[Kernel] Add NVFP4 support to the torch linear backend (#53319)

Signed-off-by: Michael Lazos <mlazos@meta.com>
Co-authored-by: OpenAI Codex <codex@openai.com>
This commit is contained in:
Michael Lazos
2026-09-08 20:30:05 -04:00
committed by GitHub
co-authored by OpenAI Codex
parent 4aadb0f14d
commit 28a73ccfba
3 changed files with 102 additions and 0 deletions
+2
View File
@@ -94,6 +94,7 @@ SM_100_NVFP4_BACKENDS = [
"flashinfer_cudnn",
"flashinfer_trtllm",
"flashinfer_cutlass",
"torch",
]
@@ -107,6 +108,7 @@ SM_100_NVFP4_BACKENDS = [
"flashinfer_cudnn",
"flashinfer_trtllm", # the small seq_len ensures trtllm_8x4_layout backend is used
"flashinfer_cutlass",
"torch",
],
)
def test_nvfp4(vllm_runner, model, eager, backend):