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 supported62 Subscribers
View all SubscribersAdd a CodeTriage badge to vllm
Help out
- Issues
- [Feature][Frontend] Add optional fastapi-guard security middleware
- [KV Cache] Size packed groups by pooled bytes per request, not layer padding
- [Bug] DFlash draft model: fp8 `ignored_layers` are not aliased to global layer indices, so excluded bf16 projections are loaded into fp8 parameters (acceptance drops from 4.3 to 1.0)
- [ROCm][KV Connector] Support sleep mode with MoRIIOConnector over RDMA
- [Test][V1] Cover input-queue arrivals during uncapped prefill
- [Kernel][XPU] Mamba2 SSD: tensor-descriptor path for _chunk_scan_fwd_kernel and _chunk_state_fwd_kernel
- [Bug]: DeepSeek-V4.1-Flash worker dies with `CUDA error: unspecified launch failure` during long-context decode with dspark spec decode (DP=3, TP=2, EP); Anthropic `/v1/messages` adapter then fails on the error chunk
- [Bugfix][Attention] Fall back when FlashInfer attention JIT is unavailable
- [Bugfix] Preserve timed-trace chunk identity across benchmark runs
- [Feature]: Allow pooling tasks (token_embed / token_classify) to reuse prefix cache while returning full hidden-state sequences
- Docs
- Python not yet supported