Skip to content
ONNX Runtime CUDA Builds

[CUDA] Support FP8 (E4M3) KV Cache for Group Query Attention #10463

Sign in to view logs

[CUDA] Support FP8 (E4M3) KV Cache for Group Query Attention

[CUDA] Support FP8 (E4M3) KV Cache for Group Query Attention #10463

Triggered via pull request February 14, 2026 04:12
@tianleiwutianleiwu
synchronize #27321
tlwu/20260211/gqa_fp8_kv_cache
Status Success
Total duration 1h 21m 26s
Artifacts 1

windows_cuda.yml

on: pull_request
Windows GPU CUDA CI Pipeline
44m 9s
Windows GPU CUDA CI Pipeline
Windows GPU CUDA CI Pipeline Test Job
27m 6s
Windows GPU CUDA CI Pipeline Test Job
Fit to window
Zoom out
Zoom in

Annotations

6 warnings
Windows GPU CUDA CI Pipeline: onnxruntime/core/mlas/lib/amd64/QgemmU8X8KernelAvx2.asm#L1234
epilog offset from end of function exceeds 4095
Windows GPU CUDA CI Pipeline: onnxruntime/core/mlas/lib/amd64/QgemmU8X8KernelAvx2.asm#L1227
epilog offset from end of function exceeds 4095
Windows GPU CUDA CI Pipeline: onnxruntime/core/mlas/lib/amd64/QgemmU8X8KernelAvx2.asm#L1220
epilog offset from end of function exceeds 4095
Windows GPU CUDA CI Pipeline: onnxruntime/core/mlas/lib/amd64/QgemmU8X8KernelAvx2.asm#L1213
epilog offset from end of function exceeds 4095
Windows GPU CUDA CI Pipeline: onnxruntime/core/mlas/lib/amd64/QgemmU8X8KernelAvx2.asm#L1206
epilog offset from end of function exceeds 4095
Windows GPU CUDA CI Pipeline: onnxruntime/core/mlas/lib/amd64/QgemmU8X8KernelAvx2.asm#L1199
epilog offset from end of function exceeds 4095

Artifacts

Produced during runtime
Name Size Digest
build-artifacts
2 GB
sha256:8f1934677753de310beb81a15df387b4e4cb31f84c0d24975865aab4d979c4dc