vLLM v0.6.1 — Release Notes & Engineering Impact

## Highlights ### Model Support * Added support for Pixtral (`mistralai/Pixtral-12B-2409`). (#8377, #8168) * Added support for Llava-Next-Video (#7559), Qwen-VL (#8029), Qwen2-VL (#7905) * Multi-input support for LLaVA (#8238), InternVL2 models (#8201) ### Performance Enhancements * Memory optimization for awq_gemm and awq_dequantize, 2x throughput (#8248) ### Production Engine * Support load and unload LoRA in api server (#6566) * Add progress reporting to batch runner (#8060) * Add support for NVIDIA ModelOpt static scaling checkpoints. (#6112) ### Others * Update the docker image to use Python 3.12 for small performance bump. (#8133) * Added CODE_OF_CONDUCT.md (#8161) ## What's Changed * [Doc] [Misc] Create CODE_OF_CONDUCT.md by @mmcelaney in https://github.com/vllm-project/vllm/pull/8161 * [bugfix] Upgrade minimum OpenAI version by @SolitaryThinker in https://github.com/vllm-project/vllm/pull/8169 * [Misc] Clean up RoPE forward_native by @WoosukKwon in https://github.com/vllm-project/vllm/pull/8076 * [ci] Mark LoRA test as soft-fail by @khluu in https://github.com/vllm-project/vllm/pull/8160 * [Core/Bugfix] Add query dtype as per FlashInfer API requirements. by @elfiegg in https://github.com/vllm-project/vllm/pull/8173 * [Doc] Add multi-image input example and update supported models by @DarkLight1337 in https://github.com/vllm-project/vllm/pull/8181 * Inclusion of InternVLChatModel In PP_SUPPORTED_MODELS(Pipeline Parallelism) by @Manikandan-Thangaraj-ZS0321 in https://github.com/vllm-project/vllm/pull/7860 * [MODEL] Qwen Multimodal Support (Qwen-VL / Qwen-VL-Chat) by @alex-jw-brooks in https://github.com/vllm-project/vllm/pull/8029 * Move verify_marlin_supported to GPTQMarlinLinearMethod by @mgoin in https://github.com/vllm-project/vllm/pull/8165 * [Documentation][Spec Decode] Add documentation about lossless guarantees in Speculative Decoding in vLLM by @sroy745 in https://github.com/vllm-project/vllm/pull/7962 * [Core] Support load and unload LoRA in api server by @Jeffwan in https://github.com/vllm-project/vllm/pull/6566 * [BugFix] Fix Granite model configuration by @njhill in https://github.com/vllm-project/vllm/pull/8216 * [Frontend] Add --logprobs argument to `benchmark_serving.py` by @afeldman-nm in https://github.com/vllm-project/vllm/pull/8191 * [Misc] Use ray[adag] dependency instead of cuda by @ruisearch42 in https://github.com/vllm-project/vllm/pull/7938 * [CI/Build] Increasing timeout for multiproc worker tests by @alexeykondrat in https://github.com/vllm-project/vllm/pull/8203 * [Kernel] [Triton] Memory optimization for awq_gemm and awq_dequantize, 2x throughput by @rasmith in https://github.com/vllm-project/vllm/pull/8248 * [Misc] Remove `SqueezeLLM` by @dsikka in https://github.com/vllm-project/vllm/pull/8220 * [Model] Allow loading from original Mistral format by @patrickvonplaten in https://github.com/vllm-project/vllm/pull/8168 * [misc] [doc] [frontend] LLM torch profiler support by @SolitaryThinker in https://github.com/vllm-project/vllm/pull/7943 * [Bugfix] Fix Hermes tool call chat template bug by @K-Mistele in https://github.com/vllm-project/vllm/pull/8256 * [Model] Multi-input support for LLaVA and fix embedding inputs for multi-image models by @DarkLight1337 in https://github.com/vllm-project/vllm/pull/8238 * Enable Random Prefix Caching in Serving Profiling Tool (benchmark_serving.py) by @wschin in https://github.com/vllm-project/vllm/pull/8241 * [tpu][misc] fix typo by @youkaichao in https://github.com/vllm-project/vllm/pull/8260 * [Bugfix] Fix broken OpenAI tensorizer test by @DarkLight1337 in https://github.com/vllm-project/vllm/pull/8258 * [Model][VLM] Support multi-images inputs for InternVL2 models by @Isotr0py in https://github.com/vllm-project/vllm/pull/8201 * [Model][VLM] Decouple weight loading logic for `Paligemma` by @Isotr0py in https://github.com/vllm-project/vllm/pull/8269 * ppc64le: Dockerfile fixed, and a script for buildkite by @sumitd2 in https://github.com/vllm-project/vllm/pull/8026 * [CI/Build] Use python 3.12 in cuda image by @joerunde in https://github.com/vllm-project/vllm/pull/8133 * [Bugfix] Fix async postprocessor in case of preemption by @alexm-neuralmagic in https://github.com/vllm-project/vllm/pull/8267 * [Bugfix] Streamed tool calls now more strictly follow OpenAI's format; ensures Vercel AI SDK compatibility by @K-Mistele in https://github.com/vllm-project/vllm/pull/8272 * [Frontend] Add progress reporting to run_batch.py by @alugowski in https://github.com/vllm-project/vllm/pull/8060 * [Bugfix] Correct adapter usage for cohere and jamba by @vladislavkruglikov in https://github.com/vllm-project/vllm/pull/8292 * [Misc] GPTQ Activation Ordering by @kylesayrs in https://github.com/vllm-project/vllm/pull/8135 * [Misc] Fused MoE Marlin support for GPTQ by @dsikka in https://github.com/vllm-project/vllm/pull/8217 * Add NVIDIA Meetup slides, announce AMD meetup, and add contact info by @simon-mo in https://github.com/vllm-project/vllm/pull/8319 * [Bugfix] Fix missing `post_layernorm` in CLIP by @DarkLight1337 in https://github.com/vllm-project/vllm/pull/8155 * [CI/Build] enable ccache/scccache for HIP builds by @dtrifiro in https://github.com/vllm-project/vllm/pull/8327 * [Frontend] Clean up type annotations for mistral tokenizer by @DarkLight1337 in https://github.com/vllm-project/vllm/pull/8314 * [CI/Build] Enabling kernels tests for AMD, ignoring some of then that fail by @alexeykondrat in https://github.com/vllm-project/vllm/pull/8130 * Fix ppc64le buildkite job by @sumitd2 in https://github.com/vllm-project/vllm/pull/8309 * [Spec Decode] Move ops.advance_step to flash attn advance_step by @kevin314 in https://github.com/vllm-project/vllm/pull/8224 * [Misc] remove peft as dependency for prompt models by @prashantgupta24 in https://github.com/vllm-project/vllm/pull/8162 * [MISC] Keep chunked prefill enabled by default with long context when prefix caching is enabled by @comaniac in https://github.com/vllm-project/vllm/pull/8342 * [Bugfix] Ensure multistep lookahead allocation is compatible with cuda graph max capture by @alexm-neuralmagic in https://github.com/vllm-project/vllm/pull/8340 * [Core/Bugfix] pass VLLM_ATTENTION_BACKEND to ray workers by @SolitaryThinker in https://github.com/vllm-project/vllm/pull/8172 * [CI/Build][Kernel] Update CUTLASS to 3.5.1 tag by @tlrmchlsmth in https://github.com/vllm-project/vllm/pull/8043 * [Misc] Skip loading extra bias for Qwen2-MOE GPTQ models by @jeejeelee in https://github.com/vllm-project/vllm/pull/8329 * [Bugfix] Fix InternVL2 vision embeddings process with pipeline parallel by @Isotr0py in https://github.com/vllm-project/vllm/pull/8299 * [Hardware][NV] Add support for ModelOpt static scaling checkpoints. by @pavanimajety in https://github.com/vllm-project/vllm/pull/6112 * [model] Support for Llava-Next-Video model by @TKONIY in https://github.com/vllm-project/vllm/pull/7559 * [Frontend] Create ErrorResponse instead of raising exceptions in run_batch by @pooyadavoodi in https://github.com/vllm-project/vllm/pull/8347 * [Model][VLM] Add Qwen2-VL model support by @fyabc in https://github.com/vllm-project/vllm/pull/7905 * [Hardware][Intel] Support compressed-tensor W8A8 for CPU backend by @bigPYJ1151 in https://github.com/vllm-project/vllm/pull/7257 * [CI/Build] Excluding test_moe.py from AMD Kernels tests for investigation by @alexeykondrat in https://github.com/vllm-project/vllm/pull/8373 * [Bugfix] Add missing attributes in mistral tokenizer by @DarkLight1337 in https://github.com/vllm-project/vllm/pull/8364 * [Kernel][Misc] Add meta functions for ops to prevent graph breaks by @bnellnm in https://github.com/vllm-project/vllm/pull/6917 * [Misc] Move device options to a single place by @akx in https://github.com/vllm-project/vllm/pull/8322 * [Speculative Decoding] Test refactor by @LiuXiaoxuanPKU in https://github.com/vllm-project/vllm/pull/8317 * Pixtral by @patrickvonplaten in https://github.com/vllm-project/vllm/pull/8377 * Bump version to v0.6.1 by @simon-mo in https://github.com/vllm-project/vllm/pull/8379 ## New Contributors * @mmcelaney made their first contribution in https://github.com/vllm-project/vllm/pull/8161 * @elfiegg made their first contribution in https://github.com/vllm-project/vllm/pull/8173 * @Manikandan-Thangaraj-ZS0321 made their first contribution in https://github.com/vllm-project/vllm/pull/7860 * @sumitd2 made their first contribution in https://github.com/vllm-project/vllm/pull/8026 * @alugowski made their first contribution in https://github.com/vllm-project/vllm/pull/8060 * @vladislavkruglikov made their first contribution in https://github.com/vllm-project/vllm/pull/8292 * @kevin314 made their first contribution in https://github.com/vllm-project/vllm/pull/8224 * @TKONIY made their first contribution in https://github.com/vllm-project/vllm/pull/7559 * @akx made their first contribution in https://github.com/vllm-project/vllm/pull/8322 **Full Changelog**: https://github.com/vllm-project/vllm/compare/v0.6.0...v0.6.1