Skip to content
ONNX Runtime CUDA Builds

Add zero point support to dp4a 2-bit dequantization in the WebGPU MatMulNbits #10443

Sign in to view logs

Add zero point support to dp4a 2-bit dequantization in the WebGPU MatMulNbits

Add zero point support to dp4a 2-bit dequantization in the WebGPU MatMulNbits #10443

Triggered via pull request February 13, 2026 18:24
@HectorSVCHectorSVC
synchronize #27325
HectorSVC:hecli_webgpu_2bit_improvement
Status Success
Total duration 1h 36m 12s
Artifacts 1

windows_cuda.yml

on: pull_request
Windows GPU CUDA CI Pipeline
44m 12s
Windows GPU CUDA CI Pipeline
Windows GPU CUDA CI Pipeline Test Job
28m 33s
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
1.99 GB
sha256:ca4402da0f192f7ab30df6e103ef16716c72051c6b0f4188077268746751bda6