## Refactor - [ ] Refactor logprob processor layer @Qiaolin-Yu https://github.com/sgl-project/sglang/pull/20071 - [ ] Refactor logprob test kit @Qiaolin-Yu https://github.com/sgl-project/sglang/pull/20558 ## Improvement - [x] Support return_logprob for spec v2 (overlap safe) @Qiaolin-Yu https://github.com/sgl-project/sglang/pull/19801 - [x] Remove sync in get_token_ids_logprobs @Qiaolin-Yu https://github.com/sgl-project/sglang/pull/20972 - [ ] Remove sync in get_token_ids_logprobs_batch_optimized @chanh https://github.com/sgl-project/sglang/pull/20064 - [x] Add logprob related params in bench_serving @Qiaolin-Yu https://github.com/sgl-project/sglang/pull/20218 - [x] Enable piecewise-cuda-graph when logprob_start_len = -1 @Qiaolin-Yu https://github.com/sgl-project/sglang/pull/19453 ## Fix - [ ] Fix spec v2 + logprob when max_num_token is set @Qiaolin-Yu https://github.com/sgl-project/sglang/pull/20799 - [x] Fix spec v1 token_ids_logprobs @Qiaolin-Yu https://github.com/sgl-project/sglang/pull/20718
Refactor
Improvement
Fix