[NVIDIA] Add SM100 Flashinfer MoE per tensor scale fp8 backend (#21458)

Signed-off-by: Amir Klein <203507526+amirkl94@users.noreply.github.com>
Signed-off-by: mgoin <mgoin64@gmail.com>
Co-authored-by: mgoin <mgoin64@gmail.com>
This commit is contained in:
amirkl94
2025-07-31 06:00:01 -07:00
committed by GitHub
co-authored by mgoin
parent 5daffe7cf6
commit 207b750e19
5 changed files with 269 additions and 49 deletions
+2
View File
@@ -66,6 +66,8 @@ def _lazy_import_wrapper(module_name: str,
# Create lazy wrappers for each function
flashinfer_trtllm_fp8_block_scale_moe = _lazy_import_wrapper(
"flashinfer.fused_moe", "trtllm_fp8_block_scale_moe")
flashinfer_trtllm_fp8_per_tensor_scale_moe = _lazy_import_wrapper(
"flashinfer.fused_moe", "trtllm_fp8_per_tensor_scale_moe")
flashinfer_cutlass_fused_moe = _lazy_import_wrapper("flashinfer.fused_moe",
"cutlass_fused_moe")
fp4_quantize = _lazy_import_wrapper("flashinfer", "fp4_quantize")