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 supported36 Subscribers
View all SubscribersAdd a CodeTriage badge to vllm
Help out
- Issues
- [ROCm][Perf] gfx942: use FlyDSL fp8 MQA logits kernel (ROCm/aiter#3913)
- [RFC]: Speculative recompute fallback for slow remote KV cache loads
- FlashInfer + spec-decode silently downgrades to PIECEWISE cudagraphs (−16% measured); warning understates the cost
- [Performance]: Dynamic speculative decoding (num_speculative_tokens_per_batch_size) causes catastrophic aggregate-throughput collapse under concurrency at the batch-size threshold (MTP, V1/PIECEWISE)
- [Bug]: MTP draft model ignores checkpoint per-layer quantization config (block_name_to_quantize mapped inconsistently vs target)
- [Bug]: In-tree fallback LMCache adapter does not propagate retrieve failures to the engine
- [Bugfix][MoE] Filter packed expert weights during EP loading
- [Bug]: QWEN 3.6-35Ba3B+DFlash (Train by Speculator) when vllm0.25.1 use V2, the acceptance rate is 0%.
- [Feature] Support `extract_hidden_states` on Model Runner V2
- [MRv2] FlashAttention PCP support for GQA on MRv2
- Docs
- Python not yet supported