Skip to content
Other

Sloth-27B-v1

26B model for routing and decision-making

2 min readupdated

Sloth-27B-v1 by cjqhhh is a specialized model for making structured decisions from text or JSON context. Instead of generating free-form text, it picks one of the predefined options and returns a probability distribution across all candidates. Built on vllm-sr/d3, the model packs around 26.09B parameters and features a dedicated FP32 classification head (decision head). It is built for developers tackling query routing, relevance scoring, and business-rule automation without risking output formatting hallucinations.

Key features

  • No free-form generation: skips extra text and strictly outputs a typed decision (selecting from up to 255 options or returning a binary answer) alongside probabilities.
  • Batch question handling: lets you ask multiple questions against a single shared state (context) in one batch.
  • Flexible input format: accepts context as plain text or a JSON object paired with instructions and criteria for each option.
  • Text-only focus: vision weights inherited from the base model were not tested for this release.

What you can do

  • Support ticket routing: automatically dispatch incoming tickets and user messages across departments (billing, tech support, sales) based on set criteria.
  • Filtering and relevance scoring: verify whether documents or responses match specific rules via binary checks (noul) with truth-probability scores.
  • AI agent orchestration: deploy the model as a fast, strict router to pick the next step in multi-stage data processing workflows.

Example

from transformers import AutoModel

model = AutoModel.from_pretrained("./Sloth-27B-v1", trust_remote_code=True, device="cuda:0")
res = model.system_one(
    state={"message": "Please send me the invoice for my order."},
    questions={"route": {"type": "choice", "instructions": "Choose team", "criteria": {"billing": "Invoices", "sales": "Enquiries"}}}
)
print(res["answers"]["route"]["choice"])
Read moreDetails
Parameters: ~26.09B (11 BF16 safetensors shards) + FP32 readout head for 255 response codes.
Requirements: weights take up around 52.2 GB; requires Linux with a CUDA-capable GPU.
Limitations: probability calibration has not been independently evaluated; GPU inference was not re-tested for this release.
Discuss Telegram 0 30

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 49 GB · ready files54 GB
  • FP8 / INT8recommendedweights 24 GB27 GB
  • 4-bitweights 14 GB16 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.

    # cjqhhh/Sloth-27B-v1
    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 cjqhhh/Sloth-27B-v1 --local-dir ~/models/Sloth-27B-v1
    # The model is in ~/models. How to run it is in the model card (link above).
    Add more LoRAs to Sloth-27B-v1
  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

  1. Qwen3.8-27Bfamily root
  2. d3
  3. Sloth-27B-v1

New posts about versions of Qwen3.8-27B (quants, finetunes, LoRAs) will show up under the bell.

Quantised versions 12All on Hugging Face
Fine-tunes 12All on Hugging Face
LoRAs and adapters 2All on Hugging Face

About the model

cjqhhh/Sloth-27B-v1
Author
cjqhhh
Category
Other
Released
Posted

On NeyroLearn

Hugging Face

300

Ask AI about this model

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

Comments

Other

BFL

BFL