[Misc] Logits processor plugins #4769
Open
+228
−14
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Outdated suggestions cannot be applied.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.
This pull request adds support for Logits processor plugins.
This makes implementing custom Logits processors very easy, and eliminates the need to change vLLM directly to implement it.
For example with this merge request we could implement all of the guided decoding features, just by implementing a Python package and installing it in the same virtualenv as vLLM, without actually changing vLLM source code.
Example code for a logits processor plugin that given a token id multiplies its logit by 100:
And the
setup.py
file for the package should look something like this:With this merge request vLLM will load all the plugins at startup, and each inference request can specify usage of custom logits processors using the
logits_processors
field in the request body.The
parameters_model
in the plugin dictionary is used to validate and parse the request body.I will soon add to this pull request a page in the documentation explaining how to implement custom logits processors.