Use first bad_words as extra parameters, and implement min-p #1536
+116
−37
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Outdated suggestions cannot be applied.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.
An approach for implementing #1154
The user-facing classes in BatchManager and
tensorrt_llm::executor::SamplingConfig
are not open source (the constructor implementation exists in .a files and will segfault if the class is modified), so we hacked it by using the integers in the first bad_words for extra parameters.In this case, we're using the first integer reinterpreted as a float, to represent min_p (the default value is 0.0 which matches the 0 padding in bad_words).
I implemented the min-p by piggybacking on the existing logprobs calculation in cuda, so it should have no additional performance overhead beyond the logprobs calculation.