add release note for vLLM 0.9.0 release #802
There are no files selected for viewing
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
| Original file line number | Diff line number | Diff line change |
|---|---|---|
| @@ -0,0 +1,198 @@ | ||
| # Optimize LLM serving with vLLM on Intel® GPUs | ||
|
|
||
| vLLM is a fast and easy-to-use library for LLM inference and serving. It has evolved into a community-driven project with contributions from both academia and industry. Intel, as one of the community contributors, is working actively to bring satisfying performance with vLLM on Intel® platforms, including Intel® Xeon® Scalable Processors, Intel® discrete GPUs, as well as Intel® Gaud® AI accelerators. This blog focuses on Intel® discrete GPUs at this time and brings you the necessary information to get the workloads running well on your Intel® graphics cards. | ||
|
There was a problem hiding this comment. Gaud® -> Gaudi® |
||
|
|
||
| The vLLM used in the latest docker image is based on [v0.9.0](https://github.com/vllm-project/vllm/tree/v0.9.0) | ||
|
There was a problem hiding this comment. latest -> this because it will be no longer 'latest' along with the new releases. |
||
|
|
||
| ## 1. What's Supported? | ||
|
|
||
| Intel GPUs benefit from enhancements brought by [vLLM V1 engine](https://blog.vllm.ai/2025/01/27/v1-alpha-release.html), including: | ||
|
|
||
| * Optimized Execution Loop & API Server | ||
| * Simple & Flexible Scheduler | ||
| * Zero-Overhead Prefix Caching | ||
| * Clean Architecture for Tensor-Parallel Inference | ||
| * Efficient Input Preparation | ||
|
|
||
| Besides, following up vLLM V1 design, corresponding optimized kernels are implemented for Intel GPUs. | ||
|
|
||
| * chunked_prefill: | ||
|
|
||
| chunked_prefill is an optimization feature in vLLM that allows large prefill requests to be divided into small chunks and batched together with decode requests. This approach prioritizes decode requests, improving inter-token latency (ITL) and GPU utilization by combining compute-bound (prefill) and memory-bound (decode) requests in the same batch. vLLM v1 engine is built on this feature and in this release, it's also supported on intel GPUs by leveraging corresponding kernel from Intel® Extension for PyTorch\* for model execution. | ||
|
|
||
| * FP8 W8A16: | ||
|
|
||
| vLLM supports FP8 (8-bit floating point) weight using hardware acceleration on GPUs. We support weight-only online dynamic quantization with FP8, which allows for a 2x reduction in model memory requirements and up to a 1.6x improvement in throughput with minimal impact on accuracy. | ||
|
|
||
| Dynamic quantization of an original precision BF16/FP16 model to FP8 can be achieved with vLLM without any calibration data required. You can enable the feature by specifying `--quantization="fp8"` in the command line or setting `quantization="fp8"` in the LLM constructor. | ||
|
|
||
| Besides, the FP8 types typically supported in hardware have two distinct representations, each useful in different scenarios: | ||
|
|
||
| * **E4M3**: Consists of 1 sign bit, 4 exponent bits, and 3 bits of mantissa. It can store values up to +/-448 and `nan`. | ||
| * **E5M2**: Consists of 1 sign bit, 5 exponent bits, and 2 bits of mantissa. It can store values up to +/-57344, +/- `inf`, and `nan`. The tradeoff for the increased dynamic range is lower precision of the stored values. | ||
|
|
||
| We support both representations through ENV variable `VLLM_XPU_FP8_DTYPE` with default value `E5M2`. | ||
|
|
||
| :::{warning} | ||
| Currently, by default we load the model at original precision before quantizing down to 8-bits, so you need enough memory to load the whole model. To avoid this, adding `VLLM_OFFLOAD_WEIGHTS_BEFORE_QUANT=1` can allow offloading weights to cpu before quantization and quantized weights will be kept in device. | ||
|
|
||
| ::: | ||
|
|
||
| * Multi Modility Support for Qwen2.5-VL Models | ||
|
|
||
|
|
||
| In this release, image/audio input can be processed using Qwen2.5-VL Models, like Qwen/Qwen2.5-VL-32B-Instruct on 4 BMG cards. | ||
|
|
||
| Besides, we also have some experimental features supported, including: | ||
|
|
||
| * **pipeline parallelism**: Works on on single node as only backend `mp` is supported for now. | ||
| * **torch.compile**: Can be enabled for both FP16 and online FP8 quantization path. | ||
| * **speculative decoding**: Supports methods `n-gram`, `EAGLE` and `EAGLE3`. | ||
|
|
||
| ## Optimizations | ||
|
|
||
| * tensor parallel inference: Intel® oneAPI Collective Communications Library(oneCCL) is optimized to provide boosted performance in Intel® Arc™ B-Series graphics cards. | ||
|
|
||
| * GQA kernel optimization: An optimized version of Grouped-Query Attention(GQA) kernel is adopted and obvious perf improvement is observed in models like Qwen and Llama. | ||
|
There was a problem hiding this comment. 10-16% end-to-end throughput improvement for 8B/14B/32B fp8 workloads in 1024/512 input/output lengths. |
||
| * Other: long context length (>4k) optimization for output token latency, which brings 1.8x perf gain on next token for 40K seq length, 1.6x for 20K, 1.4x for 12K. | ||
|
|
||
| ## Supported Models | ||
|
|
||
| The table below lists models that have been verified by Intel. However, there should be broader models that are supported by vLLM work on Intel® GPUs. | ||
|
|
||
| | Model Type | Model (company/model name) | Dynamic Online FP8 | | ||
|
There was a problem hiding this comment. We have support for fp16 and w4a8 model list as well: There was a problem hiding this comment. Hi @rogerxfeng8, one open here, the int4 models we tested are not public but get from ccg. How can we add them here? |
||
| | ---------- | -------------------------- | --- | | ||
| | Text Generation | deepseek-ai/DeepSeek-R1-Distill-Llama-8B |✅︎| | ||
| | Text Generation | deepseek-ai/DeepSeek-R1-Distill-Qwen-14B |✅︎| | ||
| | Text Generation | deepseek-ai/DeepSeek-R1-Distill-Qwen-32B |✅︎| | ||
| | Text Generation | deepseek-ai/DeepSeek-R1-Distill-Llama-70B |✅︎| | ||
| | Text Generation | Qwen/Qwen2.5-72B-Instruct |✅︎| | ||
| | Text Generation | Qwen/Qwen3-32B |✅︎| | ||
| | Text Generation | Qwen/Qwen3-30B-A3B |✅︎| | ||
| | Text Generation | meta-llama/Llama-3.1-8B-Instruct |✅︎| | ||
| | Text Generation | baichuan-inc/Baichuan2-13B-Chat |✅︎| | ||
| | Text Generation | THUDM/GLM-4-9B-chat |✅︎| | ||
| | Text Generation | THUDM/CodeGeex4-All-9B |✅︎| | ||
| | Text Generation | chuhac/TeleChat2-35B |✅︎| | ||
| | Text Generation | 01-ai/Yi1.5-34B-Chat |✅︎| | ||
| | Text Generation | THUDM/CodeGeex4-All-9B |✅︎| | ||
| | Text Generation | deepseek-ai/DeepSeek-Coder-33B-base |✅︎| | ||
| | Multi Modality | Qwen/Qwen2.5-VL-72B-Instruct |✅︎| | ||
| | Multi Modality | Qwen/Qwen2.5-VL-32B-Instruct |✅︎| | ||
|
|
||
| ## 2. Limitations | ||
|
|
||
| Some of vLLM V1 features may need extra support, including LoRA(Low-Rank Adaptation), pipeline parallel on Ray, EP(Expert Parallelism)/TP(Tensor Parallelism) MoE(Mixture of Experts), DP(Data Parallelism) Attention and MLA(Multi-head Latent Attention). | ||
|
|
||
| The following issues are known issues: | ||
|
|
||
| * Memory reservation increases in vLLM 0.9.0 and it may cause OOM to multi-modility models like Qwen/Qwen2-VL-7B-Instruct, Qwen/Qwen2.5-VL-72B-Instruct and Qwen/Qwen2.5-VL-32B-Instruct. We need downgrade `gpu-memory-utilization` from default value `0.9` to `0.85`. | ||
|
|
||
| * W8A8 quantized models through llm_compressor are not supported yet, like RedHatAI/DeepSeek-R1-Distill-Qwen-32B-FP8-dynamic. | ||
|
|
||
| ## 3. How to Get Started | ||
|
|
||
| ### 3.1. Prerequisite | ||
|
|
||
| | OS | Hardware | | ||
| | ---------- | ---------- | | ||
| | Ubuntu 25.04 | Intel® Arc™ B-Series | | ||
|
|
||
| ### 3.2. Prepare a Serving Environment | ||
|
|
||
| 1. Get the released docker image with command `docker pull intel/vllm:0.9.0-xpu` | ||
| 2. Instantiate a docker container with command `docker run -t -d --shm-size 10g --net=host --ipc=host --privileged -v /dev/dri/by-path:/dev/dri/by-path --name=vllm-test --device /dev/dri:/dev/dri --entrypoint= intel/vllm:0.9.0-xpu /bin/bash` | ||
| 3. Source openapi envs to ensure correct variables set with command `docker exec vllm-test /bin/bash -c "source /opt/intel/oneapi/setvars.sh --force"` | ||
| 4. Run command `docker exec -it vllm-test bash` in 2 separate terminals to enter container environments for the server and the client respectively. | ||
|
|
||
| \* Starting from here, all commands are expected to be run inside the docker container, if not explicitly noted. | ||
|
|
||
| In both environments, you may then wish to set a `HUGGING_FACE_HUB_TOKEN` environment variable to make sure necessary files can be downloaded from the HuggingFace website. | ||
|
|
||
| ```bash | ||
| export HUGGING_FACE_HUB_TOKEN=xxxxxx | ||
| ``` | ||
|
|
||
| ### 3.3. Launch Workloads | ||
|
|
||
| #### 3.3.1. Launch Server in the Server Environment | ||
|
|
||
| Command: | ||
|
|
||
| ```bash | ||
| TORCH_LLM_ALLREDUCE=1 VLLM_USE_V1=1 VLLM_WORKER_MULTIPROC_METHOD=spawn python3 -m vllm.entrypoints.openai.api_server --model deepseek-ai/DeepSeek-R1-Distill-Qwen-32B --dtype=float16 --device=xpu --enforce-eager --port 8000 --block-size 64 --gpu-memory-util 0.9 --no-enable-prefix-caching --trust-remote-code --disable-sliding-window --disable-log-requests --max_num_batched_tokens=8192 --max_model_len 4096 -tp=4 --quantization fp8 | ||
| ``` | ||
|
|
||
| Note that by default fp8 online quantization will use `e5m2` and you can switch to use `e4m3` by explicitly add env `VLLM_XPU_FP8_DTYPE=e4m3`. If there is not enough memory to hold the whole model before quantization to fp8, you can use `VLLM_OFFLOAD_WEIGHTS_BEFORE_QUANT=1` to offload weights to cpu first. | ||
|
|
||
| Expected output: | ||
|
|
||
| ```bash | ||
| INFO 02-20 03:20:29 api_server.py:937] Starting vLLM API server on http://0.0.0.0:8000 | ||
| INFO 02-20 03:20:29 launcher.py:23] Available routes are: | ||
| INFO 02-20 03:20:29 launcher.py:31] Route: /openapi.json, Methods: HEAD, GET | ||
| INFO 02-20 03:20:29 launcher.py:31] Route: /docs, Methods: HEAD, GET | ||
| INFO 02-20 03:20:29 launcher.py:31] Route: /docs/oauth2-redirect, Methods: HEAD, GET | ||
| INFO 02-20 03:20:29 launcher.py:31] Route: /redoc, Methods: HEAD, GET | ||
| INFO 02-20 03:20:29 launcher.py:31] Route: /health, Methods: GET | ||
| INFO 02-20 03:20:29 launcher.py:31] Route: /ping, Methods: POST, GET | ||
| INFO 02-20 03:20:29 launcher.py:31] Route: /tokenize, Methods: POST | ||
| INFO 02-20 03:20:29 launcher.py:31] Route: /detokenize, Methods: POST | ||
| INFO 02-20 03:20:29 launcher.py:31] Route: /v1/models, Methods: GET | ||
| INFO 02-20 03:20:29 launcher.py:31] Route: /version, Methods: GET | ||
| INFO 02-20 03:20:29 launcher.py:31] Route: /v1/chat/completions, Methods: POST | ||
| INFO 02-20 03:20:29 launcher.py:31] Route: /v1/completions, Methods: POST | ||
| INFO 02-20 03:20:29 launcher.py:31] Route: /v1/embeddings, Methods: POST | ||
| INFO 02-20 03:20:29 launcher.py:31] Route: /pooling, Methods: POST | ||
| INFO 02-20 03:20:29 launcher.py:31] Route: /score, Methods: POST | ||
| INFO 02-20 03:20:29 launcher.py:31] Route: /v1/score, Methods: POST | ||
| INFO 02-20 03:20:29 launcher.py:31] Route: /v1/audio/transcriptions, Methods: POST | ||
| INFO 02-20 03:20:29 launcher.py:31] Route: /rerank, Methods: POST | ||
| INFO 02-20 03:20:29 launcher.py:31] Route: /v1/rerank, Methods: POST | ||
| INFO 02-20 03:20:29 launcher.py:31] Route: /v2/rerank, Methods: POST | ||
| INFO 02-20 03:20:29 launcher.py:31] Route: /invocations, Methods: POST | ||
| INFO: Started server process [1636943] | ||
| INFO: Waiting for application startup. | ||
| INFO: Application startup complete. | ||
| ``` | ||
|
|
||
| It may take some time. Showing `INFO: Application startup complete.` indicates that the server is ready. | ||
|
|
||
| #### 3.3.2. Raise Requests for Benchmarking in the Client Environment | ||
|
|
||
| We leverage a [benchmarking script](https://github.com/vllm-project/vllm/blob/main/benchmarks/benchmark_serving.py) which is provided in vLLM to perform performance benchmarking. You can use your own client scripts as well. | ||
|
|
||
| Use the command below to shoot serving requests: | ||
|
|
||
| ```bash | ||
| python3 benchmarks/benchmark_serving.py --model deepseek-ai/DeepSeek-R1-Distill-Qwen-32B --dataset-name random --random-input-len=1024 --random-output-len=1024 --ignore-eos --num-prompt 1 --max-concurrency 16 --request-rate inf --backend vllm --port=8000 --host 0.0.0.0 | ||
| ``` | ||
|
|
||
| The command uses model `deepseek-ai/DeepSeek-R1-Distill-Qwen-32B`. Both input and output token sizes are set to `1024`. Maximally `16` requests are processed concurrently in the server. | ||
|
|
||
| Expected output: | ||
|
|
||
| ```bash | ||
| Maximum request concurrency: 16 | ||
| ============ Serving Benchmark Result ============ | ||
| Successful requests: 1 | ||
| Benchmark duration (s): xxx | ||
| Total input tokens: 1024 | ||
| Total generated tokens: 1024 | ||
| Request throughput (req/s): xxx | ||
| Output token throughput (tok/s): xxx | ||
| Total Token throughput (tok/s): xxx | ||
| ---------------Time to First Token---------------- | ||
| Mean TTFT (ms): xxx | ||
| Median TTFT (ms): xxx | ||
| P99 TTFT (ms): xxx | ||
| -----Time per Output Token (excl. 1st token)------ | ||
| Mean TPOT (ms): xxx | ||
| Median TPOT (ms): xxx | ||
| P99 TPOT (ms): xxx | ||
| ---------------Inter-token Latency---------------- | ||
| Mean ITL (ms): xxx | ||
| Median ITL (ms): xxx | ||
| P99 ITL (ms): xxx | ||
| ================================================== | ||
| ``` | ||
|
|
||
| ## 5. Need Assistance? | ||
|
|
||
| Should you encounter any issues or have any questions, please submit an issue ticket at [vLLM Github Issues](https://github.com/vllm-project/vllm/issues). Include the text `[Intel GPU]` in the issue title to ensure it gets noticed. | ||
Oops, something went wrong.
Oops, something went wrong.
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Uh oh!
There was an error while loading. Please reload this page.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
We don't update the history release notes. just need to add this 0.9.0-xpu.md