biondizzle/vllm
1
0
Fork 0
Code Issues Pull Requests Actions 2 Packages Projects Releases Wiki Activity

[Feature][OCP MX] Support mxfp6 and mixed mxfp6-mxfp4 (#21166)

Browse Source
This commit is contained in:
fxmarty-amd
2025-10-07 15:35:26 +02:00
committed by GitHub
parent 08d26a1b7e
commit 41f1cf38f2
18 changed files with 656 additions and 180 deletions
Download Patch File Download Diff File Expand all files Collapse all files

4
vllm/utils/flashinfer.py
View File

@@ -89,7 +89,7 @@ flashinfer_trtllm_fp8_per_tensor_scale_moe = _lazy_import_wrapper(
flashinfer_cutlass_fused_moe = _lazy_import_wrapper(
"flashinfer.fused_moe", "cutlass_fused_moe"
)
fp4_quantize = _lazy_import_wrapper("flashinfer", "fp4_quantize")
flashinfer_fp4_quantize = _lazy_import_wrapper("flashinfer", "fp4_quantize")
nvfp4_block_scale_interleave = _lazy_import_wrapper(
"flashinfer", "nvfp4_block_scale_interleave"
)
@@ -442,7 +442,7 @@ __all__ = [
"has_flashinfer",
"flashinfer_trtllm_fp8_block_scale_moe",
"flashinfer_cutlass_fused_moe",
"fp4_quantize",
"flashinfer_fp4_quantize",
"nvfp4_block_scale_interleave",
"trtllm_fp4_block_scale_moe",
"autotune",