Skip to content
Other

e2h-polaris-gemma-4-e4b-it-full-all_rb

GRPO-чекпойнт для математики

2 min read

Чекпойнт step 200 из full-split раунда E2H Polaris: языковую часть google/gemma-4-E4B-it дообучили GRPO на математических задачах Polaris-Dataset-53K. Обучение шло через upstream verl 0.9.1 (v0 trainer, FSDP2), а задачи делили по 16-сэмпловому pass rate базовой модели. Vision- и audio-башни не трогали — они остались от базового мультимодального чекпойнта.

Ключевые особенности

  • GRPO-дообучение языковой части Gemma 4 E4B на математике; reward — правило проверки ответа с эталоном.
  • Данные: train_all из 29 394 задач (train_easy + train_hard), разбитых по pass rate базовой модели.
  • 200 шагов по 64 промпта × 8 сэмплов, lr 1e-6, KL loss 0.001 (low_var_kl), ответы до 2048 токенов с thinking.
  • Сохранённые шаги каждые 50 — отдельные ветки; main — финальный чекпойнт.
  • Веса bf16, собраны из FSDP-шардов verl; K/V в последних 18 слоях Gemma 4 общие, их k_proj/v_proj/k_norm скопированы из базы без изменений.

Пример

from transformers import AutoModelForImageTextToText

model = AutoModelForImageTextToText.from_pretrained(
    "davidafrica/e2h-polaris-gemma-4-e4b-it-full-all_rb",
    revision="step-200"
)
ПодробнееОценка: eval_all — 5 304 задачи, eval_hard — 10 765 задач, которые база не решила ни разу из 16.
Данные: train_all — 29 394 задачи; обучение — 200 шагов, 64 промпта × 8 сэмплов, lr 1e-6, KL 0.001, до 2048 токенов с thinking.
База: google/gemma-4-E4B-it; обучен только языковой модуль, vision/audio без изменений.
Веса bf16 из FSDP-шардов; загрузка по revision="step-200", main — финальный шаг.
Discuss Telegram 0 22

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.

    # davidafrica/e2h-polaris-gemma-4-e4b-it-full-all_rb
    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 davidafrica/e2h-polaris-gemma-4-e4b-it-full-all_rb --local-dir ~/models/e2h-polaris-gemma-4-e4b-it-full-all_rb
    # 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-E4Bfamily root
  2. gemma-4-E4B-it
  3. e2h-polaris-gemma-4-e4b-it-full-all_rb

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

davidafrica/e2h-polaris-gemma-4-e4b-it-full-all_rb
Author
davidafrica
Category
Other
Released
Posted

On NeyroLearn

Hugging Face

220

Ask AI about this model

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

Comments