Fix the unit test errors / enable accuracy tests #150
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
lucaslie
reviewed
Oct 1, 2025
| # For generate-only (s == 1), caches must carry prior state. | ||
| if num_prefill > 0 and slot_idx_decode.numel() > 0: | ||
| zero_rows = torch.zeros_like(conv_state_cache.index_select(0, slot_idx_decode)) | ||
| conv_state_cache.index_copy_(0, slot_idx_decode, zero_rows) |
There was a problem hiding this comment.
why would we reset cache state for decode?
lucaslie
reviewed
Oct 2, 2025
Comment on lines
189
to
197
| slot_idx_decode = slot_idx[num_prefill:].to(torch.long) | ||
| y_dec = causal_conv1d_update( | ||
| x_decode, # [batch, dim] | ||
| conv_state_cache, | ||
| w2d, | ||
| bias, | ||
| activation=None, | ||
| cache_seqlens=None, | ||
| conv_state_indices=slot_idx[num_prefill:].to(torch.int32), | ||
| conv_state_indices=slot_idx_decode.to(torch.int32), |
There was a problem hiding this comment.
there is two type casts here?
Comment on lines
+245
to
+246
| # Initialize to zeros so brand-new sequences start from a clean state. | ||
| return torch.zeros( |
There was a problem hiding this comment.
nit: not needed when we correctly index caches
|
see NVIDIA#8133 for the accuracy test |
* [None][auto_deploy] Bamba Signed-off-by: William Zhang <[email protected]> * debugging export accuracy diff for bamba Signed-off-by: Lucas Liebenwein <[email protected]> --------- Signed-off-by: William Zhang <[email protected]> Signed-off-by: Lucas Liebenwein <[email protected]> Co-authored-by: William Zhang <[email protected]>
Signed-off-by: Lucas Liebenwein <[email protected]>
Signed-off-by: Chenghao Zhang <[email protected]>
* Fix the bamba unit test Signed-off-by: Chenghao Zhang <[email protected]> * none: Add triton backend for ssm_transform and cuda backend for conv Signed-off-by: Chenghao Zhang <[email protected]> * Fully Use the TRT LLM kernels Signed-off-by: Chenghao Zhang <[email protected]> * Add fake version for ssm transform op Signed-off-by: Chenghao Zhang <[email protected]> * Fix the datatype error in fake op Signed-off-by: Chenghao Zhang <[email protected]> * Fix the conv test error Signed-off-by: Chenghao Zhang <[email protected]> * Fix the triton ssm error Signed-off-by: Chenghao Zhang <[email protected]> --------- Signed-off-by: Chenghao Zhang <[email protected]>
…es with better reset/sizing (#140) Signed-off-by: Lucas Liebenwein <[email protected]>
Signed-off-by: Lucas Liebenwein <[email protected]>
Signed-off-by: Lucas Liebenwein <[email protected]>
* Fix the bamba unit test Signed-off-by: Chenghao Zhang <[email protected]> * none: Add triton backend for ssm_transform and cuda backend for conv Signed-off-by: Chenghao Zhang <[email protected]> * Fully Use the TRT LLM kernels Signed-off-by: Chenghao Zhang <[email protected]> * Add fake version for ssm transform op Signed-off-by: Chenghao Zhang <[email protected]> * Fix the datatype error in fake op Signed-off-by: Chenghao Zhang <[email protected]> * Fix the conv test error Signed-off-by: Chenghao Zhang <[email protected]> * Fix the triton ssm error Signed-off-by: Chenghao Zhang <[email protected]> * Fix the DemoLLM sampler mismatch Signed-off-by: Chenghao Zhang <[email protected]> * Update the implementation for triton/cuda kernels Signed-off-by: Chenghao Zhang <[email protected]> * Fix the d2d memcpy for decode Signed-off-by: Chenghao Zhang <[email protected]> * Revert the generator and remove the redundant code Signed-off-by: Chenghao Zhang <[email protected]> --------- Signed-off-by: Chenghao Zhang <[email protected]> Signed-off-by: Suyog Gupta <[email protected]> Co-authored-by: Suyog Gupta <[email protected]>
* [None][feat] Add patches for NemotronH Signed-off-by: William Zhang <[email protected]> * [None][test] unittest for nemotron_h Signed-off-by: William Zhang <[email protected]> * nemotron-h support finished Signed-off-by: Lucas Liebenwein <[email protected]> * added anticapted path for new models on llm_models trt-llm CI Signed-off-by: Lucas Liebenwein <[email protected]> --------- Signed-off-by: William Zhang <[email protected]> Signed-off-by: Lucas Liebenwein <[email protected]> Co-authored-by: William Zhang <[email protected]>
Signed-off-by: Lucas Liebenwein <[email protected]>
Signed-off-by: Chenghao Zhang <[email protected]>
Signed-off-by: Chenghao Zhang <[email protected]>
Signed-off-by: Chenghao Zhang <[email protected]>
Signed-off-by: Chenghao Zhang <[email protected]>
Signed-off-by: Chenghao Zhang <[email protected]>
Signed-off-by: Chenghao Zhang <[email protected]>
Signed-off-by: Chenghao Zhang <[email protected]>
Signed-off-by: Lucas Liebenwein <[email protected]>
This reverts commit 67ee3d8.
Signed-off-by: Lucas Liebenwein <[email protected]>
Signed-off-by: Chenghao Zhang <[email protected]>
Signed-off-by: Chenghao Zhang <[email protected]>
nvchenghaoz
force-pushed
the
chenghao/fix-causal-conv
branch
from
1199afe to
fe22f1c
Compare
Signed-off-by: Chenghao Zhang <[email protected]>
Signed-off-by: Chenghao Zhang <[email protected]>
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
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.
@coderabbitai summary
Several updates
Enable the accuracy testing for nemotron-h, added tests for MMLU and gsm8k
Fix two unit tests.