Skip to content

[TRTLLM-9661][chore] Further reduce tuning time for cuteDSL nvFP4 dense gemm. #10339

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

Open
hyukn wants to merge 3 commits into
base: main
Choose a base branch
from
+8 −1
Open

[TRTLLM-9661][chore] Further reduce tuning time for cuteDSL nvFP4 dense gemm. #10339

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
3 commits
Select commit Hold shift + click to select a range
e9fb82a
[TRTLLM-9661][chore] Further reduce tuning time for cuteDSL nvFP4 den…
hyukn Dec 30, 2025
a256ba9
Merge branch 'main' into feat/autotuner_cute_dsl_fp4_gemm_tune_time
hyukn Dec 31, 2025
8336412
Merge branch 'main' into feat/autotuner_cute_dsl_fp4_gemm_tune_time
hyukn Jan 1, 2026
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
4 changes: 3 additions & 1 deletion tensorrt_llm/_torch/custom_ops/cute_dsl_custom_ops.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -454,7 +454,9 @@ def get_valid_tactics(
(4, 4),
]
swap_ab_candidates = [True, False]
use_prefetch_candidates = [True, False]
# prune: prefetch is beneficial only when K is large enough
use_prefetch_candidates = [True, False
] if real_k >= 16384 else [False]

valid_tactics = []
for mma_tiler_mn, cluster_shape_mn, swap_ab, use_prefetch in itertools.product(
Expand Down
5 changes: 5 additions & 0 deletions tensorrt_llm/_torch/modules/linear.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -2105,6 +2105,11 @@ def __init__(
'cutlass', 'cublaslt', 'cuda_core'
]

# Add cutedsl to the allowed backends if tp size is greater than or equal to 4,
# because distributed tuning can decrease the tuning time by tp_size.
if self.tp_size >= 4 and 'cutedsl' not in self.nvfp4_allowed_backends:
self.nvfp4_allowed_backends.append('cutedsl')

local_in_features = in_features
local_out_features = out_features

Expand Down