Skip to content

[https://nvbugs/6244474][fix] AutoDeploy: skip explicit shape-prop after MLIR elementwise fusion #14795

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

Merged
MrGeva merged 2 commits into from
Jun 4, 2026
+8 −9
Merged

[https://nvbugs/6244474][fix] AutoDeploy: skip explicit shape-prop after MLIR elementwise fusion #14795

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
ebe640c
[nvbugs/6244474][fix] Skip explicit shape-prop after MLIR elementwise…
tensorrt-cicd May 31, 2026
30979b5
[nvbugs/6244474][chore] Remove stale waiver after fix
tensorrt-cicd May 31, 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
16 changes: 8 additions & 8 deletions tensorrt_llm/_torch/auto_deploy/transform/library/mlir_elementwise_fusion.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -166,15 +166,15 @@ def _min_output_rank(sg):
back_converter = MLIRToFXConverter(gm)
new_gm = back_converter.convert(mlir_module, converter.metadata)

# Step 6: Re-validate the graph and propagate shapes.
from ...utils._graph import canonicalize_graph, run_shape_prop

canonicalize_graph(new_gm)
run_shape_prop(new_gm)

# Step 6: Defer graph cleanup and shape-prop to the framework. Running
# fake-tensor shape-prop here is unsafe at post_load_fusion: the graph
# may be in a deliberately invalid intermediate state (e.g.
# fuse_rope_into_trtllm_attention rewires Q/K/V to a fused-QKV tensor
# whose rank does not match torch_attention.register_fake; the op swap
# happens later at cache_init).
return new_gm, TransformInfo(
skipped=False,
num_matches=num_replaced,
is_clean=True,
has_valid_shapes=True,
is_clean=False,
has_valid_shapes=False,
)
1 change: 0 additions & 1 deletion tests/integration/test_lists/waives.txt
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -299,7 +299,6 @@ perf/test_perf_sanity.py::test_e2e[aggr_upload-k25_thinking_fp4_2_nodes_grace_bl
perf/test_perf_sanity.py::test_e2e[aggr_upload-k25_thinking_fp4_blackwell-k25_thinking_fp4_dep8_32k8k] SKIP (https://nvbugs/6236094)
perf/test_perf_sanity.py::test_e2e[aggr_upload-k25_thinking_fp4_blackwell-k25_thinking_fp4_dep8_8k1k] SKIP (https://nvbugs/6227472)
perf/test_perf_sanity.py::test_e2e[aggr_upload-k25_thinking_fp4_blackwell-k25_thinking_fp4_tep8_32k8k] SKIP (https://nvbugs/6227472)
perf/test_perf_sanity.py::test_e2e[aggr_upload-llama3_1_8b_fp8_ad_hopper-llama3_1_8b_ad_ws1_1k1k] SKIP (https://nvbugs/6244474)
perf/test_perf_sanity.py::test_e2e[aggr_upload-super_ad_blackwell-super_ad_ws1_1k1k] SKIP (https://nvbugs/6153575)
perf/test_perf_sanity.py::test_e2e[disagg_upload-e2e-gb200_deepseek-r1-fp4_128k8k_con128_ctx1_pp8_gen1_dep16_eplb0_mtp1_ccb-NIXL] SKIP (https://nvbugs/6215844)
perf/test_perf_sanity.py::test_e2e[disagg_upload-e2e-gb200_kimi-k25-thinking-fp4_8k1k_con4096_ctx1_dep4_gen1_dep16_eplb0_mtp0_ccb-NIXL] SKIP (https://nvbugs/6179661)
Expand Down
Loading