Skip to content
Computer vision

Grafted Gemma-4-E4B

PR-описания без забывания

1 min readupdated

Это 8B-модель image-text-to-text на базе google/gemma-4-E4B-it, адаптированная под анализ diff и генерацию описаний pull request. Автор применил Grafting: дообучил donor-чекпоинт google/gemma-4-e4b на GitHub PR reasoning traces и перенёс task vector в post-trained модель.

Ключевые особенности

  • Grafting с масштабом λ=1.0: task vector переносится из donor-модели в gemma-4-E4B-it.
  • Заморожены высокочувствительные эмбеддинги на 3.49B параметров, включая Layer 0 и 42-layer Engram.
  • Сохранены представления управляющих токенов <|think|>, <|turn|>, <|channel|> и <|tool_call|>.
  • Donor обучен 4-bit QLoRA: r=16, α=16, 258 модулей attention и MLP.
  • Запуск через transformers и Gemma4ForConditionalGeneration.

Пример

from transformers import AutoTokenizer, Gemma4ForConditionalGeneration

model_id = "carlesoctav/grafting-gh_pr_task"
tok = AutoTokenizer.from_pretrained(model_id)
model = Gemma4ForConditionalGeneration.from_pretrained(model_id, torch_dtype="auto", device_map="auto")
messages = [{"role": "user", "content": "Analyze the diff and write a PR description."}]
prompt = tok.apply_chat_template(messages, tokenize=False, add_generation_prompt=True)
inputs = tok(prompt, return_tensors="pt").to(model.device)
# outputs = model.generate(**inputs, max_new_tokens=2048, temperature=0.7, top_p=0.95)
ПодробнееОбучение donor: 12 шагов, 3 эпохи, 4×NVIDIA L4 DDP; loss 2.926 → 0.9242, mean loss 1.434.
Перенесены 258 линейных проекций; эмбеддинги на 3.49B параметров остались в post-trained значениях.
Бенчмарков и требований к VRAM в карточке нет.
Discuss Telegram 0 20

Will it run?

Runs in BF16

Fits without quantisation.

  • BF16weights 15 GB · ready files17 GB
  • FP8 / INT8recommendedweights 7.4 GB8.7 GB
  • 4-bitweights 4.2 GB5.1 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 3080 · BF16 — from RTX 4090

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.

    # carlesoctav/grafting-gh_pr_task
    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 carlesoctav/grafting-gh_pr_task --local-dir ~/models/grafting-gh_pr_task
    # The model is in ~/models. How to run it is in the model card (link above).
    Add more LoRAs to grafting-gh_pr_task
  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-E4Bfamily root
  2. gemma-4-E4B-it
  3. grafting-gh_pr_task

New posts about versions of gemma-4-E4B (quants, finetunes, LoRAs) will show up under the bell.

Quantised versions 12All on Hugging Face
Fine-tunes 12All on Hugging Face
LoRAs and adapters 8All on Hugging Face

About the model

carlesoctav/grafting-gh_pr_task
Author
carlesoctav
Category
Computer vision
Released
Posted

On NeyroLearn

Hugging Face

200

Ask AI about this model

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

Comments