Skip to content

vLLM Adds Day-One Support for Tencent Hunyuan's Hy4 Flash

vLLM · XFrom the source· @vllm_project·· 3 hours agoPicks

547 views · 5 likes · 1 reposts (at collection)

Original

What happened

vLLM adds day-one support for Tencent Hunyuan's Hy4 flash, with a Docker image and deployment recipe for running it on NVIDIA GPUs. The post describes Hy4 flash as a 224B-parameter model, including 64B Engram, with 12B active, a 1M-token context, and multimodal understanding for research, documents, slides, and computer use. Integration work was done by Inferact.

Why it matters

vLLM's day-one support gives teams a deployment path for Hy4 flash on NVIDIA GPUs, with a release image and recipe covering a model that combines a 1M-token context with multimodal understanding.

Article

🎉 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: recipes.vllm.ai/tencent/Hy4-f…

Quoted postTencent Hy@TencentHunyuan

Hy4 flash is now open-weight, including Hy4 flash-FP8 and Hy4 flash-Base.

224B (12B active), 1M context, natively multimodal.
Fast, efficient and built for everyday productivity.

It's yours now. And stay tuned for Hy4 coming up next!

🤗: huggingface.co/tencent/Hy4-fl…
🐙: github.com/Tencent-Hunyuan/Hy…
📄: hy.tencent.ai/research/hy4-fl…

Released under the Tencent Hy License.
See full terms here: huggingface.co/tencent/Hy4-fl…

View the quoted post on X

The story so far

vLLM has added day-one support for Tencent Hunyuan's Hy4 Flash model series, providing a Docker image and deployment recipes for running the models on NVIDIA GPUs. The integration work was carried out by Inferact. Tencent Hunyuan simultaneously made the Hy4 Flash model series available under the Tencent Hy License. The release includes FP8 and Base variants. Tencent lists the model as having 224 billion total parameters, 12 billion active parameters, a 1 million token context window, and native multimodality, with weights and research details published on Hugging Face, GitHub, and Tencent's research portal.

How the story developed · 2

View full event
  1. 11/18 13:23 (UTC+8)

    vLLM Adds Day-One Support for Tencent Hunyuan's Hy4 Flash This report

  2. 11/18 13:23 (UTC+8)
    Tencent Makes Hy4 Flash Open-Weight in FP8 and Base VariantsTencent Hunyuan

Read next

See the full storyvLLM Adds Day-One Support for Tencent Hunyuan Hy4 Flash Models