[bong-water-water-bong]

428400 text-generation models on HF, 340020 with an architecture (2890 architectures).

	models	share
mapped (a backend's code accepts the architecture)	321373	94.52%
checked (passed serve_e2e on Strix Halo)	216851	63.78%

backend	mapped	checked
hrx	94.52%	63.78%
npu	53.67%	9.41%

Unmapped architectures with the most models:

• Step1MoEForCausalLM: 2882
• ParlerTTSForConditionalGeneration: 1585
• ChessForCausalLM: 970
• LlavaLlamaForCausalLM: 453
• DynamicAlibiForCausalLM: 245
• OpenLMForCausalLM: 245
• DynamicForgettingForCausalLM: 240
• BertLMHeadModel: 212
• MobileLLMForCausalLM: 176
• BartForConditionalGeneration: 174

[github-actions]

PR Reviewer Guide 🔍

Here are some key observations to aid the review process:

⏱️ Estimated effort to review: 2 🔵🔵⚪⚪⚪

🧪 No relevant tests

🔒 No security concerns identified

⚡ Recommended focus areas for review Data Consistency The census data shows an increase in the number of models with architectures (340020) and mapped models (321373) compared to the previous day, but the percentage of mapped models (94.52%) is slightly lower than the previous day's 94.53%. This discrepancy could indicate inconsistencies in how models are counted or classified, especially if the increase in total models (428400 vs 428065) is not proportionally matched by an increase in architectures or mapped models. "models": 428400, "with_architecture": 340020, "architectures_seen": 2890, "mapped_models": 321373, "mapped_pct": 94.52, "checked_models": 216851, "checked_pct": 63.78, Backend Coverage The NPU backend shows a significant drop in checked models (31993 vs 31938) and a lower percentage of checked models (9.41% vs 9.4%) compared to the previous day, while the HRX backend shows a slight increase in both mapped and checked models. This could indicate a regression or a change in the testing pipeline for the NPU backend. "npu": { "mapped": 182487, "checked": 31993, "mapped_pct": 53.67, "checked_pct": 9.41 }