forked from Karylab-cklius/vllm
Luciano Martins
9b4e83934d
Model (gemma4_mtp.py): - Q-only attention layers sharing KV cache with target model via kv_sharing_target_layer_name (no K/V projections or norms) - pre_projection(2*backbone_dim -> draft_dim) -> decoder layers -> norm -> post_projection(draft_dim -> backbone_dim) - forward() returns (draft_hidden, backbone_hidden) tuple for compute_logits and hidden-state feedback buffer respectively - Embeddings shared with target model; lm_head tied to original draft-dim embed_tokens and preserved across sharing Centroids masking (Gemma4MTPMaskedEmbedder): - Centroid-based sparse logit computation for E2B/E4B assistants (use_ordered_embeddings=True), inactive for 26B/31B - Centroid projection (hidden_size -> num_centroids) selects top-K centroids, gathers candidate embeddings, computes sparse dot products - Shared pipeline in _select_and_score serves both forward() (full-vocab scatter) and get_top_tokens() (sparse argmax) - TP>1 support via all-gather of sharded lm_head.weight - CUDA graph acceleration: capture graphs at batch sizes [1,2,4,8,16,32,64] during load_model, replay in _greedy_sample to eliminate per-step kernel launch overhead Proposer (gemma4.py): - constant_draft_positions: all draft steps reuse last target position - Multi-group KV cache: per-group block tables with correct block_table_tensor per attention group (sliding vs full) - Cross-model KV sharing: maps each draft layer to last non-KV-shared target layer of same attention type - Override _maybe_share_lm_head to preserve draft-dim lm_head - Override _create_draft_vllm_config to carry target's forced TRITON_ATTN backend to draft layers (prevents FLASH_ATTN fallback for sliding attention with KV-shared cache) Framework changes (llm_base_proposer.py): - Extract _update_positions_dependent_metadata helper from draft loop - Cache attention metadata when constant_draft_positions is True Signed-off-by: Luciano Martins <lucianommartins@users.noreply.github.com>
Easy, fast, and cheap LLM serving for everyone
| Documentation | Blog | Paper | Twitter/X | User Forum | Developer Slack |
🔥 We have built a vLLM website to help you get started with vLLM. Please visit vllm.ai to learn more. For events, please visit vllm.ai/events to join us.
About
vLLM is a fast and easy-to-use library for LLM inference and serving.
Originally developed in the Sky Computing Lab at UC Berkeley, vLLM has grown into one of the most active open-source AI projects built and maintained by a diverse community of many dozens of academic institutions and companies from over 2000 contributors.
vLLM is fast with:
- State-of-the-art serving throughput
- Efficient management of attention key and value memory with PagedAttention
- Continuous batching of incoming requests, chunked prefill, prefix caching
- Fast and flexible model execution with piecewise and full CUDA/HIP graphs
- Quantization: FP8, MXFP8/MXFP4, NVFP4, INT8, INT4, GPTQ/AWQ, GGUF, compressed-tensors, ModelOpt, TorchAO, and more
- Optimized attention kernels including FlashAttention, FlashInfer, TRTLLM-GEN, FlashMLA, and Triton
- Optimized GEMM/MoE kernels for various precisions using CUTLASS, TRTLLM-GEN, CuTeDSL
- Speculative decoding including n-gram, suffix, EAGLE, DFlash
- Automatic kernel generation and graph-level transformations using torch.compile
- Disaggregated prefill, decode, and encode
vLLM is flexible and easy to use with:
- Seamless integration with popular Hugging Face models
- High-throughput serving with various decoding algorithms, including parallel sampling, beam search, and more
- Tensor, pipeline, data, expert, and context parallelism for distributed inference
- Streaming outputs
- Generation of structured outputs using xgrammar or guidance
- Tool calling and reasoning parsers
- OpenAI-compatible API server, plus Anthropic Messages API and gRPC support
- Efficient multi-LoRA support for dense and MoE layers
- Support for NVIDIA GPUs, AMD GPUs, and x86/ARM/PowerPC CPUs. Additionally, diverse hardware plugins such as Google TPUs, Intel Gaudi, IBM Spyre, Huawei Ascend, Rebellions NPU, Apple Silicon, MetaX GPU, and more.
vLLM seamlessly supports 200+ model architectures on Hugging Face, including:
- Decoder-only LLMs (e.g., Llama, Qwen, Gemma)
- Mixture-of-Expert LLMs (e.g., Mixtral, DeepSeek-V3, Qwen-MoE, GPT-OSS)
- Hybrid attention and state-space models (e.g., Mamba, Qwen3.5)
- Multi-modal models (e.g., LLaVA, Qwen-VL, Pixtral)
- Embedding and retrieval models (e.g., E5-Mistral, GTE, ColBERT)
- Reward and classification models (e.g., Qwen-Math)
Find the full list of supported models here.
Getting Started
Install vLLM with uv (recommended) or pip:
uv pip install vllm
Or build from source for development.
Visit our documentation to learn more.
Contributing
We welcome and value any contributions and collaborations. Please check out Contributing to vLLM for how to get involved.
Citation
If you use vLLM for your research, please cite our paper:
@inproceedings{kwon2023efficient,
title={Efficient Memory Management for Large Language Model Serving with PagedAttention},
author={Woosuk Kwon and Zhuohan Li and Siyuan Zhuang and Ying Sheng and Lianmin Zheng and Cody Hao Yu and Joseph E. Gonzalez and Hao Zhang and Ion Stoica},
booktitle={Proceedings of the ACM SIGOPS 29th Symposium on Operating Systems Principles},
year={2023}
}
Contact Us
- For technical questions and feature requests, please use GitHub Issues
- For discussing with fellow users, please use the vLLM Forum
- For coordinating contributions and development, please use Slack
- For security disclosures, please use GitHub's Security Advisories feature
- For collaborations and partnerships, please contact us at collaboration@vllm.ai
Media Kit
- If you wish to use vLLM's logo, please refer to our media kit repo