Skip to content
Other

Connais 1

a compact encoder for fast decision-making and classification

2 min readupdated

Connais 1 is a small, specialized 67-million-parameter model (the weights take up 267 MB) built on the ModernBERT architecture (jhu-clsp/ettin-encoder-68m). Unlike generative language models (LLMs), it doesn't generate text. Instead, it evaluates the input in a single encoder pass and returns probabilities for your predefined answer options. It is ideal for agent tool selection, request routing, and strict classification. The model is text-only and processes up to 32,768 tokens (text chunks) per request.

Key features

  • Bidirectional encoder: analyzes the entire input text at once using 19 bidirectional layers.
  • Flexible answer types: supports multiple choice (choice), yes/no evaluation (noul), and numeric scales (score).
  • CPU and GPU efficiency: runs fast even on consumer hardware like the Apple M4.
  • No text generation: returns clean probabilities and confidence scores for each option without extra reasoning.

What you can do

  • Agent integration: set up automatic selection of the right tool or function based on the user's request description.
  • Request routing: sort incoming customer tickets by category and intent using accurate probabilities.
  • Binary and scale classification: quickly check if text meets specific criteria or rate sentiment along predefined scales.

Example

from connais import load_engine

engine = load_engine('connais-1')
response, timing = engine(
    'User: is it raining in Zurich?',
    {'tool': {'type': 'choice', 'instructions': 'Which tool?',
              'criteria': {'get_weather': 'Current weather', 'none': None}}},
)
Read moreDetails
• Architecture: 66.8M parameters, including a 42.36M base and a tabular encoder.
• Hardware: tested on Apple M4; light on compute and memory.
• Limitations: limited factual knowledge, text-only (primarily English), and long contexts over 8k tokens rely mostly on text near the question.
• License: weights under CC BY-NC 4.0, code under Apache License 2.0.
Examples
Discuss Telegram 1 34

Will it run?

Runs in F32

Fits without quantisation.

  • F32weights 0.2 GB · ready files0.8 GB
  • BF16weights 0.1 GB0.6 GB
  • FP8 / INT8recommendedweights 0.1 GB0.6 GB
  • 4-bitweights 0.0 GB0.5 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 2060 · F32 — 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.

    # Connais/connais-1-68m
    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 Connais/connais-1-68m --local-dir ~/models/connais-1-68m
    # The model is in ~/models. How to run it is in the model card (link above).
    Add more LoRAs to connais-1-68m
  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 ettin-encoder-68m (quants, finetunes, LoRAs) will show up under the bell.

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

About the model

Connais/connais-1-68m
Author
Connais
Category
Other
Released
Posted

On NeyroLearn

Hugging Face

341

Ask AI about this model

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

Comments

Other

Argus V3

Argus V3