Skip to content
Other

Mini CLM: A compact contrastive language model without text generation

2 min read

Mini CLM is an independent implementation of the contrastive language model concept, which evaluates response option probabilities in a single pass instead of token-by-token text generation. The model is built on a frozen Qwen/Qwen3-0.6B base, supplemented with two small trained heads (state head and action head) to match situations and options via cosine similarity. The project is designed for multiple-choice tasks and ranking evaluation on a laptop GPU. It will be useful for researchers and developers exploring alternative approaches to evaluating LLM (large language model) responses.

Key features

  • Qwen3 base: Uses the compact Qwen/Qwen3-0.6B model as a frozen backbone.
  • Contrastive approach: Returns the probability for each option in a single forward pass without generating tokens.
  • Trained heads: Contains a state head and an action head trained in three stages using InfoNCE loss.
  • Specialization: Excels at semantic matching on datasets like ARC-Easy and CommonsenseQA, but does not perform step-by-step reasoning.

What you can do

  • Perform multiple-choice tasks and rank predefined response options to questions.
  • Test action-selection scenarios for agents (based on datasets like neulab/adp-v2).
  • Run the ready evaluation script using the included clm.py module and a demo example.

Example

import sys
from huggingface_hub import snapshot_download

path = snapshot_download("ManiShekhar/mini-clm")
sys.path.insert(0, path)
import clm

scorer = clm.Scorer()
print(scorer.rank("Question: What is the capital of Japan?", ["Tokyo", "Osaka", "Paris"]))
Read moreDetails
• Requirements: Requires a CUDA GPU and local execution via clm.py.
• Training: 3 stages (pre, mid, post) with a learning rate ranging from 3e-4 to 5e-5.
• Limitations: The model cannot perform step-by-step reasoning and does not output numerical scores.
Discuss Telegram 0 12

Will it run?

Working out the memory needs

The bot is still collecting this model's weight sizes; the estimate will appear here in a few minutes. Meanwhile you can rent a GPU with headroom.

Massed ComputeOpen Massed Compute

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.

    # ManiShekhar/mini-clm
    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 ManiShekhar/mini-clm --local-dir ~/models/mini-clm
    # The model is in ~/models. How to run it is in the model card (link above).
    Add more LoRAs to mini-clm
  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 mini-clm (quants, finetunes, LoRAs) will show up under the bell.

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

About the model

ManiShekhar/mini-clm
Author
ManiShekhar
Category
Other
Released
Posted

On NeyroLearn

Hugging Face

120

Ask AI about this model

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

Comments