vLLM 首日支持 Reflection Beam
AI 总结
vLLM 博客发布了对 Reflection 的 Beam 模型的首日推理支持，并在 NVIDIA Blackwell 上做了专门优化。Beam 是 501B 参数的 MoE 模型，每 token 激活 23B 参数；vLLM 表示推理、工具调用和结构化输出开箱可用，并支持张量并行、数据并行与专家并行，同时提供了 Docker 镜像和启动命令。文章称编码智能体自主优化了实现，使 batch-1 解码比最初版本快 2.71 倍、prefill 快 1.87 倍；使用 Beam 的 DFlash2 草稿头时，TP8 下单请求达到 708 tokens/s，接受长度为 3.7。vLLM 还与 SkyRL 团队合作提供首日强化学习支持，文中称在 50 个训练步后 AIME25 平均准确率从约 61% 升至约 79%。
深度见解
对需要自行部署大型开放权重 MoE 的团队而言，vLLM 称 Beam 在发布首日即可获得推理、工具调用与 Blackwell 内核优化支持，并附带 SkyRL 的强化学习路径。
TL;DR: Reflection's Beam is a 501B-parameter MoE with 23B active parameters per token. vLLM supports it from day 0 and optimizes it on NVIDIA Blackwell:
- Full feature support. Reasoning, tool calling, and structured outputs work out of the box, and you can run the model with tensor, data, or expert parallelism.
- Agent-driven optimization. Coding agents optimized the model autonomously, making batch-1 decode 2.71× faster and prefill 1.87× faster than our first implementation.
- DFlash2 with adaptive verification. With Beam's DFlash2 drafter, a single request reaches 708 tokens/s on TP8, with an acceptance length of 3.7. Adaptive verification, which decides each step how many draft tokens are worth verifying, adds 10% more throughput at high concurrency.
To try it, use our Docker image vllm/vllm-openai:reflection-beam:
docker run --gpus all --ipc=host -p 8000:8000 \
-v ~/.cache/huggingface:/root/.cache/huggingface \
vllm/vllm-openai:reflection-beam \
reflection/Beam-1.0-501B-A23B-NVFP4 \
-tp 4 \
--max-num-seqs 128 \
--reasoning-parser beam \
--tool-call-parser beam --enable-auto-tool-choice \
--kv-cache-dtype fp8 \
--attention-config '{"backend":"FLASHINFER"}' \
--kernel-config '{"enable_flashinfer_autotune":false}' \
--speculative-config '{
"method": "dflash",
"model": "reflection/Beam-1.0-501B-A23B-DFlash2",
"num_speculative_tokens": 7,
"attention_backend": "FLASH_ATTN"
}'#Reflection#Beam#vLLM#NVIDIA#Inferact#SkyRL#Blackwell#GB300#FlashInfer#DeepSeek-V3