Skip to content
Computer vision

Hy4-Flash-WV-v6

run a massive multimodal model on a single GPU

2 min readupdated

This new 2.50 bpw quantized build of the popular tencent/Hy4-flash multimodal model is tailored specifically for Blackwell GPUs (SM120 architecture). While the original checkpoint exceeds 500 GB and requires a server cluster, this version shrinks the weights down to 52.5 GB so it can run on a single consumer card like an RTX 5090 (with VRAM limits) or an RTX PRO 6000. It supports text, images, reasoning chains, and tool calls—making it ideal for developers and hobbyists who want to test a heavy modern architecture locally.

Key features

  • Single-GPU execution: Compresses backbone weights to 52.51 GB while offloading heavy Engram data to host memory (mmap) without consuming VRAM.
  • Speculative decoding: A built-in DFly2 (0.85B) draft model speeds up generation to 55–229 tokens per second.
  • Multimodal support: Full image processing capabilities powered by an integrated SigLIP2 ViT.
  • OpenAI API compatibility: Ships with a built-in server for quick deployment and seamless web UI integration.

What you can do

  • Spin up a local AI server on an RTX 5090 using a dedicated VRAM optimization flag.
  • Test structured JSON outputs alongside streaming reasoning chains (reasoning_effort).
  • Benchmark the quantized build against the original checkpoint using built-in memory and accuracy tests.

Example

pip install wheels/qfn_sm120_unified-22.3.0-cp313-cp313-linux_x86_64.whl

# Запуск на потребительской карте RTX 5090 с ограничением памяти:
python -m qfn_sm120_unified.serve --ckpt . --host 0.0.0.0 --port 8000 --force-32gb --inline-think
Read moreDetails
• Backbone size: 52.51 GB (117.19 GB total on disk)
• Hardware requirements: 1x 96GB GPU or RTX 5090 (32 GB)
• Generation speed: up to 229.54 tokens/s (at batch size 32)
• License: Tencent Hy License
Discuss Telegram 0 16

Will it run?

Runs in BF16

Fits without quantisation.

  • BF16weights 15 GB · ready files17 GB
  • FP8 / INT8recommendedweights 7.5 GB8.8 GB
  • 4-bitweights 4.2 GB5.2 GB

Tip: go 8-bit.FP8 / INT8 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

Your own GPU: FP8 / INT8 — from RTX 3080 · BF16 — from RTX 4090

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.

    # Grawnch/Hy4-Flash-WV-v6-2.50bpw-SM120
    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 Grawnch/Hy4-Flash-WV-v6-2.50bpw-SM120 --local-dir ~/models/Hy4-Flash-WV-v6-2.50bpw-SM120
    # 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-flash (quants, finetunes, LoRAs) will show up under the bell.

Quantised versions 12All on Hugging Face
Fine-tunes 12All on Hugging Face

About the model

Grawnch/Hy4-Flash-WV-v6-2.50bpw-SM120
Author
Grawnch
Category
Computer vision
Released
Posted

On NeyroLearn

Hugging Face

160

Ask AI about this model

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

Comments