Skip to content
Other

MiniMax-H3-Prompt-Rewriter-Qwen3.5-9B-Ref2VA-Q3_K_M-GGUF

a local prompt rewriter in GGUF format

2 min readupdated

This is a GGUF build (a format for running models locally on CPUs and GPUs) of lightx2v/MiniMax-H3-Prompt-Rewriter-Qwen3.5-9B-Ref2VA, converted by Bakanayatsu using llama.cpp. The original model is built for rewriting and enhancing prompts—judging by the name, for Ref2VA tasks in the MiniMax-H3 lineup based on the 9B Qwen3.5 model. Q3_K_M quantization (3-bit weight compression) significantly lowers memory requirements compared to the full-size checkpoint. It is a solid fit if you want to run a prompt generator and enhancer locally without heavy server hardware.

Key features

  • Q3_K_M quantization — a compact 3-bit variant of the 9B model designed to run on consumer PCs.
  • GGUF format with native support for llama.cpp, llama-cli, and llama-server.
  • Direct loading from the Hugging Face repo via the --hf-repo and --hf-file flags with no manual download required.
  • Converted using the official GGUF-my-repo tool from ggml.ai.

What you can do

  • Spin up a local API server with llama-server and plug it into generation pipelines as an automated prompt-enhancement step.
  • Run it in CLI mode via llama-cli to quickly refine text prompts offline.

Example

llama-cli --hf-repo Bakanayatsu/MiniMax-H3-Prompt-Rewriter-Qwen3.5-9B-Ref2VA-Q3_K_M-GGUF \
  --hf-file minimax-h3-prompt-rewriter-qwen3.5-9b-ref2va-q3_k_m.gguf \
  -p "The meaning to life and the universe is"
Read moreDetails
Base model: lightx2v/MiniMax-H3-Prompt-Rewriter-Qwen3.5-9B-Ref2VA.
Format and quantization: GGUF, Q3_K_M (converted via llama.cpp / GGUF-my-repo).
Detailed architecture notes, benchmarks, and system prompts can be found on the original model card.
Discuss Telegram 0 115

Will it run?

Runs in GGUF Q3_K_M

The most precise of the ready quantised files fits.

  • GGUF Q3_K_Mweights 4.4 GB · ready files5.4 GB

What to get for this model

Massed ComputeOpen Massed Compute

Your own GPU: GGUF Q3_K_M — from RTX 2060

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.

    # Bakanayatsu/MiniMax-H3-Prompt-Rewriter-Qwen3.5-9B-Ref2VA-Q3_K_M-GGUF
    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 Bakanayatsu/MiniMax-H3-Prompt-Rewriter-Qwen3.5-9B-Ref2VA-Q3_K_M-GGUF --local-dir ~/models/MiniMax-H3-Prompt-Rewriter-Qwen3.5-9B-Ref2VA-Q3_K_M-GGUF
    # 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.

MiniMax H3 prompts

All prompts

Ready prompts with the resulting video: copy and try.

Model family

New posts about versions of MiniMax-H3-Prompt-Rewriter-Qwen3.5-9B-Ref2VA (quants, finetunes, LoRAs) will show up under the bell.

Quantised versions 3All on Hugging Face

About the model

Bakanayatsu/MiniMax-H3-Prompt-Rewriter-Qwen3.5-9B-Ref2VA-Q3_K_M-GGUF
Author
Bakanayatsu
Category
Other
Released
Posted

On NeyroLearn

Hugging Face

1150

Ask AI about this model

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

Comments