Skip to content
Other

FlagRelease/Hy4-W8A8-iluvatar-FlagOS

Tencent Hunyuan multimodal model for Iluvatar chips

2 min readupdated

Tencent Hunyuan has officially open-sourced its Hy4 multimodal model, and the FlagOS community has performed cross-hardware adaptation and prepared a W8A8 quantized version. The model is tailored for real-world performance scenarios, supporting visual understanding, computer use, complex task execution, and long-text processing. Built on the FlagOS 2.2 stack, the version is optimized to run on alternative hardware platforms while delivering high inference speed. The release is intended for developers and engineers deploying large multimodal models on heterogeneous hardware.

Key features

  • Cross-hardware adaptation: Official integration of the Hy4 model into the FlagOS ecosystem with multi-vendor support.
  • Performance optimization: Using the vLLM combination and the FlagGems operator library to accelerate computations.
  • W8A8 quantization: Lower memory requirements and higher throughput while preserving generation quality.
  • Multimodality: Full support for text and visual tasks, including the analysis of complex charts and interfaces.

What you can do

  • Deploy an inference server: Launch a distributed cluster based on 16 Iluvatar BI-V150 cards using a ready-made Docker image.
  • Integrate into chatbots: Connect the model to external client applications and services via the vLLM OpenAI-compatible API.
  • Process images: Send multimodal prompts with graphical data for visual analysis and image-based answers.

Example

curl http://localhost:8400/v1/chat/completions \
  -H "Content-Type: application/json" \
  -d '{
    "model": "hy4",
    "messages": [{"role": "user", "content": "中国的首都是哪个城市?"}],
    "max_tokens": 64,
    "temperature": 0
  }'
Read moreDetails
• Architecture: HyV4VLForConditionalGeneration (246 shards, 793 GB)
• Hardware: 2 nodes × 16 Iluvatar BI-V150 cards (32 GB)
• Configuration: Tensor Parallelism 8, Pipeline Parallelism 4
• Limitations: it is critically important to use the PIECEWISE graph mode and not clear the /root/.triton cache
Discuss Telegram 0

Will it run?

Won't run

Not enough memory even for 4-bit. Below are GPUs you can rent.

  • I8recommendedweights 735 GB · ready files808 GB
  • 4-bitweights 406 GB447 GB

Tip: go 8-bit.I8 is within about a percent of full precision in quality, takes half the memory and runs faster on RTX 40/50 and H100. Full precision only pays off for fine-tuning.

What to get for this model

Massed ComputeOpen Massed Compute

No single instance holds the whole model: it needs RAM offload or a cluster.

Your own GPU: Won't fit any home GPU — renting is cheaper.

Massed Compute hourly prices, November 2026. The discounted price applies with the coupon at checkout.

Estimate: weights plus working memory. Long videos, high resolutions and long contexts need more.

Run it on Massed Compute

The script downloads the model to the rented machine. How to run it is in the model card.

  1. 1. Rent a machine with a fitting GPU

    Open Massed Compute
  2. 2. Paste the script into the machine's terminal

    Connect over SSH (the Massed Compute panel shows the command) or open a terminal on the machine's desktop.

    HF_TOKEN / CIVITAI_TOKENThe script starts with token lines: paste yours between the quotes. The Hugging Face token is for gated models, the Civitai one for Civitai models.

    # FlagRelease/Hy4-W8A8-iluvatar-FlagOS
    mkdir -p ~/models
    curl -LsSf https://astral.sh/uv/install.sh | sh
    source "$HOME/.local/bin/env"
    uv tool install "huggingface_hub[cli]"
    hf download FlagRelease/Hy4-W8A8-iluvatar-FlagOS --local-dir ~/models/Hy4-W8A8-iluvatar-FlagOS
    # The model is in ~/models. How to run it is in the model card (link above).
  3. 3. Use it

    The model is in ~/models. How to run it is in the model card (link above).

Stop the machine in the panel when you are done: it bills by the minute while it runs.

How to speed it up

Picked for this model type. Choose your hardware above to see what fits you exactly.

  • FlashAttention 44.0.0b40from November 18, 2026new release

    fastest exact attention on B200

    Kernels for Blackwell (beta). Exact attention without quantisation — fine for everything, LLMs included.

    How to enable: pip install flash-attn-4; vLLM and SGLang pick it up themselves.

    Earlier generations (2)
    • FlashAttention 3 — The Hopper (H100/H200) version with FP8.
    • FlashAttention 2 — The standard for RTX 30/40/50 and A100: exact attention without the memory cost of long contexts.
  • torch.compile2.15.0from October 28, 2026new release

    10–30% for free

    Compiles the model for your GPU. Slower first run, faster afterwards.

    How to enable: model = torch.compile(model); in ComfyUI the TorchCompileModel node.

Model family

New posts about versions of Hy4-W8A8-iluvatar-FlagOS (quants, finetunes, LoRAs) will show up under the bell.

No known relatives of this model yet. All derivatives on Hugging Face

About the model

FlagRelease/Hy4-W8A8-iluvatar-FlagOS
Author
FlagRelease
Category
Other
Released
Posted

On NeyroLearn

Hugging Face

00

Ask AI about this model

Answered by Gemini 4 · the chat is kept only in this tab

Comments