vllm
https://github.com/vllm-project/vllm
Python
A high-throughput and memory-efficient inference and serving engine for LLMs
Triage Issues!
When you volunteer to triage issues, you'll receive an email each day with a link to an open issue that needs help in this project. You'll also receive instructions on how to triage issues.
Triage Docs!
Receive a documented method or class from your favorite GitHub repos in your inbox every day. If you're really pro, receive undocumented methods or classes and supercharge your commit history.
Python not yet supported51 Subscribers
Add a CodeTriage badge to vllm
Help out
- Issues
- [Bug]: PD-admitted Mamba request is spec-padded before prefill completes, then align split truncates the 8-token window to 5
- [Feature][KV-offloading]: OffloadingConnector rejects hybrid models whose KV groups have blocks smaller than one hash unit (GLM-5.3-Flash) — per-group blocks_per_chunk implementation attached
- [Feature][KV-offloading]: recent-window state groups can never participate in restores — per-group offload exclusion + hit-boundary rollback (GLM-5.3 tail_cache)
- [Perf] Fuse temperature scaling into the sampler's softmax
- [Bugfix][Spec Decode] Honor draft eager mode in MRV2
- [Frontend] Add Responses SSE keepalive
- [Bugfix] Don't fail the request when a LoRA resolver raises
- [Bugfix] Abort encode() requests when the caller stops early
- [Perf][ROCm] Sparse-MLA/kpool backends fall back to rebuilding attention metadata per draft step, capping useful MTP depth at k=4
- [ROCm] Fused shared experts unavailable for GLM-5.3-Flash: Fp8Config not handled by the FSE gate, and glm5next never wires it
- Docs
- Python not yet supported