f32f32_preprocessor_1s _{RSS @ ready · cpu}	`i9-9900k`	16.3 MB → 20.5 MB
⚠️ +25.4%	hey_snips_v31 _{RSS @ ready · 400ms}	`i9-9900k`	14.4 MB → 18 MB
⚠️ +25.3%	parakeet_tdt_600m_v3_f32f32_preprocessor_1s _{RSS @ load · cpu}	`i9-9900k`	15.6 MB → 19.6 MB
⚠️ +24.8%	parakeet_tdt_600m_v3_f32f32_preprocessor_1s _{RSS @ ready · cpu}	`i9-9900k`	16.5 MB → 20.6 MB
⚠️ +24.8%	hey_snips_v31 _{RSS @ load · 400ms}	`i9-9900k`	13.9 MB → 17.4 MB
⚠️ +24.3%	nemotron_3_5_asr_streaming_0_6b_f32f32_preprocessor_pulse100ms _{RSS @ ready · cpu}	`i9-9900k`	17.1 MB → 21.3 MB
⚠️ +23.7%	parakeet_tdt_600m_v3_f32f32_preprocessor_1s _{RSS @ load · cuda}	`jetson-orin-nx`	20.7 MB → 25.6 MB
⚠️ +22.9%	nemotron_3_5_asr_streaming_0_6b_f32f32_preprocessor_pulse100ms _{RSS @ load · cpu}	`i9-9900k`	16.7 MB → 20.5 MB
⚠️ +21.6%	voicecom_fake_quant _{RSS @ load · 2sec}	`i9-9900k`	17.8 MB → 21.6 MB
⚠️ +20.9%	speaker_id _{RSS @ ready · pulse8}	`i9-9900k`	17.2 MB → 20.8 MB
⚠️ +20.5%	speaker_id _{RSS @ load · pulse8}	`i9-9900k`	16.5 MB → 19.8 MB
⚠️ +18.9%	voicecom_float _{RSS @ load · 2sec}	`i9-11900kb_rtx-4060`	24.7 MB → 29.3 MB
⚠️ +18.4%	nemotron_3_5_asr_streaming_0_6b_f32f32_preprocessor_1s _{RSS @ load · cuda}	`i9-11900kb_rtx-4060`	24.2 MB → 28.6 MB
⚠️ +17.8%	arm_ml_kws_cnn_m _{RSS @ ready · pass}	`i9-9900k`	15.3 MB → 18.1 MB
⚠️ +17.7%	hey_snips_v1 _{RSS @ load · 400ms}	`i9-11900kb_rtx-4060`	22 MB → 25.8 MB
⚠️ +17.5%	nemotron_3_5_asr_streaming_0_6b_f32f32_preprocessor_1s _{RSS @ load · cpu}	`i9-11900kb_rtx-4060`	24.2 MB → 28.5 MB
⚠️ +16.9%	parakeet_tdt_600m_v3_f32f32_preprocessor_1s _{RSS @ load · cpu}	`i9-11900kb_rtx-4060`	24.5 MB → 28.6 MB
⚠️ +16.9%	parakeet_tdt_600m_v3_f32f32_preprocessor_1s _{RSS @ ready · cpu}	`i9-11900kb_rtx-4060`	26 MB → 30.4 MB
⚠️ +16.7%	arm_ml_kws_cnn_m _{RSS @ load · pass}	`i9-11900kb_rtx-4060`	23.1 MB → 27 MB
⚠️ +16.1%	parakeet_tdt_600m_v3_f32f32_preprocessor_1s _{RSS @ load · cuda}	`i9-11900kb_rtx-4060`	24.5 MB → 28.4 MB
⚠️ +15.8%	nemotron_3_5_asr_streaming_0_6b_f32f32_preprocessor_1s _{RSS @ ready · cpu}	`i9-11900kb_rtx-4060`	25.9 MB → 29.9 MB
⚠️ +15.7%	nemotron_3_5_asr_streaming_0_6b_f32f32_preprocessor_pulse100ms _{RSS @ load · cuda}	`i9-11900kb_rtx-4060`	26.4 MB → 30.5 MB
⚠️ +15.5%	nemotron_3_5_asr_streaming_0_6b_f32f32_preprocessor_pulse100ms _{RSS @ ready · cpu}	`i9-11900kb_rtx-4060`	27.4 MB → 31.6 MB
⚠️ +15.5%	speaker_id _{RSS @ load · pulse8}	`cortex-a9`	18.4 MB → 21.2 MB
⚠️ +15.2%	voicecom_fake_quant _{RSS @ load · 2sec}	`cortex-a9`	20.1 MB → 23.1 MB
⚠️ +14.9%	nemotron_3_5_asr_streaming_0_6b_f32f32_preprocessor_pulse100ms _{RSS @ load · cpu}	`i9-11900kb_rtx-4060`	26.4 MB → 30.4 MB
⚠️ +14.8%	voicecom_fake_quant _{RSS @ load · 2sec}	`i9-11900kb_rtx-4060`	26.8 MB → 30.8 MB
⚠️ +14.3%	hey_snips_v31 _{RSS @ load · 400ms}	`i9-11900kb_rtx-4060`	23.9 MB → 27.3 MB
⚠️ +14.2%	hey_snips_v31 _{RSS @ ready · 400ms}	`i9-11900kb_rtx-4060`	25.5 MB → 29.1 MB
⚠️ +13.7%	en_tdnn_15M _{load · 2600ms}	`cortex-a53`	1.43 s → 1.62 s
⚠️ +13.3%	voicecom_fake_quant _{RSS @ load · 2sec}	`cortex-a7`	20 MB → 22.7 MB
⚠️ +12.7%	speaker_id _{RSS @ load · pulse8}	`cortex-a7`	18.5 MB → 20.8 MB
⚠️ +12.7%	speaker_id _{RSS @ ready · pulse8}	`i9-11900kb_rtx-4060`	27.7 MB → 31.2 MB
⚠️ +12.7%	hey_snips_v1 _{load+optimize · 400ms}	`cortex-a53`	63 ms → 71 ms
⚠️ +12.7%	parakeet_tdt_600m_v3_f32f32_decoder_pass _{RSS @ load · cuda}	`i9-11900kb_rtx-4060`	69.1 MB → 77.8 MB
⚠️ +12.5%	hey_snips_v1 _{RSS @ ready · 400ms}	`i9-11900kb_rtx-4060`	24.3 MB → 27.4 MB
⚠️ +12.4%	en_tdnn_lstm_bn_q7 _{RSS @ load · pulse_240ms}	`i9-9900k`	19.5 MB → 22 MB
⚠️ +12.2%	hey_snips_v1 _{load · 400ms}	`cortex-a53`	41 ms → 46 ms
⚠️ +12.0%	arm_ml_kws_cnn_m _{load+optimize · pass}	`cortex-a55`	50 ms → 56 ms
⚠️ +11.7%	hey_snips_v4_model17_nnef _{RSS @ load · pulse8}	`i9-9900k`	16.3 MB → 18.2 MB
⚠️ +11.5%	trunet _{RSS @ ready · pulse1_f32}	`i9-9900k`	22 MB → 24.5 MB
⚠️ +11.5%	trunet _{RSS @ load · pulse1_f32}	`i9-9900k`	20.7 MB → 23.1 MB
⚠️ +11.4%	voicecom_fake_quant _{RSS @ load · 2sec}	`cortex-a55`	22.2 MB → 24.8 MB
⚠️ +11.3%	en_tdnn_15M _{load+optimize · 2600ms}	`cortex-a53`	1.74 s → 1.94 s
⚠️ +11.3%	en_tdnn_15M _{load · pulse_240ms}	`cortex-a53`	1.68 s → 1.87 s
⚠️ +11.3%	en_tdnn_15M _{load · pulse_120ms}	`cortex-a53`	1.68 s → 1.87 s
⚠️ +11.3%	speaker_id _{RSS @ load · pulse8}	`i9-11900kb_rtx-4060`	26.3 MB → 29.2 MB
⚠️ +11.3%	en_tdnn_lstm_bn_q7 _{RSS @ load · 2600ms}	`i9-9900k`	19.1 MB → 21.3 MB
⚠️ +11.2%	parakeet_tdt_600m_v3_f32f32_decoder_pass _{RSS @ load · cpu}	`i9-9900k`	61.9 MB → 68.9 MB
⚠️ +10.8%	voicecom_fake_quant _{RSS @ load · 2sec}	`cortex-a53`	22.2 MB → 24.6 MB
⚠️ +10.2%	en_tdnn_15M _{load · 2600ms}	`cortex-a55`	1.04 s → 1.15 s
⚠️ +9.9%	arm_ml_kws_cnn_m _{load+optimize · pass}	`cortex-a53`	71 ms → 78 ms
⚠️ +9.8%	parakeet_tdt_600m_v3_f32f32_decoder_pass _{RSS @ load · cpu}	`i9-11900kb_rtx-4060`	69.2 MB → 76 MB
⚠️ +9.6%	en_tdnn_15M _{load+optimize · pulse_240ms}	`cortex-a53`	1.99 s → 2.18 s
⚠️ +9.5%	parakeet_tdt_600m_v3_f32f32_decoder_pass _{RSS @ load · cuda}	`jetson-orin-nx`	65.8 MB → 72 MB
⚠️ +9.5%	en_tdnn_15M _{load+optimize · pulse_120ms}	`cortex-a53`	2 s → 2.19 s
⚠️ +9.4%	en_tdnn_15M _{load · pulse_240ms}	`cortex-a55`	1.24 s → 1.36 s
⚠️ +9.2%	hey_snips_v1 _{RSS @ ready · 400ms}	`i9-9900k`	16.1 MB → 17.6 MB
⚠️ +9.2%	mobilenet_v1_1 _{RSS @ ready · pass}	`orangepi-rv2`	34.3 MB → 37.5 MB
⚠️ +9.0%	en_tdnn_15M _{load · pulse_120ms}	`cortex-a55`	1.24 s → 1.35 s
⚠️ +8.9%	hey_snips_v4_model17_nnef _{RSS @ load · pulse8}	`i9-11900kb_rtx-4060`	24.4 MB → 26.6 MB
⚠️ +8.9%	mdl_en_2019_Q3_librispeech_onnx _{RSS @ load · 2600ms}	`i9-9900k`	27.4 MB → 29.8 MB
⚠️ +8.8%	mdl_en_2019_Q3_librispeech_onnx _{RSS @ load · pulse_240ms}	`i9-9900k`	28.1 MB → 30.5 MB
⚠️ +8.7%	hey_snips_v4_model17_nnef _{RSS @ ready · pulse8}	`i9-9900k`	18.1 MB → 19.6 MB
⚠️ +8.4%	mobilenet_v1_1 _{RSS @ ready · pass}	`beaglev-ahead`	34.3 MB → 37.2 MB
⚠️ +8.4%	en_tdnn_8M _{RSS @ ready · pulse_120ms}	`orangepi-rv2`	50.9 MB → 55.1 MB
⚠️ +8.4%	en_tdnn_8M_nnef _{RSS @ ready · pulse_240ms}	`i9-9900k`	50.6 MB → 54.8 MB
⚠️ +8.3%	en_tdnn_lstm_bn_q7 _{RSS @ ready · 2600ms}	`i9-9900k`	20.7 MB → 22.5 MB
⚠️ +8.3%	nemotron_3_5_asr_streaming_0_6b_f32f32_decoder_pass _{RSS @ load · cpu}	`i9-9900k`	74.6 MB → 80.9 MB
⚠️ +8.2%	en_tdnn_8M _{RSS @ ready · pulse_240ms}	`orangepi-rv2`	50.9 MB → 55.1 MB
⚠️ +8.2%	arm_ml_kws_cnn_m _{load · pass}	`cortex-a53`	49 ms → 53 ms
⚠️ +8.2%	en_tdnn_15M _{load · 2600ms}	`cortex-a9`	2.49 s → 2.69 s
⚠️ +8.1%	mobilenet_v1_1 _{RSS @ ready · pass}	`cortex-a9`	37.6 MB → 40.6 MB
⚠️ +8.0%	en_tdnn_8M_nnef _{RSS @ load · pulse_240ms}	`i9-9900k`	46.8 MB → 50.6 MB
⚠️ +7.9%	en_tdnn_8M_nnef _{RSS @ ready · pulse_240ms}	`orangepi-rv2`	48 MB → 51.8 MB
⚠️ +7.7%	en_tdnn_8M _{RSS @ ready · pulse_180ms}	`orangepi-rv2`	50.9 MB → 54.8 MB
⚠️ +7.6%	en_tdnn_8M_nnef _{RSS @ ready · pulse_240ms}	`i9-11900kb_rtx-4060`	59.7 MB → 64.2 MB
⚠️ +7.6%	en_tdnn_lstm_bn_q7 _{RSS @ load · 2600ms}	`i9-11900kb_rtx-4060`	28.3 MB → 30.4 MB
⚠️ +7.5%	en_tdnn_8M_nnef _{RSS @ ready · pulse_240ms}	`beaglev-ahead`	48.2 MB → 51.8 MB
⚠️ +7.5%	mobilenet_v2_1 _{load · pass_mt}	`cortex-a55`	696 ms → 748 ms
⚠️ +7.5%	nemotron_3_5_asr_streaming_0_6b_f32f32_decoder_pass _{RSS @ load · cuda}	`i9-11900kb_rtx-4060`	81.7 MB → 87.8 MB
⚠️ +7.4%	en_tdnn_8M _{RSS @ ready · pulse_180ms}	`beaglev-ahead`	51.1 MB → 54.9 MB
⚠️ +7.2%	parakeet_tdt_600m_v3_f32f32_joint_pass _{RSS @ ready · cpu}	`i9-9900k`	44.1 MB → 47.2 MB
⚠️ +7.2%	en_tdnn_8M _{RSS @ ready · pulse_240ms}	`beaglev-ahead`	51.1 MB → 54.8 MB
⚠️ +7.2%	nemotron_3_5_asr_streaming_0_6b_f32f32_decoder_pass _{RSS @ load · cpu}	`i9-11900kb_rtx-4060`	81.9 MB → 87.8 MB
⚠️ +7.2%	en_tdnn_8M _{RSS @ ready · pulse_120ms}	`beaglev-ahead`	51.1 MB → 54.8 MB
⚠️ +7.1%	en_tdnn_8M _{RSS @ ready · 2600ms}	`orangepi-rv2`	50.1 MB → 53.6 MB
⚠️ +6.9%	en_tdnn_8M_nnef _{RSS @ load · pulse_240ms}	`i9-11900kb_rtx-4060`	54.9 MB → 58.7 MB
⚠️ +6.9%	mobilenet_v2_1 _{load · pass}	`cortex-a53`	1.12 s → 1.2 s
⚠️ +6.8%	voicecom_float _{load+optimize · 2sec}	`cortex-a53`	147 ms → 157 ms
⚠️ +6.8%	en_tdnn_pyt_15M _{RSS @ ready · pulse_120ms}	`orangepi-rv2`	61.8 MB → 66 MB
⚠️ +6.7%	mobilenet_v1_1 _{load · pass}	`cortex-a53`	713 ms → 761 ms
⚠️ +6.7%	en_tdnn_8M _{RSS @ ready · 2600ms}	`cortex-a7`	53.1 MB → 56.6 MB
⚠️ +6.7%	mobilenet_v1_1 _{load+optimize · pass}	`cortex-a53`	928 ms → 990 ms
⚠️ +6.7%	nemotron_3_5_asr_streaming_0_6b_f32f32_joint_pass _{RSS @ ready · cpu}	`i9-11900kb_rtx-4060`	64.1 MB → 68.4 MB
⚠️ +6.7%	en_tdnn_8M _{RSS @ ready · pulse_240ms}	`cortex-a7`	54.3 MB → 57.9 MB
⚠️ +6.6%	en_tdnn_8M _{RSS @ ready · 2600ms}	`beaglev-ahead`	50.2 MB → 53.6 MB
⚠️ +6.5%	voicecom_float _{load · 2sec}	`cortex-a53`	92 ms → 98 ms
⚠️ +6.5%	mobilenet_v2_1 _{load+optimize · pass_mt}	`cortex-a55`	955 ms → 1.02 s
⚠️ +6.4%	mobilenet_v2_1 _{RSS @ load · pass_mt}	`orangepi-rv2`	39.9 MB → 42.5 MB
⚠️ +6.4%	en_tdnn_8M_nnef _{RSS @ ready · pulse_240ms}	`cortex-a9`	51.9 MB → 55.3 MB
⚠️ +6.3%	hey_snips_v31 _{RSS @ ready · 400ms}	`cortex-a55`	21.1 MB → 22.4 MB
⚠️ +6.3%	en_tdnn_8M _{RSS @ ready · pulse_180ms}	`cortex-a7`	54.2 MB → 57.6 MB
⚠️ +6.1%	en_tdnn_8M_nnef _{RSS @ ready · pulse_240ms}	`cortex-a7`	52.1 MB → 55.4 MB
⚠️ +6.0%	en_tdnn_15M _{load · pulse_120ms}	`cortex-a9`	2.84 s → 3.01 s
⚠️ +6.0%	en_tdnn_pyt_15M _{RSS @ ready · pulse_120ms}	`beaglev-ahead`	61.9 MB → 65.7 MB
⚠️ +6.0%	mdl_en_2019_Q3_librispeech_onnx _{RSS @ load · 2600ms}	`i9-11900kb_rtx-4060`	37.2 MB → 39.4 MB
⚠️ +5.8%	mobilenet_v2_1 _{RSS @ ready · pass_mt}	`orangepi-rv2`	42 MB → 44.4 MB
⚠️ +5.8%	en_tdnn_8M _{RSS @ ready · pulse_180ms}	`cortex-a9`	54.2 MB → 57.3 MB
⚠️ +5.7%	en_tdnn_8M _{RSS @ ready · pulse_120ms}	`cortex-a7`	54.4 MB → 57.5 MB
⚠️ +5.7%	openelm_270M_q40ef16_541 _{RSS @ ready · cpu}	`i9-9900k`	183 MB → 193 MB
⚠️ +5.6%	hey_snips_v31 _{RSS @ ready · 400ms}	`cortex-a53`	20.8 MB → 22 MB
⚠️ +5.6%	inceptionv3 _{RSS @ ready · pass_mt}	`orangepi-rv2`	120 MB → 127 MB
⚠️ +5.6%	inceptionv3 _{RSS @ ready · pass}	`orangepi-rv2`	120 MB → 126 MB
⚠️ +5.5%	inceptionv3 _{RSS @ load · pass_mt}	`orangepi-rv2`	115 MB → 121 MB
⚠️ +5.5%	nemotron_3_5_asr_streaming_0_6b_f32f32_joint_pass _{RSS @ ready · cpu}	`i9-9900k`	56.7 MB → 59.8 MB
⚠️ +5.5%	openelm_270M_q40ef16_541 _{RSS @ ready · cpu}	`i9-11900kb_rtx-4060`	193 MB → 204 MB
⚠️ +5.4%	en_tdnn_8M _{RSS @ ready · pulse_120ms}	`cortex-a9`	54.4 MB → 57.3 MB
⚠️ +5.4%	inceptionv3 _{RSS @ ready · pass_mt}	`beaglev-ahead`	120 MB → 126 MB
⚠️ +5.3%	parakeet_tdt_600m_v3_f32f32_joint_pass _{RSS @ ready · cpu}	`i9-11900kb_rtx-4060`	51.3 MB → 54 MB
⚠️ +5.2%	inceptionv3 _{RSS @ load · pass}	`orangepi-rv2`	115 MB → 121 MB
⚠️ +5.2%	en_tdnn_15M _{load · pulse_240ms}	`cortex-a9`	2.85 s → 3 s

Bench numbers come from shared CI hardware and do produce false positives: a flagged
row is not on its own evidence of a real regression, and a clean report is not proof
there is none. Contributors do not need to chase what is reported here — leave the
reading of it to the maintainers, who will say if something needs action.

github-actions · 2026-10-13T15:03:35Z

🔴 Bench vs main — 1 speed regression(s) · ⚠️ 129 secondary

Reference: 2026-10-13 morning nightly run (0d old) · full report → run

Speed — evaltime · prefill · decode

Δ	metric	device	main → PR
🔴 +14.5%	inceptionv3 _{evaltime · pass}	`orangepi-rv2`	2.17 s → 2.49 s

Improvements

Δ	metric	device	main → PR
🟢 -40.4%	mobilenet_v2_1 _{evaltime · pass_mt}	`orangepi-rv2`	252 ms → 150 ms
🟢 -38.7%	trunet _{evaltime · pulse1_f32}	`orangepi-rv2`	6.01 ms/pulse _{0.601 RTF} → 3.68 ms/pulse _{0.368 RTF}
🟢 -38.6%	parakeet_tdt_600m_v3_f32f32_preprocessor_1s _{evaltime · cpu}	`i9-9900k`	0.581 ms → 0.357 ms
🟢 -37.7%	nemotron_3_5_asr_streaming_0_6b_f32f32_preprocessor_1s _{evaltime · cpu}	`i9-11900kb_rtx-4060`	0.401 ms → 0.25 ms
🟢 -36.2%	nemotron_3_5_asr_streaming_0_6b_f32f32_preprocessor_1s _{evaltime · cpu}	`i9-9900k`	0.462 ms → 0.295 ms

+53 more improvement(s)

Δ	metric	device	main → PR
🟢 -35.4%	speaker_id _{evaltime · pulse8}	`orangepi-rv2`	2.24 ms/pulse _{0.028 RTF} → 1.44 ms/pulse _{0.0181 RTF}
🟢 -25.5%	speaker_id _{evaltime · pulse8}	`beaglev-ahead`	2.66 ms/pulse _{0.0333 RTF} → 1.98 ms/pulse _{0.0248 RTF}
🟢 -24.7%	mobilenet_v2_1 _{evaltime · pass_mt}	`beaglev-ahead`	328 ms → 247 ms
🟢 -22.5%	hey_snips_v4_model17_nnef _{evaltime · pulse8}	`beaglev-ahead`	5.19 ms/pulse _{0.0649 RTF} → 4.03 ms/pulse _{0.0503 RTF}
🟢 -22.1%	hey_snips_v4_model17 _{evaltime · 2sec}	`orangepi-rv2`	39.5 ms → 30.8 ms
🟢 -21.5%	en_tdnn_8M _{evaltime · pulse_240ms}	`orangepi-rv2`	41.9 ms/pulse _{0.175 RTF} → 32.9 ms/pulse _{0.137 RTF}
🟢 -20.4%	mobilenet_v2_1 _{evaltime · pass}	`orangepi-rv2`	556 ms → 442 ms
🟢 -20.0%	en_tdnn_8M _{evaltime · pulse_180ms}	`orangepi-rv2`	37.4 ms/pulse _{0.208 RTF} → 29.9 ms/pulse _{0.166 RTF}
🟢 -19.8%	hey_snips_v31 _{evaltime · 400ms}	`orangepi-rv2`	2.29 ms → 1.83 ms
🟢 -18.4%	en_tdnn_8M _{evaltime · pulse_120ms}	`orangepi-rv2`	29.2 ms/pulse _{0.244 RTF} → 23.8 ms/pulse _{0.199 RTF}
🟢 -17.8%	inceptionv3 _{evaltime · pass_mt}	`orangepi-rv2`	1.04 s → 852 ms
🟢 -17.8%	hey_snips_v4_model17 _{evaltime · 2sec}	`beaglev-ahead`	41.7 ms → 34.3 ms
🟢 -17.4%	trunet _{evaltime · pulse1_f32}	`beaglev-ahead`	5.32 ms/pulse _{0.532 RTF} → 4.4 ms/pulse _{0.44 RTF}
🟢 -16.7%	en_tdnn_15M _{evaltime · pulse_240ms}	`orangepi-rv2`	72 ms/pulse _{0.3 RTF} → 59.9 ms/pulse _{0.25 RTF}
🟢 -16.5%	mobilenet_v1_1 _{evaltime · pass}	`orangepi-rv2`	368 ms → 307 ms
🟢 -16.5%	hey_snips_v4_model17_nnef _{evaltime · pulse8}	`orangepi-rv2`	4.77 ms/pulse _{0.0596 RTF} → 3.98 ms/pulse _{0.0498 RTF}
🟢 -15.6%	en_tdnn_8M_nnef _{evaltime · pulse_240ms}	`orangepi-rv2`	39.6 ms/pulse _{0.165 RTF} → 33.4 ms/pulse _{0.139 RTF}
🟢 -14.9%	en_tdnn_15M _{evaltime · 2600ms}	`orangepi-rv2`	752 ms → 640 ms
🟢 -14.6%	voicecom_float _{evaltime · 2sec}	`orangepi-rv2`	60.2 ms → 51.4 ms
🟢 -14.1%	hey_snips_v4_model17 _{evaltime · pulse8}	`beaglev-ahead`	4.7 ms/pulse _{0.0587 RTF} → 4.03 ms/pulse _{0.0504 RTF}
🟢 -13.8%	en_tdnn_15M_nnef _{evaltime · pulse_240ms}	`orangepi-rv2`	71.4 ms/pulse _{0.297 RTF} → 61.5 ms/pulse _{0.256 RTF}
🟢 -13.5%	mobilenet_v2_1 _{evaltime · pass}	`beaglev-ahead`	855 ms → 739 ms
🟢 -13.3%	inceptionv3 _{evaltime · pass_mt}	`beaglev-ahead`	1.78 s → 1.54 s
🟢 -13.2%	voicecom_fake_quant _{evaltime · 2sec}	`orangepi-rv2`	65.3 ms → 56.7 ms
🟢 -13.1%	hey_snips_v4_model17 _{evaltime · pulse8}	`orangepi-rv2`	4.6 ms/pulse _{0.0575 RTF} → 4 ms/pulse _{0.05 RTF}
🟢 -12.8%	arm_ml_kws_cnn_m _{evaltime · pass}	`orangepi-rv2`	3.27 ms → 2.85 ms
🟢 -12.8%	en_tdnn_15M _{evaltime · pulse_120ms}	`orangepi-rv2`	50.7 ms/pulse _{0.423 RTF} → 44.3 ms/pulse _{0.369 RTF}
🟢 -12.3%	en_tdnn_8M _{evaltime · 2600ms}	`orangepi-rv2`	387 ms → 340 ms
🟢 -11.7%	en_tdnn_pyt_15M _{evaltime · pulse_120ms}	`orangepi-rv2`	37.7 ms/pulse _{0.314 RTF} → 33.3 ms/pulse _{0.278 RTF}
🟢 -10.9%	mobilenet_v2_1 _{evaltime · pass_mt}	`cortex-a9`	552 ms → 492 ms
🟢 -10.4%	trunet _{evaltime · pulse1_f16}	`orangepi-rv2`	7.15 ms/pulse _{0.715 RTF} → 6.41 ms/pulse _{0.641 RTF}
🟢 -10.3%	hey_snips_v4_model17 _{evaltime · pulse8}	`cortex-a9`	6.38 ms/pulse _{0.0797 RTF} → 5.72 ms/pulse _{0.0715 RTF}
🟢 -9.9%	hey_snips_v4_model17_nnef _{evaltime · pulse8}	`cortex-a9`	6.35 ms/pulse _{0.0794 RTF} → 5.72 ms/pulse _{0.0715 RTF}
🟢 -9.5%	hey_snips_v31 _{evaltime · 400ms}	`beaglev-ahead`	4.67 ms → 4.22 ms
🟢 -9.1%	mobilenet_v1_1 _{evaltime · pass}	`beaglev-ahead`	663 ms → 603 ms
🟢 -8.5%	hey_snips_v4_model17_nnef _{evaltime · pulse8}	`cortex-a7`	7.16 ms/pulse _{0.0895 RTF} → 6.55 ms/pulse _{0.0819 RTF}
🟢 -8.5%	mobilenet_v2_1 _{evaltime · pass_mt}	`cortex-a7`	534 ms → 489 ms
🟢 -8.4%	hey_snips_v4_model17 _{evaltime · pulse8}	`cortex-a55`	2.09 ms/pulse _{0.0261 RTF} → 1.91 ms/pulse _{0.0239 RTF}
🟢 -7.9%	hey_snips_v4_model17_nnef _{evaltime · pulse8}	`cortex-a55`	2.08 ms/pulse _{0.026 RTF} → 1.91 ms/pulse _{0.0239 RTF}
🟢 -7.8%	inceptionv1q _{evaltime · pass}	`orangepi-rv2`	654 ms → 603 ms
🟢 -7.6%	hey_snips_v4_model17 _{evaltime · pulse8}	`cortex-a7`	7.15 ms/pulse _{0.0893 RTF} → 6.6 ms/pulse _{0.0825 RTF}
🟢 -7.5%	voicecom_fake_quant _{evaltime · 2sec}	`i9-9900k`	5.38 ms → 4.98 ms
🟢 -7.2%	trunet _{evaltime · pulse1_f32}	`cortex-a9`	8.19 ms/pulse _{0.819 RTF} → 7.6 ms/pulse _{0.76 RTF}
🟢 -6.9%	en_tdnn_8M _{evaltime · pulse_180ms}	`beaglev-ahead`	72.6 ms/pulse _{0.403 RTF} → 67.6 ms/pulse _{0.376 RTF}
🟢 -6.9%	trunet _{evaltime · pulse1_f32}	`cortex-a7`	8.1 ms/pulse _{0.81 RTF} → 7.54 ms/pulse _{0.754 RTF}
🟢 -6.5%	en_tdnn_8M _{evaltime · pulse_120ms}	`beaglev-ahead`	40.7 ms/pulse _{0.339 RTF} → 38.1 ms/pulse _{0.317 RTF}
🟢 -6.5%	hey_snips_v4_model17_nnef _{evaltime · pulse8}	`cortex-a53`	2.72 ms/pulse _{0.034 RTF} → 2.54 ms/pulse _{0.0318 RTF}
🟢 -6.3%	en_tdnn_pyt_15M _{evaltime · pulse_120ms}	`beaglev-ahead`	60 ms/pulse _{0.5 RTF} → 56.2 ms/pulse _{0.468 RTF}
🟢 -5.8%	en_tdnn_8M _{evaltime · pulse_240ms}	`beaglev-ahead`	80.7 ms/pulse _{0.336 RTF} → 76 ms/pulse _{0.317 RTF}
🟢 -5.8%	hey_snips_v4_model17 _{evaltime · pulse8}	`cortex-a53`	2.71 ms/pulse _{0.0338 RTF} → 2.55 ms/pulse _{0.0319 RTF}
🟢 -5.7%	mobilenet_v2_1 _{evaltime · pass}	`cortex-a9`	1.4 s → 1.32 s
🟢 -5.1%	en_tdnn_15M _{evaltime · 2600ms}	`beaglev-ahead`	1.69 s → 1.6 s
🟢 -5.1%	en_tdnn_lstm_bn_q7 _{evaltime · 2600ms}	`cortex-a7`	371 ms → 352 ms

⚠️ 129 secondary regression(s)

Δ	metric	device	main → PR
⚠️ +33.6%	nemotron_3_5_asr_streaming_0_6b_f32f32_preprocessor_1s _{RSS @ load · cuda}	`jetson-orin-nx`	20.6 MB → 27.6 MB
⚠️ +33.4%	parakeet_tdt_600m_v3_f32f32_preprocessor_1s _{RSS @ load · cuda}	`jetson-orin-nx`	20.7 MB → 27.6 MB
⚠️ +28.9%	arm_ml_kws_cnn_m _{RSS @ load · pass}	`i9-9900k`	13.1 MB → 16.8 MB
⚠️ +28.3%	voicecom_float _{RSS @ load · 2sec}	`i9-9900k`	16.6 MB → 21.2 MB
⚠️ +27.8%	hey_snips_v1 _{RSS @ load · 400ms}	`i9-9900k`	13.2 MB → 16.9 MB
⚠️ +27.3%	nemotron_3_5_asr_streaming_0_6b_f32f32_preprocessor_1s _{RSS @ load · cpu}	`i9-9900k`	15.4 MB → 19.7 MB
⚠️ +26.5%	nemotron_3_5_asr_streaming_0_6b_f32f32_preprocessor_1s _{RSS @ ready · cpu}	`i9-9900k`	16.3 MB → 20.7 MB
⚠️ +26.0%	hey_snips_v31 _{RSS @ ready · 400ms}	`i9-9900k`	14.4 MB → 18.1 MB
⚠️ +26.0%	parakeet_tdt_600m_v3_f32f32_preprocessor_1s _{RSS @ load · cpu}	`i9-9900k`	15.6 MB → 19.6 MB
⚠️ +25.7%	hey_snips_v31 _{RSS @ load · 400ms}	`i9-9900k`	13.9 MB → 17.5 MB
⚠️ +25.1%	parakeet_tdt_600m_v3_f32f32_preprocessor_1s _{RSS @ ready · cpu}	`i9-9900k`	16.5 MB → 20.6 MB
⚠️ +24.4%	nemotron_3_5_asr_streaming_0_6b_f32f32_preprocessor_pulse100ms _{RSS @ load · cpu}	`i9-9900k`	16.7 MB → 20.8 MB
⚠️ +24.2%	nemotron_3_5_asr_streaming_0_6b_f32f32_preprocessor_pulse100ms _{RSS @ ready · cpu}	`i9-9900k`	17.1 MB → 21.3 MB
⚠️ +22.9%	voicecom_fake_quant _{RSS @ load · 2sec}	`i9-9900k`	17.8 MB → 21.9 MB
⚠️ +21.0%	speaker_id _{RSS @ ready · pulse8}	`i9-9900k`	17.2 MB → 20.8 MB
⚠️ +20.3%	speaker_id _{RSS @ load · pulse8}	`i9-9900k`	16.5 MB → 19.8 MB
⚠️ +19.7%	voicecom_float _{RSS @ load · 2sec}	`i9-11900kb_rtx-4060`	24.7 MB → 29.5 MB
⚠️ +18.5%	nemotron_3_5_asr_streaming_0_6b_f32f32_preprocessor_1s _{RSS @ load · cuda}	`i9-11900kb_rtx-4060`	24.2 MB → 28.6 MB
⚠️ +18.1%	arm_ml_kws_cnn_m _{RSS @ ready · pass}	`i9-9900k`	15.3 MB → 18.1 MB
⚠️ +17.7%	nemotron_3_5_asr_streaming_0_6b_f32f32_preprocessor_1s _{RSS @ load · cpu}	`i9-11900kb_rtx-4060`	24.2 MB → 28.5 MB
⚠️ +17.2%	hey_snips_v1 _{RSS @ load · 400ms}	`i9-11900kb_rtx-4060`	22 MB → 25.7 MB
⚠️ +17.0%	voicecom_fake_quant _{RSS @ load · 2sec}	`cortex-a9`	20.1 MB → 23.4 MB
⚠️ +16.9%	arm_ml_kws_cnn_m _{RSS @ load · pass}	`i9-11900kb_rtx-4060`	23.1 MB → 27.1 MB
⚠️ +16.7%	parakeet_tdt_600m_v3_f32f32_preprocessor_1s _{RSS @ load · cpu}	`i9-11900kb_rtx-4060`	24.5 MB → 28.6 MB
⚠️ +16.7%	parakeet_tdt_600m_v3_f32f32_preprocessor_1s _{RSS @ ready · cpu}	`i9-11900kb_rtx-4060`	26 MB → 30.3 MB
⚠️ +16.5%	nemotron_3_5_asr_streaming_0_6b_f32f32_preprocessor_1s _{RSS @ ready · cpu}	`i9-11900kb_rtx-4060`	25.9 MB → 30.1 MB
⚠️ +15.8%	nemotron_3_5_asr_streaming_0_6b_f32f32_preprocessor_pulse100ms _{RSS @ ready · cpu}	`i9-11900kb_rtx-4060`	27.4 MB → 31.7 MB
⚠️ +15.7%	parakeet_tdt_600m_v3_f32f32_preprocessor_1s _{RSS @ load · cuda}	`i9-11900kb_rtx-4060`	24.5 MB → 28.3 MB
⚠️ +15.7%	voicecom_fake_quant _{RSS @ load · 2sec}	`cortex-a7`	20 MB → 23.2 MB
⚠️ +15.5%	nemotron_3_5_asr_streaming_0_6b_f32f32_preprocessor_pulse100ms _{RSS @ load · cuda}	`i9-11900kb_rtx-4060`	26.4 MB → 30.5 MB
⚠️ +15.5%	voicecom_fake_quant _{RSS @ load · 2sec}	`i9-11900kb_rtx-4060`	26.8 MB → 30.9 MB
⚠️ +15.3%	nemotron_3_5_asr_streaming_0_6b_f32f32_preprocessor_pulse100ms _{RSS @ load · cpu}	`i9-11900kb_rtx-4060`	26.4 MB → 30.5 MB
⚠️ +14.9%	hey_snips_v31 _{RSS @ load · 400ms}	`i9-11900kb_rtx-4060`	23.9 MB → 27.4 MB
⚠️ +14.1%	hey_snips_v31 _{RSS @ ready · 400ms}	`i9-11900kb_rtx-4060`	25.5 MB → 29.1 MB
⚠️ +14.0%	voicecom_fake_quant _{RSS @ load · 2sec}	`cortex-a55`	22.2 MB → 25.4 MB
⚠️ +13.6%	en_tdnn_15M _{load · 2600ms}	`cortex-a55`	1.04 s → 1.19 s
⚠️ +13.6%	speaker_id _{RSS @ load · pulse8}	`cortex-a9`	18.4 MB → 20.8 MB
⚠️ +13.2%	en_tdnn_lstm_bn_q7 _{RSS @ load · 2600ms}	`i9-9900k`	19.1 MB → 21.6 MB
⚠️ +13.1%	parakeet_tdt_600m_v3_f32f32_decoder_pass _{RSS @ load · cpu}	`i9-11900kb_rtx-4060`	69.2 MB → 78.3 MB
⚠️ +13.0%	en_tdnn_lstm_bn_q7 _{RSS @ load · pulse_240ms}	`i9-9900k`	19.5 MB → 22.1 MB
⚠️ +12.8%	en_tdnn_15M _{load · 2600ms}	`cortex-a53`	1.43 s → 1.61 s
⚠️ +12.7%	hey_snips_v1 _{RSS @ ready · 400ms}	`i9-11900kb_rtx-4060`	24.3 MB → 27.4 MB
⚠️ +12.6%	speaker_id _{RSS @ ready · pulse8}	`i9-11900kb_rtx-4060`	27.7 MB → 31.2 MB
⚠️ +12.1%	trunet _{RSS @ ready · pulse1_f32}	`i9-9900k`	22 MB → 24.6 MB
⚠️ +12.1%	voicecom_fake_quant _{RSS @ load · 2sec}	`cortex-a53`	22.2 MB → 24.9 MB
⚠️ +12.0%	hey_snips_v4_model17_nnef _{RSS @ load · pulse8}	`i9-9900k`	16.3 MB → 18.3 MB
⚠️ +11.9%	parakeet_tdt_600m_v3_f32f32_decoder_pass _{RSS @ load · cpu}	`i9-9900k`	61.9 MB → 69.3 MB
⚠️ +11.8%	speaker_id _{RSS @ load · pulse8}	`cortex-a7`	18.5 MB → 20.7 MB
⚠️ +11.7%	speaker_id _{RSS @ load · pulse8}	`i9-11900kb_rtx-4060`	26.3 MB → 29.3 MB
⚠️ +11.4%	en_tdnn_15M _{load+optimize · 2600ms}	`cortex-a55`	1.25 s → 1.39 s
⚠️ +11.3%	trunet _{RSS @ load · pulse1_f32}	`i9-9900k`	20.7 MB → 23.1 MB
⚠️ +10.6%	en_tdnn_15M _{load+optimize · 2600ms}	`cortex-a53`	1.74 s → 1.92 s
⚠️ +10.5%	en_tdnn_15M _{load · pulse_240ms}	`cortex-a55`	1.24 s → 1.37 s
⚠️ +10.4%	en_tdnn_15M _{load · pulse_120ms}	`cortex-a53`	1.68 s → 1.86 s
⚠️ +10.2%	parakeet_tdt_600m_v3_f32f32_decoder_pass _{RSS @ load · cuda}	`i9-11900kb_rtx-4060`	69.1 MB → 76.1 MB
⚠️ +10.1%	en_tdnn_lstm_bn_q7 _{RSS @ ready · 2600ms}	`i9-9900k`	20.7 MB → 22.8 MB
⚠️ +10.1%	en_tdnn_15M _{load · pulse_240ms}	`cortex-a53`	1.68 s → 1.85 s
⚠️ +10.0%	mdl_en_2019_Q3_librispeech_onnx _{RSS @ load · 2600ms}	`i9-9900k`	27.4 MB → 30.1 MB
⚠️ +9.6%	en_tdnn_15M _{load+optimize · pulse_240ms}	`cortex-a55`	1.44 s → 1.58 s
⚠️ +9.4%	mdl_en_2019_Q3_librispeech_onnx _{RSS @ load · pulse_240ms}	`i9-9900k`	28.1 MB → 30.7 MB
⚠️ +9.1%	hey_snips_v1 _{RSS @ ready · 400ms}	`i9-9900k`	16.1 MB → 17.5 MB
⚠️ +9.1%	en_tdnn_8M _{RSS @ ready · pulse_120ms}	`orangepi-rv2`	50.9 MB → 55.5 MB
⚠️ +9.1%	hey_snips_v4_model17_nnef _{RSS @ ready · pulse8}	`i9-9900k`	18.1 MB → 19.7 MB
⚠️ +9.0%	en_tdnn_15M _{load · pulse_120ms}	`cortex-a55`	1.24 s → 1.35 s
⚠️ +9.0%	hey_snips_v4_model17_nnef _{RSS @ load · pulse8}	`i9-11900kb_rtx-4060`	24.4 MB → 26.6 MB
⚠️ +8.9%	en_tdnn_15M _{load+optimize · pulse_240ms}	`cortex-a53`	1.99 s → 2.17 s
⚠️ +8.8%	mobilenet_v1_1 _{RSS @ ready · pass}	`orangepi-rv2`	34.3 MB → 37.4 MB
⚠️ +8.8%	arm_ml_kws_cnn_m _{RSS @ ready · pass}	`i9-11900kb_rtx-4060`	26.5 MB → 28.9 MB
⚠️ +8.7%	hey_snips_v31 _{RSS @ ready · 400ms}	`cortex-a53`	20.8 MB → 22.6 MB
⚠️ +8.7%	mobilenet_v1_1 _{RSS @ ready · pass}	`beaglev-ahead`	34.3 MB → 37.3 MB
⚠️ +8.7%	en_tdnn_15M _{load+optimize · pulse_120ms}	`cortex-a53`	2 s → 2.17 s
⚠️ +8.6%	en_tdnn_8M_nnef _{RSS @ ready · pulse_240ms}	`beaglev-ahead`	48.2 MB → 52.3 MB
⚠️ +8.6%	en_tdnn_lstm_bn_q7 _{RSS @ load · 2600ms}	`i9-11900kb_rtx-4060`	28.3 MB → 30.7 MB
⚠️ +8.5%	nemotron_3_5_asr_streaming_0_6b_f32f32_decoder_pass _{RSS @ load · cpu}	`i9-9900k`	74.6 MB → 80.9 MB
⚠️ +8.4%	parakeet_tdt_600m_v3_f32f32_joint_pass _{RSS @ ready · cpu}	`i9-11900kb_rtx-4060`	51.3 MB → 55.6 MB
⚠️ +8.4%	en_tdnn_8M_nnef _{RSS @ ready · pulse_240ms}	`i9-9900k`	50.6 MB → 54.8 MB
⚠️ +8.3%	en_tdnn_8M _{RSS @ ready · pulse_240ms}	`orangepi-rv2`	50.9 MB → 55.1 MB
⚠️ +8.2%	en_tdnn_8M_nnef _{RSS @ load · pulse_240ms}	`i9-9900k`	46.8 MB → 50.6 MB
⚠️ +8.2%	en_tdnn_15M _{load · 2600ms}	`cortex-a9`	2.49 s → 2.69 s
⚠️ +7.9%	en_tdnn_8M_nnef _{RSS @ ready · pulse_240ms}	`i9-11900kb_rtx-4060`	59.7 MB → 64.4 MB
⚠️ +7.9%	en_tdnn_lstm_bn_q7 _{RSS @ ready · pulse_240ms}	`i9-9900k`	21.5 MB → 23.2 MB
⚠️ +7.9%	nemotron_3_5_asr_streaming_0_6b_f32f32_decoder_pass _{RSS @ load · cuda}	`i9-11900kb_rtx-4060`	81.7 MB → 88.1 MB
⚠️ +7.8%	en_tdnn_8M _{RSS @ ready · pulse_180ms}	`orangepi-rv2`	50.9 MB → 54.9 MB
⚠️ +7.8%	mobilenet_v1_1 _{RSS @ ready · pass}	`cortex-a7`	37.6 MB → 40.6 MB
⚠️ +7.8%	nemotron_3_5_asr_streaming_0_6b_f32f32_joint_pass _{RSS @ ready · cpu}	`i9-9900k`	56.7 MB → 61.1 MB
⚠️ +7.7%	en_tdnn_8M_nnef _{RSS @ ready · pulse_240ms}	`orangepi-rv2`	48 MB → 51.7 MB
⚠️ +7.6%	nemotron_3_5_asr_streaming_0_6b_f32f32_decoder_pass _{RSS @ load · cpu}	`i9-11900kb_rtx-4060`	81.9 MB → 88.2 MB
⚠️ +7.6%	en_tdnn_8M _{RSS @ ready · pulse_240ms}	`beaglev-ahead`	51.1 MB → 55 MB
⚠️ +7.6%	en_tdnn_8M _{RSS @ ready · pulse_120ms}	`beaglev-ahead`	51.1 MB → 55 MB
⚠️ +7.6%	en_tdnn_8M _{RSS @ ready · pulse_180ms}	`beaglev-ahead`	51.1 MB → 55 MB
⚠️ +7.4%	en_tdnn_8M_nnef _{RSS @ load · pulse_240ms}	`i9-11900kb_rtx-4060`	54.9 MB → 59 MB
⚠️ +7.2%	parakeet_tdt_600m_v3_f32f32_decoder_pass _{RSS @ load · cuda}	`jetson-orin-nx`	65.8 MB → 70.6 MB
⚠️ +7.1%	en_tdnn_8M_nnef _{RSS @ ready · pulse_240ms}	`cortex-a9`	51.9 MB → 55.6 MB
⚠️ +7.1%	nemotron_3_5_asr_streaming_0_6b_f32f32_joint_pass _{RSS @ ready · cpu}	`i9-11900kb_rtx-4060`	64.1 MB → 68.7 MB
⚠️ +7.1%	en_tdnn_8M _{RSS @ ready · 2600ms}	`orangepi-rv2`	50.1 MB → 53.6 MB
⚠️ +7.0%	mobilenet_v2_1 _{RSS @ load · pass_mt}	`orangepi-rv2`	39.9 MB → 42.7 MB
⚠️ +6.8%	en_tdnn_8M _{RSS @ ready · 2600ms}	`beaglev-ahead`	50.2 MB → 53.7 MB
⚠️ +6.8%	mobilenet_v2_1 _{RSS @ load · pass_mt}	`beaglev-ahead`	40 MB → 42.7 MB
⚠️ +6.7%	en_tdnn_pyt_15M _{RSS @ ready · pulse_120ms}	`orangepi-rv2`	61.8 MB → 65.9 MB
⚠️ +6.4%	mobilenet_v2_1 _{RSS @ load · pass}	`orangepi-rv2`	39.9 MB → 42.5 MB
⚠️ +6.3%	hey_snips_v1 _{load+optimize · 400ms}	`cortex-a53`	63 ms → 67 ms
⚠️ +6.2%	hey_snips_v31 _{RSS @ ready · 400ms}	`cortex-a55`	21.1 MB → 22.4 MB
⚠️ +6.2%	en_tdnn_8M _{RSS @ ready · pulse_120ms}	`cortex-a7`	54.4 MB → 57.8 MB
⚠️ +6.2%	mobilenet_v2_1 _{RSS @ ready · pass_mt}	`orangepi-rv2`	42 MB → 44.6 MB
⚠️ +6.1%	openelm_270M_q40ef16_541 _{RSS @ ready · cpu}	`i9-11900kb_rtx-4060`	193 MB → 205 MB
⚠️ +6.1%	openelm_270M_q40ef16_541 _{RSS @ ready · cpu}	`i9-9900k`	183 MB → 194 MB
⚠️ +6.1%	mobilenet_v2_1 _{RSS @ ready · pass}	`orangepi-rv2`	42 MB → 44.5 MB
⚠️ +6.0%	mdl_en_2019_Q3_librispeech_onnx _{RSS @ load · 2600ms}	`i9-11900kb_rtx-4060`	37.2 MB → 39.4 MB
⚠️ +6.0%	en_tdnn_pyt_15M _{RSS @ ready · pulse_120ms}	`beaglev-ahead`	61.9 MB → 65.6 MB
⚠️ +6.0%	en_tdnn_8M _{RSS @ ready · 2600ms}	`cortex-a7`	53.1 MB → 56.2 MB
⚠️ +5.8%	en_tdnn_15M _{load · pulse_120ms}	`cortex-a9`	2.84 s → 3.01 s
⚠️ +5.8%	parakeet_tdt_600m_v3_f32f32_joint_pass _{RSS @ ready · cpu}	`i9-9900k`	44.1 MB → 46.6 MB
⚠️ +5.8%	mobilenet_v2_1 _{RSS @ ready · pass_mt}	`beaglev-ahead`	42 MB → 44.4 MB
⚠️ +5.8%	inceptionv3 _{RSS @ ready · pass}	`orangepi-rv2`	120 MB → 127 MB
⚠️ +5.8%	mobilenet_v2_1 _{RSS @ ready · pass}	`beaglev-ahead`	42 MB → 44.4 MB
⚠️ +5.7%	en_tdnn_8M _{RSS @ ready · pulse_240ms}	`cortex-a7`	54.3 MB → 57.3 MB
⚠️ +5.6%	en_tdnn_8M_nnef _{RSS @ ready · pulse_240ms}	`cortex-a7`	52.1 MB → 55.1 MB
⚠️ +5.6%	en_tdnn_8M _{RSS @ ready · pulse_180ms}	`cortex-a7`	54.2 MB → 57.2 MB
⚠️ +5.6%	en_tdnn_15M_nnef _{RSS @ ready · pulse_240ms}	`i9-9900k`	87.3 MB → 92.2 MB
⚠️ +5.5%	inceptionv3 _{RSS @ ready · pass_mt}	`orangepi-rv2`	120 MB → 126 MB
⚠️ +5.5%	inceptionv3 _{RSS @ ready · pass}	`beaglev-ahead`	120 MB → 127 MB
⚠️ +5.5%	inceptionv3 _{RSS @ load · pass}	`orangepi-rv2`	115 MB → 121 MB
⚠️ +5.4%	en_tdnn_15M _{load · pulse_240ms}	`cortex-a9`	2.85 s → 3.01 s
⚠️ +5.4%	inceptionv3 _{RSS @ ready · pass_mt}	`beaglev-ahead`	120 MB → 126 MB
⚠️ +5.3%	en_tdnn_8M _{RSS @ ready · pulse_180ms}	`cortex-a9`	54.2 MB → 57.1 MB
⚠️ +5.3%	en_tdnn_pyt_15M _{RSS @ load · pulse_120ms}	`orangepi-rv2`	60.7 MB → 63.9 MB
⚠️ +5.2%	inceptionv3 _{RSS @ load · pass}	`beaglev-ahead`	115 MB → 121 MB
⚠️ +5.2%	inceptionv3 _{RSS @ load · pass_mt}	`orangepi-rv2`	115 MB → 121 MB
⚠️ +5.1%	inceptionv3 _{RSS @ load · pass_mt}	`beaglev-ahead`	115 MB → 121 MB

Bench numbers come from shared CI hardware and do produce false positives: a flagged
row is not on its own evidence of a real regression, and a clean report is not proof
there is none. Contributors do not need to chase what is reported here — leave the
reading of it to the maintainers, who will say if something needs action.

…nch suite runs, and a process can settle into a regime where a per-eval intermediate is mapped and unmapped on every pass, costing millions of minor faults. Use jemalloc as the CLI global allocator, wrapped in the same readings instrumentation, behind a jemalloc feature that bench-suite turns on.

…ch has no count-trailing-zeros instruction outside Zbb, and rust's compiler_builtins ships the rest of the *di2 family but not that one, so the static musl link of the riscv64 CLI left it undefined. Name libgcc explicitly in that target's rustflags.

…urged extent keeps its mapping and its dirty pages, and on some kernels those pages never come back: the CLI reported a 338 MB resident set for a graph holding 68 MB of live tensors, where the same binary on an older kernel of the same SoC stayed at 115 MB. Set retain:false in the CLI's malloc_conf, which is what jemalloc itself defaults to on 32-bit targets, so the resident set the bench suite records tracks the live one.

…plan allocates every op's output on every eval -- but tract is a library and cannot pick it. Write down what to link and how to configure it: jemalloc for the pulsed graphs whose per-turn cost is bookkeeping rather than matmul, with retain:false so a long-running process gives its purged pages back.

…thology jemalloc was brought in to work around, and the bench run made that concrete: apple-m1-max produced the only four speed regressions in the whole suite (mobilenet_v1 +9.0%, inceptionv3 +8.1% and +7.7% on two configs, llama_3_2_1B prefill +5.3%) while every improvement came from the linux boards. Exclude apple targets from the jemalloc feature so they keep the system allocator.

jemalloc keeps freed pages mapped for its dirty decay period, and loading a model frees most of what it allocates within that period, so the RSS the bench suite records at load and at model-ready reported allocator slack — up to twice the musl figure on small models, and varying run to run. Purge every arena right before those two readings, so they report what the process holds; the jemalloc setup moves into an allocator module that also carries the no-op purge for the system allocator.

kali · 2026-10-13T17:58:11Z

⚠️⚠️⚠️ Just rebased! ⚠️⚠️⚠️

github-actions · 2026-10-13T18:23:40Z

⚠️ Bench vs main — no speed regressions · 91 secondary regression(s)

Reference: 2026-10-13 morning nightly run (0d old) · full report → run

Speed — evaltime · prefill · decode

no inference-speed regressions

Improvements

Δ	metric	device	main → PR
🟢 -39.5%	mobilenet_v2_1 _{evaltime · pass_mt}	`orangepi-rv2`	252 ms → 153 ms
🟢 -39.5%	parakeet_tdt_600m_v3_f32f32_preprocessor_1s _{evaltime · cpu}	`i9-9900k`	0.581 ms → 0.352 ms
🟢 -38.3%	trunet _{evaltime · pulse1_f32}	`orangepi-rv2`	6.01 ms/pulse _{0.601 RTF} → 3.71 ms/pulse _{0.371 RTF}
🟢 -37.3%	nemotron_3_5_asr_streaming_0_6b_f32f32_preprocessor_1s _{evaltime · cpu}	`i9-9900k`	0.462 ms → 0.29 ms
🟢 -34.9%	nemotron_3_5_asr_streaming_0_6b_f32f32_preprocessor_1s _{evaltime · cpu}	`i9-11900kb_rtx-4060`	0.401 ms → 0.261 ms

+51 more improvement(s)

Δ	metric	device	main → PR
🟢 -33.8%	speaker_id _{evaltime · pulse8}	`orangepi-rv2`	2.24 ms/pulse _{0.028 RTF} → 1.48 ms/pulse _{0.0185 RTF}
🟢 -26.5%	speaker_id _{evaltime · pulse8}	`beaglev-ahead`	2.66 ms/pulse _{0.0333 RTF} → 1.95 ms/pulse _{0.0244 RTF}
🟢 -25.4%	mobilenet_v2_1 _{evaltime · pass_mt}	`beaglev-ahead`	328 ms → 244 ms
🟢 -21.6%	hey_snips_v4_model17 _{evaltime · 2sec}	`orangepi-rv2`	39.5 ms → 31 ms
🟢 -21.3%	en_tdnn_8M _{evaltime · pulse_240ms}	`orangepi-rv2`	41.9 ms/pulse _{0.175 RTF} → 33 ms/pulse _{0.137 RTF}
🟢 -21.2%	hey_snips_v4_model17_nnef _{evaltime · pulse8}	`beaglev-ahead`	5.19 ms/pulse _{0.0649 RTF} → 4.09 ms/pulse _{0.0511 RTF}
🟢 -20.1%	en_tdnn_8M _{evaltime · pulse_180ms}	`orangepi-rv2`	37.4 ms/pulse _{0.208 RTF} → 29.9 ms/pulse _{0.166 RTF}
🟢 -18.3%	trunet _{evaltime · pulse1_f32}	`beaglev-ahead`	5.32 ms/pulse _{0.532 RTF} → 4.35 ms/pulse _{0.435 RTF}
🟢 -18.2%	inceptionv3 _{evaltime · pass_mt}	`orangepi-rv2`	1.04 s → 848 ms
🟢 -18.1%	hey_snips_v31 _{evaltime · 400ms}	`orangepi-rv2`	2.29 ms → 1.87 ms
🟢 -17.9%	hey_snips_v4_model17 _{evaltime · 2sec}	`beaglev-ahead`	41.7 ms → 34.3 ms
🟢 -16.8%	mobilenet_v2_1 _{evaltime · pass}	`orangepi-rv2`	556 ms → 462 ms
🟢 -16.3%	mobilenet_v1_1 _{evaltime · pass}	`orangepi-rv2`	368 ms → 308 ms
🟢 -16.1%	inceptionv3 _{evaltime · pass}	`orangepi-rv2`	2.17 s → 1.82 s
🟢 -16.1%	en_tdnn_8M _{evaltime · pulse_120ms}	`orangepi-rv2`	29.2 ms/pulse _{0.244 RTF} → 24.5 ms/pulse _{0.205 RTF}
🟢 -15.9%	hey_snips_v4_model17_nnef _{evaltime · pulse8}	`orangepi-rv2`	4.77 ms/pulse _{0.0596 RTF} → 4.01 ms/pulse _{0.0501 RTF}
🟢 -15.8%	en_tdnn_15M _{evaltime · pulse_240ms}	`orangepi-rv2`	72 ms/pulse _{0.3 RTF} → 60.6 ms/pulse _{0.253 RTF}
🟢 -15.7%	en_tdnn_8M_nnef _{evaltime · pulse_240ms}	`orangepi-rv2`	39.6 ms/pulse _{0.165 RTF} → 33.4 ms/pulse _{0.139 RTF}
🟢 -15.4%	en_tdnn_15M_nnef _{evaltime · pulse_240ms}	`orangepi-rv2`	71.4 ms/pulse _{0.297 RTF} → 60.4 ms/pulse _{0.252 RTF}
🟢 -14.4%	en_tdnn_pyt_15M _{evaltime · pulse_120ms}	`orangepi-rv2`	37.7 ms/pulse _{0.314 RTF} → 32.3 ms/pulse _{0.269 RTF}
🟢 -13.3%	voicecom_float _{evaltime · 2sec}	`orangepi-rv2`	60.2 ms → 52.2 ms
🟢 -13.1%	arm_ml_kws_cnn_m _{evaltime · pass}	`orangepi-rv2`	3.27 ms → 2.84 ms
🟢 -13.0%	hey_snips_v4_model17 _{evaltime · pulse8}	`beaglev-ahead`	4.7 ms/pulse _{0.0587 RTF} → 4.09 ms/pulse _{0.0511 RTF}
🟢 -12.7%	voicecom_fake_quant _{evaltime · 2sec}	`orangepi-rv2`	65.3 ms → 57 ms
🟢 -12.5%	hey_snips_v4_model17 _{evaltime · pulse8}	`orangepi-rv2`	4.6 ms/pulse _{0.0575 RTF} → 4.02 ms/pulse _{0.0503 RTF}
🟢 -12.3%	en_tdnn_15M _{evaltime · 2600ms}	`orangepi-rv2`	752 ms → 659 ms
🟢 -11.8%	inceptionv3 _{evaltime · pass_mt}	`beaglev-ahead`	1.78 s → 1.57 s
🟢 -10.4%	mobilenet_v2_1 _{evaltime · pass_mt}	`cortex-a9`	552 ms → 494 ms
🟢 -10.3%	hey_snips_v4_model17 _{evaltime · pulse8}	`cortex-a7`	7.15 ms/pulse _{0.0893 RTF} → 6.41 ms/pulse _{0.0801 RTF}
🟢 -10.3%	trunet _{evaltime · pulse1_f16}	`orangepi-rv2`	7.15 ms/pulse _{0.715 RTF} → 6.41 ms/pulse _{0.641 RTF}
🟢 -9.8%	en_tdnn_8M _{evaltime · 2600ms}	`orangepi-rv2`	387 ms → 349 ms
🟢 -9.6%	hey_snips_v4_model17 _{evaltime · pulse8}	`cortex-a9`	6.38 ms/pulse _{0.0797 RTF} → 5.76 ms/pulse _{0.0721 RTF}
🟢 -9.4%	hey_snips_v4_model17_nnef _{evaltime · pulse8}	`cortex-a9`	6.35 ms/pulse _{0.0794 RTF} → 5.76 ms/pulse _{0.0719 RTF}
🟢 -9.3%	hey_snips_v4_model17_nnef _{evaltime · pulse8}	`cortex-a7`	7.16 ms/pulse _{0.0895 RTF} → 6.49 ms/pulse _{0.0811 RTF}
🟢 -9.2%	en_tdnn_15M _{evaltime · pulse_120ms}	`orangepi-rv2`	50.7 ms/pulse _{0.423 RTF} → 46.1 ms/pulse _{0.384 RTF}
🟢 -9.2%	mobilenet_v2_1 _{evaltime · pass}	`beaglev-ahead`	855 ms → 776 ms
🟢 -9.1%	inceptionv1q _{evaltime · pass}	`orangepi-rv2`	654 ms → 595 ms
🟢 -8.9%	hey_snips_v31 _{evaltime · 400ms}	`beaglev-ahead`	4.67 ms → 4.25 ms
🟢 -8.2%	mobilenet_v2_1 _{evaltime · pass_mt}	`cortex-a7`	534 ms → 491 ms
🟢 -8.1%	mobilenet_v1_1 _{evaltime · pass}	`beaglev-ahead`	663 ms → 610 ms
🟢 -7.0%	en_tdnn_8M _{evaltime · pulse_240ms}	`beaglev-ahead`	80.7 ms/pulse _{0.336 RTF} → 75.1 ms/pulse _{0.313 RTF}
🟢 -6.9%	en_tdnn_8M _{evaltime · pulse_120ms}	`beaglev-ahead`	40.7 ms/pulse _{0.339 RTF} → 37.9 ms/pulse _{0.316 RTF}
🟢 -6.8%	trunet _{evaltime · pulse1_f32}	`cortex-a9`	8.19 ms/pulse _{0.819 RTF} → 7.63 ms/pulse _{0.763 RTF}
🟢 -6.5%	hey_snips_v4_model17 _{evaltime · pulse8}	`cortex-a53`	2.71 ms/pulse _{0.0338 RTF} → 2.53 ms/pulse _{0.0316 RTF}
🟢 -6.4%	en_tdnn_pyt_15M _{evaltime · pulse_120ms}	`beaglev-ahead`	60 ms/pulse _{0.5 RTF} → 56.1 ms/pulse _{0.468 RTF}
🟢 -6.4%	en_tdnn_8M _{evaltime · pulse_180ms}	`beaglev-ahead`	72.6 ms/pulse _{0.403 RTF} → 68 ms/pulse _{0.378 RTF}
🟢 -6.2%	hey_snips_v4_model17 _{evaltime · pulse8}	`cortex-a55`	2.09 ms/pulse _{0.0261 RTF} → 1.96 ms/pulse _{0.0245 RTF}
🟢 -6.2%	hey_snips_v4_model17_nnef _{evaltime · pulse8}	`cortex-a53`	2.72 ms/pulse _{0.034 RTF} → 2.55 ms/pulse _{0.0319 RTF}
🟢 -5.9%	en_tdnn_lstm_bn_q7 _{evaltime · 2600ms}	`cortex-a7`	371 ms → 349 ms
🟢 -5.6%	trunet _{evaltime · pulse1_f32}	`cortex-a7`	8.1 ms/pulse _{0.81 RTF} → 7.64 ms/pulse _{0.764 RTF}
🟢 -5.2%	speaker_id _{evaltime · pulse8}	`cortex-a9`	2.5 ms/pulse _{0.0312 RTF} → 2.37 ms/pulse _{0.0296 RTF}

⚠️ 91 secondary regression(s)

Δ	metric	device	main → PR
⚠️ +32.4%	nemotron_3_5_asr_streaming_0_6b_f32f32_preprocessor_1s _{RSS @ load · cuda}	`jetson-orin-nx`	20.6 MB → 27.3 MB
⚠️ +25.4%	nemotron_3_5_asr_streaming_0_6b_f32f32_encoder_1s _{load · cpu}	`apple-m1-max`	3.44 s → 4.31 s
⚠️ +25.2%	nemotron_3_5_asr_streaming_0_6b_f32f32_encoder_1s _{load · metal}	`apple-m1-max`	3.44 s → 4.31 s
⚠️ +24.8%	nemotron_3_5_asr_streaming_0_6b_f32f32_encoder_1s _{load · cpu_mt}	`apple-m1-max`	3.45 s → 4.31 s
⚠️ +23.7%	nemotron_3_5_asr_streaming_0_6b_f32f32_encoder_1s _{load+optimize · metal}	`apple-m1-max`	3.63 s → 4.49 s
⚠️ +22.6%	nemotron_3_5_asr_streaming_0_6b_f32f32_encoder_1s _{load+optimize · cpu}	`apple-m1-max`	3.87 s → 4.74 s
⚠️ +22.3%	parakeet_tdt_600m_v3_f32f32_preprocessor_1s _{RSS @ load · cuda}	`jetson-orin-nx`	20.7 MB → 25.3 MB
⚠️ +21.8%	nemotron_3_5_asr_streaming_0_6b_f32f32_encoder_1s _{load+optimize · cpu_mt}	`apple-m1-max`	3.89 s → 4.74 s
⚠️ +16.6%	voicecom_float _{RSS @ load · 2sec}	`i9-9900k`	16.6 MB → 19.3 MB
⚠️ +16.3%	voicecom_fake_quant _{RSS @ load · 2sec}	`cortex-a9`	20.1 MB → 23.3 MB
⚠️ +14.8%	voicecom_fake_quant _{RSS @ load · 2sec}	`cortex-a7`	20 MB → 23 MB
⚠️ +14.4%	speaker_id _{RSS @ load · pulse8}	`cortex-a9`	18.4 MB → 21 MB
⚠️ +13.9%	speaker_id _{RSS @ load · pulse8}	`cortex-a7`	18.5 MB → 21.1 MB
⚠️ +13.2%	arm_ml_kws_cnn_m _{RSS @ load · pass}	`i9-9900k`	13.1 MB → 14.8 MB
⚠️ +12.7%	en_tdnn_15M _{load · 2600ms}	`cortex-a53`	1.43 s → 1.61 s
⚠️ +12.6%	voicecom_float _{RSS @ load · 2sec}	`i9-11900kb_rtx-4060`	24.7 MB → 27.8 MB
⚠️ +12.3%	nemotron_3_5_asr_streaming_0_6b_f32f32_preprocessor_1s _{RSS @ load · cpu}	`i9-9900k`	15.4 MB → 17.3 MB
⚠️ +12.3%	en_tdnn_15M _{load · 2600ms}	`cortex-a55`	1.04 s → 1.17 s
⚠️ +11.9%	nemotron_3_5_asr_streaming_0_6b_f32f32_preprocessor_1s _{RSS @ ready · cpu}	`i9-9900k`	16.3 MB → 18.3 MB
⚠️ +11.5%	voicecom_fake_quant _{RSS @ load · 2sec}	`i9-9900k`	17.8 MB → 19.8 MB
⚠️ +11.2%	voicecom_fake_quant _{RSS @ load · 2sec}	`cortex-a53`	22.2 MB → 24.7 MB
⚠️ +11.1%	parakeet_tdt_600m_v3_f32f32_preprocessor_1s _{RSS @ load · cpu}	`i9-9900k`	15.6 MB → 17.3 MB
⚠️ +11.0%	en_tdnn_15M _{load · pulse_120ms}	`cortex-a55`	1.24 s → 1.37 s
⚠️ +10.8%	voicecom_fake_quant _{RSS @ load · 2sec}	`cortex-a55`	22.2 MB → 24.7 MB
⚠️ +10.8%	parakeet_tdt_600m_v3_f32f32_decoder_pass _{RSS @ load · cuda}	`i9-11900kb_rtx-4060`	69.1 MB → 76.5 MB
⚠️ +10.7%	parakeet_tdt_600m_v3_f32f32_preprocessor_1s _{RSS @ ready · cpu}	`i9-9900k`	16.5 MB → 18.3 MB
⚠️ +10.6%	en_tdnn_15M _{load · pulse_120ms}	`cortex-a53`	1.68 s → 1.86 s
⚠️ +10.4%	en_tdnn_15M _{load+optimize · 2600ms}	`cortex-a55`	1.25 s → 1.38 s
⚠️ +10.4%	en_tdnn_15M _{load+optimize · 2600ms}	`cortex-a53`	1.74 s → 1.92 s
⚠️ +10.0%	nemotron_3_5_asr_streaming_0_6b_f32f32_preprocessor_pulse100ms _{RSS @ ready · cpu}	`i9-9900k`	17.1 MB → 18.9 MB
⚠️ +9.8%	en_tdnn_15M _{load · pulse_240ms}	`cortex-a53`	1.68 s → 1.84 s
⚠️ +9.4%	hey_snips_v31 _{RSS @ ready · 400ms}	`i9-9900k`	14.4 MB → 15.7 MB
⚠️ +9.3%	nemotron_3_5_asr_streaming_0_6b_f32f32_preprocessor_pulse100ms _{RSS @ load · cpu}	`i9-9900k`	16.7 MB → 18.2 MB
⚠️ +9.2%	parakeet_tdt_600m_v3_f32f32_decoder_pass _{RSS @ load · cuda}	`jetson-orin-nx`	65.8 MB → 71.8 MB
⚠️ +9.1%	en_tdnn_15M _{load+optimize · pulse_120ms}	`cortex-a53`	2 s → 2.18 s
⚠️ +9.1%	mobilenet_v1_1 _{RSS @ ready · pass}	`orangepi-rv2`	34.3 MB → 37.5 MB
⚠️ +9.1%	en_tdnn_15M _{load+optimize · pulse_120ms}	`cortex-a55`	1.45 s → 1.58 s
⚠️ +8.8%	speaker_id _{RSS @ ready · pulse8}	`i9-9900k`	17.2 MB → 18.7 MB
⚠️ +8.7%	mobilenet_v1_1 _{RSS @ ready · pass}	`beaglev-ahead`	34.3 MB → 37.3 MB
⚠️ +8.7%	arm_ml_kws_cnn_m _{RSS @ load · pass}	`i9-11900kb_rtx-4060`	23.1 MB → 25.2 MB
⚠️ +8.6%	hey_snips_v31 _{RSS @ load · 400ms}	`i9-9900k`	13.9 MB → 15.1 MB
⚠️ +8.5%	hey_snips_v31 _{RSS @ ready · 400ms}	`cortex-a53`	20.8 MB → 22.6 MB
⚠️ +8.2%	en_tdnn_15M _{load+optimize · pulse_240ms}	`cortex-a53`	1.99 s → 2.15 s
⚠️ +8.2%	parakeet_tdt_600m_v3_f32f32_decoder_pass _{RSS @ load · cpu}	`i9-9900k`	61.9 MB → 67 MB
⚠️ +8.2%	en_tdnn_8M_nnef _{RSS @ ready · pulse_240ms}	`orangepi-rv2`	48 MB → 51.9 MB
⚠️ +8.0%	en_tdnn_8M _{RSS @ ready · pulse_180ms}	`beaglev-ahead`	51.1 MB → 55.2 MB
⚠️ +8.0%	mobilenet_v1_1 _{RSS @ ready · pass}	`cortex-a9`	37.6 MB → 40.6 MB
⚠️ +8.0%	en_tdnn_8M _{RSS @ ready · pulse_240ms}	`orangepi-rv2`	50.9 MB → 55 MB
⚠️ +7.9%	en_tdnn_8M _{RSS @ ready · pulse_120ms}	`orangepi-rv2`	50.9 MB → 54.9 MB
⚠️ +7.8%	mobilenet_v1_1 _{RSS @ ready · pass}	`cortex-a7`	37.6 MB → 40.6 MB
⚠️ +7.4%	speaker_id _{RSS @ load · pulse8}	`i9-9900k`	16.5 MB → 17.7 MB
⚠️ +7.3%	parakeet_tdt_600m_v3_f32f32_decoder_pass _{RSS @ load · cpu}	`i9-11900kb_rtx-4060`	69.2 MB → 74.3 MB
⚠️ +7.3%	en_tdnn_8M _{RSS @ ready · pulse_180ms}	`orangepi-rv2`	50.9 MB → 54.6 MB
⚠️ +7.2%	en_tdnn_8M _{RSS @ ready · 2600ms}	`orangepi-rv2`	50.1 MB → 53.7 MB
⚠️ +7.0%	en_tdnn_8M _{RSS @ ready · pulse_120ms}	`beaglev-ahead`	51.1 MB → 54.7 MB
⚠️ +7.0%	en_tdnn_8M _{RSS @ ready · 2600ms}	`beaglev-ahead`	50.2 MB → 53.7 MB
⚠️ +6.8%	en_tdnn_8M_nnef _{RSS @ ready · pulse_240ms}	`beaglev-ahead`	48.2 MB → 51.4 MB
⚠️ +6.8%	mobilenet_v2_1 _{RSS @ load · pass}	`orangepi-rv2`	39.9 MB → 42.6 MB
⚠️ +6.8%	en_tdnn_15M _{load · 2600ms}	`cortex-a9`	2.49 s → 2.66 s
⚠️ +6.7%	en_tdnn_8M _{RSS @ ready · pulse_240ms}	`beaglev-ahead`	51.1 MB → 54.5 MB
⚠️ +6.7%	hey_snips_v31 _{RSS @ ready · 400ms}	`cortex-a55`	21.1 MB → 22.5 MB
⚠️ +6.6%	mobilenet_v2_1 _{RSS @ load · pass_mt}	`beaglev-ahead`	40 MB → 42.7 MB
⚠️ +6.6%	mobilenet_v2_1 _{RSS @ load · pass}	`beaglev-ahead`	40 MB → 42.7 MB
⚠️ +6.5%	mobilenet_v2_1 _{RSS @ load · pass_mt}	`orangepi-rv2`	39.9 MB → 42.5 MB
⚠️ +6.3%	hey_snips_v1 _{load+optimize · 400ms}	`cortex-a53`	63 ms → 67 ms
⚠️ +6.3%	en_tdnn_8M _{RSS @ ready · pulse_180ms}	`cortex-a9`	54.2 MB → 57.6 MB
⚠️ +6.2%	en_tdnn_pyt_15M _{RSS @ ready · pulse_120ms}	`orangepi-rv2`	61.8 MB → 65.7 MB
⚠️ +6.2%	en_tdnn_8M _{RSS @ ready · pulse_180ms}	`cortex-a7`	54.2 MB → 57.6 MB
⚠️ +6.1%	inceptionv3 _{RSS @ ready · pass}	`orangepi-rv2`	120 MB → 127 MB
⚠️ +6.0%	mobilenet_v2_1 _{RSS @ ready · pass_mt}	`orangepi-rv2`	42 MB → 44.5 MB
⚠️ +6.0%	en_tdnn_8M _{RSS @ ready · 2600ms}	`cortex-a7`	53.1 MB → 56.2 MB
⚠️ +5.9%	en_tdnn_8M _{RSS @ ready · pulse_120ms}	`cortex-a7`	54.4 MB → 57.6 MB
⚠️ +5.8%	inceptionv3 _{RSS @ ready · pass}	`beaglev-ahead`	120 MB → 127 MB
⚠️ +5.8%	en_tdnn_8M_nnef _{RSS @ ready · pulse_240ms}	`cortex-a9`	51.9 MB → 55 MB
⚠️ +5.8%	mobilenet_v2_1 _{RSS @ ready · pass_mt}	`beaglev-ahead`	42 MB → 44.4 MB
⚠️ +5.7%	mobilenet_v2_1 _{RSS @ ready · pass}	`beaglev-ahead`	42 MB → 44.4 MB
⚠️ +5.7%	inceptionv3 _{RSS @ load · pass}	`orangepi-rv2`	115 MB → 121 MB
⚠️ +5.7%	inceptionv3 _{RSS @ ready · pass_mt}	`orangepi-rv2`	120 MB → 127 MB
⚠️ +5.6%	nemotron_3_5_asr_streaming_0_6b_f32f32_decoder_pass _{RSS @ load · cpu}	`i9-9900k`	74.6 MB → 78.8 MB
⚠️ +5.6%	en_tdnn_8M _{RSS @ ready · pulse_240ms}	`cortex-a9`	54.5 MB → 57.5 MB
⚠️ +5.5%	en_tdnn_8M _{RSS @ ready · pulse_120ms}	`cortex-a9`	54.4 MB → 57.4 MB
⚠️ +5.5%	nemotron_3_5_asr_streaming_0_6b_f32f32_decoder_pass _{RSS @ load · cuda}	`i9-11900kb_rtx-4060`	81.7 MB → 86.1 MB
⚠️ +5.5%	inceptionv3 _{RSS @ ready · pass_mt}	`beaglev-ahead`	120 MB → 127 MB
⚠️ +5.4%	nemotron_3_5_asr_streaming_0_6b_f32f32_decoder_pass _{RSS @ load · cpu}	`i9-11900kb_rtx-4060`	81.9 MB → 86.3 MB
⚠️ +5.4%	inceptionv3 _{RSS @ load · pass_mt}	`orangepi-rv2`	115 MB → 121 MB
⚠️ +5.3%	en_tdnn_8M _{RSS @ ready · pulse_240ms}	`cortex-a7`	54.3 MB → 57.1 MB
⚠️ +5.3%	inceptionv3 _{RSS @ load · pass}	`beaglev-ahead`	115 MB → 121 MB
⚠️ +5.2%	inceptionv3 _{RSS @ load · pass_mt}	`beaglev-ahead`	115 MB → 121 MB
⚠️ +5.1%	parakeet_tdt_600m_v3_f32f32_joint_pass _{RSS @ ready · cpu}	`i9-9900k`	44.1 MB → 46.3 MB
⚠️ +5.1%	en_tdnn_15M _{load · pulse_120ms}	`cortex-a9`	2.84 s → 2.99 s
⚠️ +5.0%	en_tdnn_pyt_15M _{RSS @ ready · pulse_120ms}	`beaglev-ahead`	61.9 MB → 65.1 MB

Bench numbers come from shared CI hardware and do produce false positives: a flagged
row is not on its own evidence of a real regression, and a clean report is not proof
there is none. Contributors do not need to chase what is reported here — leave the
reading of it to the maintainers, who will say if something needs action.

kali force-pushed the perf/cli-jemalloc branch from 6274fd9 to 96c9902 Compare September 7, 2026 15:08

kali force-pushed the perf/cli-jemalloc branch from 96c9902 to 8d22bcc Compare September 8, 2026 07:35

kali marked this pull request as draft September 8, 2026 08:47

kali force-pushed the perf/cli-jemalloc branch from 8d22bcc to 284c07a Compare October 13, 2026 11:54

kali marked this pull request as ready for review October 13, 2026 11:54

kali force-pushed the perf/cli-jemalloc branch from 284c07a to c485301 Compare October 13, 2026 14:23

kali added 6 commits October 13, 2026 19:58

kali force-pushed the perf/cli-jemalloc branch from c485301 to ebcf0a2 Compare October 13, 2026 17:58

kali merged commit a4c1f6e into main Oct 13, 2026
78 checks passed

Navigation Menu

Repository navigation

cli: glibc's allocator carries a lot of overhead on the graphs the be… - #2789

cli: glibc's allocator carries a lot of overhead on the graphs the be…#2789
kali merged 6 commits into
mainfrom
perf/cli-jemalloc

kali commented Sep 7, 2026

Uh oh!

github-actions Bot commented Sep 7, 2026

Uh oh!

github-actions Bot commented Sep 7, 2026

Uh oh!

github-actions Bot commented Sep 7, 2026

Uh oh!

github-actions Bot commented Sep 7, 2026

Uh oh!

kali commented Sep 8, 2026

Uh oh!

github-actions Bot commented Sep 8, 2026

Uh oh!

github-actions Bot commented Oct 13, 2026

Uh oh!

github-actions Bot commented Oct 13, 2026

Uh oh!

kali commented Oct 13, 2026

Uh oh!

github-actions Bot commented Oct 13, 2026

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

1 participant

Conversation

kali commented Sep 7, 2026

Uh oh!

github-actions Bot commented Sep 7, 2026

Uh oh!

github-actions Bot commented Sep 7, 2026

Uh oh!

github-actions Bot commented Sep 7, 2026

Uh oh!

github-actions Bot commented Sep 7, 2026

Uh oh!

kali commented Sep 8, 2026

Uh oh!

github-actions Bot commented Sep 8, 2026

Uh oh!

github-actions Bot commented Oct 13, 2026

Uh oh!

github-actions Bot commented Oct 13, 2026

Uh oh!

kali commented Oct 13, 2026

Uh oh!

github-actions Bot commented Oct 13, 2026

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

1 participant