d-flex flex-column width-full">

Try fix with kHasMask and kHasSink (mha_varlen_fwd_kernels.cu)

75ab129

Select commit

Loading

Failed to load commit list.

Uh oh!

There was an error while loading. Please reload this page.

PR Welcome Comment on: pull_request_target 1

comment

Kimi Downstream Test on: pull_request

Kimi-K2.5 (${{ matrix.backend }})

Kimi Perf Downstream on: pull_request

Kimi-K2.5 perf (${{ matrix.backend }})

Sglang Downstream Test on: pull_request

check-signal

select-sglang-tests

Sglang ${{ matrix.model }} ${{ matrix.test_type }} (8 GPU)

Atom Test on: pull_request

check-signal

load-atom-models

Accuracy (${{ matrix.display_name }}, ${{ matrix.runner }})

Flash Attention Integration on: pull_request 2

check-signal

prechecks

Flash Attention - Triton / ${{ matrix.label }} (1 GPU)

Windows Smoke Test

Triton-only Smoke

Flash Attention - CK (1 GPU)

OPUS Test on: pull_request 3

check-signal

OPUS Tests (MI35X)

OPUS Tests (MI300X)

Checks on: pull_request 2

Check Repository Dependency

Check Code Style with Black

Check Code Style with Ruff

Aiter Test on: pull_request 22

check-signal

build_aiter_wheels (3.10, rocm/pytorch:rocm7.2.4_ubuntu22.04_py3.10_pytorch_release_2.10.0, false)

build_aiter_wheels (3.12, rocm/pytorch:latest, true)

prepare_triton_wheel / prepare

Upload wheels and manifest to S3

split_aiter_tests

Multi-GPU Tests (8 GPU)

Standard Tests (1 GPU) (linux-aiter-mi35x-1, MI35X, 8, 0)

Standard Tests (1 GPU) (linux-aiter-mi35x-1, MI35X, 8, 1)

Standard Tests (1 GPU) (linux-aiter-mi35x-1, MI35X, 8, 2)

Standard Tests (1 GPU) (linux-aiter-mi35x-1, MI35X, 8, 3)

Standard Tests (1 GPU) (linux-aiter-mi35x-1, MI35X, 8, 4)

Standard Tests (1 GPU) (linux-aiter-mi35x-1, MI35X, 8, 5)

Standard Tests (1 GPU) (linux-aiter-mi35x-1, MI35X, 8, 6)

Standard Tests (1 GPU) (linux-aiter-mi35x-1, MI35X, 8, 7)

Standard Tests (1 GPU) (aiter-1gpu-runner, MI300X, 8, 0)

Standard Tests (1 GPU) (aiter-1gpu-runner, MI300X, 8, 1)

Standard Tests (1 GPU) (aiter-1gpu-runner, MI300X, 8, 2)

Standard Tests (1 GPU) (aiter-1gpu-runner, MI300X, 8, 3)

Standard Tests (1 GPU) (aiter-1gpu-runner, MI300X, 8, 4)

Standard Tests (1 GPU) (aiter-1gpu-runner, MI300X, 8, 5)

Standard Tests (1 GPU) (aiter-1gpu-runner, MI300X, 8, 6)

Standard Tests (1 GPU) (aiter-1gpu-runner, MI300X, 8, 7)

Standard Test Results

Tuned Op Bench

Kimi Downstream Test on: pull_request 2

Kimi-K2.5 (vllm)

Kimi-K2.5 (sglang)

Atom Test on: pull_request 5

check-signal

load-atom-models

Accuracy (DeepSeek-R1-0528, linux-aiter-do-mi350x-8)

Accuracy (gpt-oss-120b, linux-aiter-do-mi350x-8)

Accuracy (Kimi-K2.7-Code-MXFP4, linux-aiter-do-mi350x-4)

vLLM Benchmark on: pull_request 5

check-signal

Build Aiter wheel

vLLM Benchmark (deepseek-ai/DeepSeek-V4-Pro, linux-aiter-do-mi350x-8, fp8_kvcache, 8, --quantizat...

vLLM Benchmark (amd/MiniMax-M3-MXFP4, linux-aiter-do-mi350x-8, default_kvcache, 4, --block-size 1...

vLLM Benchmark (amd/MiniMax-M3-MXFP4, linux-aiter-do-mi350x-8, default_kvcache, 8, --block-size 1...

Sglang Downstream Test on: pull_request 3

check-signal

select-sglang-tests

Sglang DeepSeek-R1-MXFP4 Accuracy (8 GPU)

Kimi Perf Downstream on: pull_request 2

Kimi-K2.5 perf (vllm)

Kimi-K2.5 perf (sglang)

Try fix with kHasMask and kHasSink (mha_varlen_fwd_kernels.cu)

Uh oh!

Select a check to view from the sidebar