Models

vLLM 首日支持 Tencent Hunyuan Hy4 flash,1M 上下文可部署

Why we picked it

vLLM 发布当天就支持了腾讯混元的 Hy4 flash,224B 参数只激活 12B,还有 1M 上下文和多模态。有 NVIDIA GPU 的可以直接拉官方 Docker 镜像跑起来,想做长文档解析或 computer use agent 的别错过部署配方。

vLLM 宣布对腾讯混元 Hy4 flash 提供 Day-0 支持,用户可在 NVIDIA GPU 上直接通过 vLLM 部署该模型。官方提供 Docker 镜像 vllm/vllm-openai:hy4-flash 和部署配方,集成工作由 @inferact 完成。

Body

Hy4 flash 总参数量 224B,其中包含 64B Engram,激活参数仅 12B。架构上采用混合注意力、Latent MoE 和 Engram 机制,主打推理效率。模型支持 1M token 上下文窗口,具备多模态理解能力,适用场景包括科研、文档与幻灯片解析以及计算机操作(computer use)。

对开发者来说,这意味着一个高上下文、多模态的开源友好模型可以零等待接入 vLLM 生态,与 DeepSeek、Qwen 等已有 vLLM 支持的中国模型形成同一路径。依赖长文档处理或 agent 式计算机操作的团队可以直接用官方镜像起步。

Original · vLLM

🎉 Day-0 support for @TencentHunyuan’s Hy4 flash is here in vLLM!

A 1M-token context and multimodal understanding for research, documents, slides and computer use.

224B total parameters, including 64B Engram, with 12B active. Hybrid attention, Latent MoE and Engram put efficiency at the center of the architecture.

Run Hy4 flash on @NVIDIA GPUs with vLLM, with integration work by @inferact. Get started with the release image and deployment recipe:

Docker image: vllm/vllm-openai:hy4-flash Recipe: https://t.co/J3SSwNPj8p

Industry Coordinates & Source Verification

This report is categorized under AITOP's "Models" desk, focusing on China's AI industry and domestic foundation model ecosystem. Within our cross-regional analytical framework, it directly maps to AI developer toolchains, coding agents, and engineering workflows and US-China foundation model benchmarks and commercialization comparison. Key entities and technical topics under continuous tracking include Hunyuan, vLLM, open-source models, multimodal.

Source attribution: curated from vLLM (@vllm_project).