Skip to content
Computer vision

Gemma 4 31B-it Thinkless R2: Speeding up responses through weight arithmetic

2 min readupdated

A modification of the base google/gemma-4-31B-it model created through direct weight manipulation without additional training. In this version, the output layer is separated from the input embeddings (tie_word_embeddings: false), and a special vector is added to the end-of-thinking token, which reduces excessive model reasoning. This approach cuts down the number of generated tokens and speeds up response retrieval while maintaining accuracy on question-answering tasks. The model is suitable for researchers and developers who value fast large language model (LLM) performance without lengthy thinking generation phases.

Key features

  • Built on google/gemma-4-31B-it without a retraining stage.
  • Output layer weights modified and a vector added to the end-of-thinking token.
  • Reduces tokens per correct answer to 0.638x at zero temperature.

What you can do

  • Use for efficient question-answering systems where fast results are required.
  • Run via the vLLM engine to test optimized text generation.

Example

Example prompt for the model: "Explain the basic principles of thermodynamics in simple terms." The model produces a correct answer faster and with fewer intermediate "thoughts" than the standard Gemma 4 version with full reasoning enabled.

Read moreDetails
Benchmarks at temperature 0: tokens per correct answer — 0.638x, accuracy — +0.05 points.
Limitations: accuracy drops in multi-step tool calling (BFCL v4) and in LiveCodeBench.
Hardware requirements: running in FP8 on load (for example, on an NVIDIA B200 GPU via the vLLM engine).
Discuss Telegram 0 7

Will it run?

Runs in 4-bit

Full precision does not fit, a quantised version does: quality barely changes down to 8 bits, more noticeably at 4 bits.

  • BF16weights 61 GB · ready files67 GB
  • FP8 / INT8recommendedweights 30 GB34 GB
  • 4-bitweights 17 GB19 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.

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.

    # omunaman/gemma-4-31b-it-thinkless-r2
    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 omunaman/gemma-4-31b-it-thinkless-r2 --local-dir ~/models/gemma-4-31b-it-thinkless-r2
    # 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. gemma-4-31Bfamily root
  2. gemma-4-31B-it
  3. gemma-4-31b-it-thinkless-r2

New posts about versions of gemma-4-31B (quants, finetunes, LoRAs) will show up under the bell.

Quantised versions 12All on Hugging Face
Fine-tunes 12All on Hugging Face
Merges 3All on Hugging Face

About the model

omunaman/gemma-4-31b-it-thinkless-r2
Author
omunaman
Category
Computer vision
Released
Posted

On NeyroLearn

Hugging Face

70

Ask AI about this model

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

Comments

Language models

bison-1

bison-1