Configuration to allow output of special tokens #970
Closed
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
…d of always skipping them.
|
This is an approach to addressing #893. |
|
Hi @blahblahasdf, sorry for the late response. Could you elaborate more on the reason you add it as a model parameter instead of a sampling parameter? |
|
No worries @WoosukKwon . For my use cases I want the special tokens for every request. To me, this meant it should be a configuration of the model more than the request. As I said I could easily see it the other way and I'd be happy to change this to an option on |
|
@blahblahasdf Thanks! I believe HF transformers also views it as a generation parameter, rather than a model's hyper parameter. Could you fix the code? |
|
Great, will do! |
|
I created a new PR for the different approach. #1186 |
amy-why-3459
pushed a commit
to amy-why-3459/vllm
that referenced
this pull request
Sep 15, 2025
…ject#970) ### What this PR does / why we need it? Optimize the performance of calculation logic in sampler and deepseekv2. ### Does this PR introduce _any_ user-facing change? Added VLLM_ENABLE_TOPK_OPTIMZE config in sampler ### How was this patch tested? pytest test_sampler.py Signed-off-by: wangxiaoxin (A) <wangxiaoxin7@huawei.com> Co-authored-by: wangxiaoxin (A) <wangxiaoxin7@huawei.com> Co-authored-by: ZhengWG <zwg0606@gmail.com>
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Outdated suggestions cannot be applied.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.
I have a model that generates special tokens with important meaning to generation tasks. At present there is no way to get these special tokens back in the generated text because of the hardcoded input to
detokenize_incrementallyinllm_engine.This PR adds an option at start up to
--keep-special-tokenswhich resolves this limitation.I elected to resolve this with a
ModelConfigchange instead of changingSamplingParamsbut I could see an argument the other way.