[nvbugs/5333742] fix MTP illegal memory access in cuda graph warmup #5947
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Signed-off-by: Fanrong Li <[email protected]>
litaotju
approved these changes
Jul 11, 2025
Signed-off-by: Fanrong Li <[email protected]>
|
/bot run |
|
PR_Github #11644 [ run ] triggered by Bot |
|
PR_Github #11644 [ run ] completed with state |
|
/bot run |
|
PR_Github #11662 [ run ] triggered by Bot |
|
PR_Github #11662 [ run ] completed with state |
achartier
approved these changes
Jul 11, 2025
mikeiovine
approved these changes
Jul 11, 2025
|
/bot run |
|
PR_Github #11702 [ run ] triggered by Bot |
|
PR_Github #11702 [ run ] completed with state |
k-l-lambda
pushed a commit
to k-l-lambda/TensorRT-LLM
that referenced
this pull request
Jul 14, 2025
…VIDIA#5947) Signed-off-by: Fanrong Li <[email protected]>
lfr-0531
added a commit
to lfr-0531/TensorRT-LLM
that referenced
this pull request
Jul 14, 2025
…VIDIA#5947) Signed-off-by: Fanrong Li <[email protected]>
lfr-0531
added a commit
to lfr-0531/TensorRT-LLM
that referenced
this pull request
Jul 14, 2025
…VIDIA#5947) Signed-off-by: Fanrong Li <[email protected]>
|
@lfr-0531 can you please add an Mr description and Pr title please? |
lfr-0531
added a commit
to lfr-0531/TensorRT-LLM
that referenced
this pull request
Jul 15, 2025
…VIDIA#5947) Signed-off-by: Fanrong Li <[email protected]>
lfr-0531
added a commit
that referenced
this pull request
Jul 15, 2025
Signed-off-by: Fanrong Li <[email protected]>
|
Can you fill the description and add a title? Thanks. |
evezhier
pushed a commit
to evezhier/TensorRT-LLM
that referenced
this pull request
Jul 16, 2025
Signed-off-by: Fanrong Li <[email protected]>
Wanli-Jiang
pushed a commit
to Wanli-Jiang/TensorRT-LLM
that referenced
this pull request
Jul 17, 2025
…VIDIA#5947) Signed-off-by: Fanrong Li <[email protected]> Signed-off-by: Wanli Jiang <[email protected]>
dc3671
pushed a commit
to dc3671/TensorRT-LLM
that referenced
this pull request
Jul 21, 2025
…VIDIA#5947) Signed-off-by: Fanrong Li <[email protected]>
4 tasks
4 tasks
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
[nvbugs/5333742] fix MTP illegal memory access in cuda graph warmup
Description
We found an illegal memory access issue in the CUDA graph warmup. The root cause:
When enabling the overlap scheduler, the generation requests should use device tensors from the last iteration to prepare inputs. But in warmup, those dummy generation requests don't have the previous
new_tensors_device. So the tensorprevious_pos_id_offsets_cudaandprevious_kv_lens_offsets_cudawon't be set to zerosTensorRT-LLM/tensorrt_llm/_torch/pyexecutor/model_engine.py
Line 1441 in fbb4cc7
_preprocess_inputsthekv_lens_cudawill be set to a wrong valueTensorRT-LLM/tensorrt_llm/_torch/pyexecutor/model_engine.py
Line 1172 in fbb4cc7
applyMLARopeAndAssignQKVKernelGenerationkernel, there will be illegal memory access.Test Coverage
A new test to L0: