Skip to content
Other

SalgrAI-Decider

mini model for choice selection and classification

2 min read

SalgrAI-Decider is an ultracompact 25-million parameter model that does not generate text, instead handling multi-class classification and choosing from 2 to 8 options. The model takes a situation, a question, and a list of options as input, outputting probabilities for each choice in a single forward pass. Its architecture is built on an 11-layer SalgrAI-v1.2 transformer featuring bidirectional attention and a special pointer head. This project will appeal to developers and researchers exploring compact routing models and Russian text classification.

Key features

  • Ultracompact size (just 25M parameters, runs locally without heavy hardware).
  • Bidirectional attention architecture instead of a generative language head.
  • Fast probability output for a list of 2 to 8 options in a single pass.
  • Core language base and training focused on Russian text.

What you can do

  • Integrate it into your pipeline for fast user message routing and intent classification.
  • Use it to classify dialog topics or determine when to trigger a refusal response.
  • Experiment with probability calibration and the behavior of ultra-small decision-making models.

Example

import sys
from huggingface_hub import snapshot_download

path = snapshot_download("salgr/SalgrAI_Decider_1")
sys.path.insert(0, path)
from salgrai_decider import SalgrDecider

d = SalgrDecider.from_pretrained(path)
for r in d.decide("Как установить питон на компьютер?",
                  ["программирование", "еда, здоровье и быт", "география и история"],
                  criteria="О чём этот диалог?"):
    print(f"{r['prob']:.2f}  {r['option']}")
Read moreDetails
Benchmarks: message type accuracy 0.98, dialog topic accuracy 0.89.
Hardware requirements: minimal thanks to the 25M parameter size.
Limitations: unreliable outside of trained tasks, not suitable for moderation or safety.
Training: 3 epochs on 81,534 automatically generated tasks.
Discuss Telegram 0

Will it run?

Runs in F32

Fits without quantisation.

  • F32weights 0.1 GB · ready files0.6 GB
  • BF16weights 0.0 GB0.6 GB
  • FP8 / INT8recommendedweights 0.0 GB0.5 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.

    # salgr/SalgrAI_Decider_1
    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 salgr/SalgrAI_Decider_1 --local-dir ~/models/SalgrAI_Decider_1
    # The model is in ~/models. How to run it is in the model card (link above).
    Add more LoRAs to SalgrAI_Decider_1
  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 SalgrAI_Decider_1 (quants, finetunes, LoRAs) will show up under the bell.

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

About the model

salgr/SalgrAI_Decider_1
Author
salgr
Category
Other
Released
Posted

On NeyroLearn

Hugging Face

00

Ask AI about this model

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

Comments