Skip to content
Other

Huihui-Qwen3.8-27B-Comfy

quantized text encoders for ComfyUI

2 min readupdated

This repository provides ready-to-use, single-file ComfyUI text encoders for the Huihui-Qwen3.8-27B-abliterated model, repacked from the original Hugging Face checkpoint and quantized into five precision formats. While the full original model spans 18 shards at roughly 52 GB, these files are optimized into standalone encoders. They are built for anyone running complex ComfyUI workflows who wants to save disk space when loading heavy text encoders.

Key features

  • Five quantization options: Available in int8_convrot, w6a8, w4a8, mxfp8, and nvfp4 formats with varying compression ratios and quality levels.
  • ComfyUI compatibility: Each file is bundled into a single .safetensors package for quick drop-in use without manually stitching shards together.
  • Selective quantization: Only the language model, lm_head, and MTP head are quantized, while the vision components and embedding table remain in BF16.
  • Hardware optimization: Encoders unpack to BF16 on older GPUs, while Blackwell and newer architectures take advantage of native FP4, FP8, and INT8 execution paths.

What you can do

  • Install in ComfyUI: Drop your chosen file into the ComfyUI/models/text_encoders/ folder to use it in your pipelines.
  • Strike the right balance: Run the int8_convrot variant for the best trade-off between file size and output quality.
  • Save disk space: Go with the lightest nvfp4 version (around 18 GB) if you are short on storage.

Example

Place the downloaded file (for example, huihui-qwen38-27b-abl-int8_convrot.safetensors) into your ComfyUI/models/text_encoders/ directory and select it as the text encoder in your workflow.

Read moreDetails
Base checkpoint: huihui-ai/Huihui-Qwen3.8-27B-abliterated
File sizes: 18.12 GB (nvfp4) to 55.56 GB (bf16)
PPL (perplexity) score: 4.0737 (w6a8) to 4.1585 (nvfp4)
Quantized modules: model.language_model.*, lm_head.*, and MTP draft head
Discuss Telegram 1

Will it run?

Runs in 4-bit

Full precision does not fit, a quantised version does: quality barely changes down to 8 bits, more noticeably at 4 bits.

  • BF16weights 52 GB · ready files57 GB
  • FP8 / INT8recommendedweights 26 GB29 GB
  • 4-bitweights 15 GB17 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.

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.

    # eastmoe/Huihui-Qwen3.8-27B-Comfy
    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 eastmoe/Huihui-Qwen3.8-27B-Comfy --local-dir ~/models/Huihui-Qwen3.8-27B-Comfy
    # 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 Huihui-Qwen3.8-27B-Comfy (quants, finetunes, LoRAs) will show up under the bell.

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

About the model

eastmoe/Huihui-Qwen3.8-27B-Comfy
Author
eastmoe
Category
Other
Released
Posted

On NeyroLearn

Hugging Face

01

Ask AI about this model

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

Comments