Skip to content
Other

SWIFT Step Probe for Qwen3-8B

reasoning step verifier

2 min readupdated

This lightweight step verifier pairs a 32,770-parameter linear head with a frozen Qwen/Qwen3-8B base model. Built on the SWIFT methodology and trained on PRM800K datasets to evaluate external generations, it scores step-by-step math and logic solutions, calculating a reward for each step to pinpoint the very first mistake in a reasoning chain. It is a useful tool for researchers and developers working to improve the reliability and verifiability of large language models (LLMs).

Key features

  • Lightweight linear head (just 32,770 parameters) built on top of a frozen Qwen3-8B.
  • Evaluates each solution step by analyzing the last four hidden states (layers 33–36).
  • Predicts the index of the first error in a reasoning chain, or returns -1 if the solution is error-free.
  • Trained on filtered PRM800K data using step-level binary cross-entropy (BCE).

What you can do

  • Score step-by-step math solutions generated by other LLMs on benchmarks like ProcessBench.
  • Integrate the verifier into generation quality evaluation pipelines as a Process Reward Model (PRM).
  • Run evaluation scripts via the SWIFT repository using local weights and configurations.

Example

git clone -b step-probe [https://github.com/aster2024/SWIFT.git](https://github.com/aster2024/SWIFT.git) && cd SWIFT
python step/eval_processbench.py Qwen/Qwen3-8B   out.json
Read moreDetails
Benchmarks: 28.9 average F1 on seed 42 (GSM8K: 36.8, MATH: 31.5, OlympiadBench: 23.9, Omni-MATH: 23.6).
Requirements: runs on top of Qwen/Qwen3-8B with a 0.45 activation threshold.
Training: 1 epoch, AdamW (lr 1e-4, weight decay 1e-5), up to 16,384 tokens per batch.
Discuss Telegram 0 28

Will it run?

Runs in BF16

Fits without quantisation.

  • BF16weights 0.0 GB · ready files0.5 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 · BF16 — 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.

    # Aster2024/swift-step-probe-qwen3-8b
    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 Aster2024/swift-step-probe-qwen3-8b --local-dir ~/models/swift-step-probe-qwen3-8b
    # 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

  1. Qwen3-8B-Basefamily root
  2. Qwen3-8B
  3. swift-step-probe-qwen3-8b

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

Quantised versions 12All on Hugging Face
Fine-tunes 12All on Hugging Face
LoRAs and adapters 4All on Hugging Face

About the model

Aster2024/swift-step-probe-qwen3-8b
Author
Aster2024
Category
Other
Released
Posted

On NeyroLearn

Hugging Face

280

Ask AI about this model

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

Comments