Skip to content
Computer vision

Qwen3.5-4B 4k GRPO maturebass

a math reasoning model

1 min readupdated

This model is a specialized fine-tune of the Qwen/Qwen3.5-4B base LLM (large language model), trained using GRPO (group relative policy optimization, a reinforcement learning method). It is built to solve math problems within a strict 4,096-token response limit. Trained on the DeepScaleR dataset to sharpen logical, step-by-step deduction, it is a solid fit for researchers and developers looking for a compact model with stronger math reasoning.

Key features

  • Base model: Qwen/Qwen3.5-4B, roughly 4.5B parameters.
  • Training algorithm: GRPO with token-level loss and group reward normalization.
  • Generation limit: fixed token budget with max_new_tokens set to 4096.
  • Correctness reward: binary scoring based on the final answer inside \boxed{} tags.

What you can do

  • Run it in services or local scripts via the transformers library to tackle complex math problems.
  • Spin up inference with vLLM for fast benchmarking and testing.
  • Study how the model handles step-by-step deduction on competition-level or engineering calculations.

Example

from transformers import AutoModelForCausalLM, AutoTokenizer

repo = "budget-internalization-iclr2027/qwen3.5-4b-4k-grpo-multiseed-seed42-maturebass-s300"
tok = AutoTokenizer.from_pretrained(repo)
model = AutoModelForCausalLM.from_pretrained(repo, torch_dtype="auto", device_map="auto")
Read moreDetails
Algorithm: GRPO (leave-one-out baseline, group reward normalization)
Data: DeepScaleR (math), up to 3 epochs
Optimizer: Adam, cosine LR, peak LR 5e-7
Checkpoint: step 300, seed 42, BF16 weights
Discuss Telegram 0 19

Will it run?

Runs in BF16

Fits without quantisation.

  • BF16weights 8.5 GB · ready files9.8 GB
  • FP8 / INT8recommendedweights 4.2 GB5.2 GB
  • 4-bitweights 2.4 GB3.1 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 3080

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.

    # budget-internalization-iclr2027/qwen3.5-4b-4k-grpo-multiseed-seed42-maturebass-s300
    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 budget-internalization-iclr2027/qwen3.5-4b-4k-grpo-multiseed-seed42-maturebass-s300 --local-dir ~/models/qwen3.5-4b-4k-grpo-multiseed-seed42-maturebass-s300
    # 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. Qwen3.5-4B-Basefamily root
  2. Qwen3.5-4B
  3. qwen3.5-4b-4k-grpo-multiseed-seed42-maturebass-s300

New posts about versions of Qwen3.5-4B-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 12All on Hugging Face
Merges 1All on Hugging Face

About the model

budget-internalization-iclr2027/qwen3.5-4b-4k-grpo-multiseed-seed42-maturebass-s300
Author
budget-internalization-iclr2027
Category
Computer vision
Released
Posted

On NeyroLearn

Hugging Face

190

Ask AI about this model

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

Comments