onnxruntime
fix(webgpu): use f32 accumulators in fp16 MatMul/MatMulNBits to prevent overflow (#26732)
#29599

Open

fix(webgpu): use f32 accumulators in fp16 MatMul/MatMulNBits to prevent overflow (#26732) #29599

RobertoReale wants to merge 4 commits into microsoft:main from RobertoReale:fix/webgpu-f16-overflow

fix(webgpu): use f32 accumulators in fp16 MatMul/MatMulNBits to preve…

4d83b9b6

github-advanced-security commented on 2026-07-07

tianleiwu commented on 2026-07-08

Address review feedback: extend f32 accumulation to fused native kern…

2f2a4b3a

Extend f32 accumulation to the generic dp4a_matmul kernel

53c9320c

Fix vec4 constructor type mismatch in dp4a qualcomm store path

2cf1e378

Reviewers

tianleiwu

Assignees

No one assigned

Labels

None yet

Milestone

No milestone