Skip to content

[proof-of-concept] Enable NVFP4 recipe #517

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Draft
matthiasdiener wants to merge 1 commit into
base: mdiener/nvfp4-cast_fusion
Choose a base branch
from
Draft

[proof-of-concept] Enable NVFP4 recipe #517

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
6 changes: 3 additions & 3 deletions transformer_engine/common/CMakeLists.txt
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -214,7 +214,8 @@ list(APPEND transformer_engine_cuda_sources
fused_router/fused_topk_with_score_function.cu
recipe/current_scaling.cu
recipe/delayed_scaling.cu
recipe/fp8_block_scaling.cu)
recipe/fp8_block_scaling.cu
recipe/nvfp4.cu)

list(APPEND transformer_engine_cuda_arch_specific_sources
cast/cast.cu
Expand All @@ -238,8 +239,7 @@ if(USE_CUDA)
fused_attn/fused_attn_fp8.cu
fused_attn/utils.cu
swizzle/swizzle.cu
swizzle/swizzle_block_scaling.cu
recipe/nvfp4.cu)
swizzle/swizzle_block_scaling.cu)
list(APPEND transformer_engine_cuda_arch_specific_sources
gemm/cutlass_grouped_gemm.cu
transpose/quantize_transpose_square_blockwise.cu)
Expand Down
2 changes: 1 addition & 1 deletion transformer_engine/pytorch/quantization.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -87,7 +87,7 @@ def check_mxfp8_support() -> Tuple[bool, str]:
@functools.lru_cache(maxsize=None)
def check_nvfp4_support() -> Tuple[bool, str]:
if IS_HIP_EXTENSION:
return False, "ROCm TE currently not supporting NVFP4"
return True, ""
"""Return if nvfp4 support is available"""
if get_device_compute_capability() >= (10, 0): # blackwell and above
return True, ""
Expand Down