Skip to content
Other

NanoReasoner-300M

компактная reasoning/code LLM на 352 млн параметров

1 min readupdated

NanoReasoner-300M — reasoning/code LLM на 352 321 602 параметра, обученная с нуля в три этапа: 25 млрд токенов предобучения, SFT на цепочках <think>...</think> и GRPO с верификаторами математики и Python-тестов. Архитектура: GQA, FlexAttention, U-Net-скипы, x0-миксины, кастомный SuperBPE-токенизатор на 49 152 токена и контекст 2048.

Ключевые особенности

  • GQA 16 query / 4 KV heads, QK-norm и окна внимания 256/2048 токенов.
  • U-Net-скипы между 12 encoder- и 12 decoder-слоями, x0 residual mixin и Value Embeddings.
  • Обучение на FineWeb-Edu, Cosmopedia v2, FineMath 4+ и CodeParrot Clean.
  • SFT на Bespoke-Stratos-17k, OpenMathInstruct-2 и OpenCodeInstruct, затем GRPO на GSM8K и MBPP.
  • Свой токенизатор: 49 152 токена, спецтокены для <think> и отдельных цифр.
  • Полный пайплайн: скрипты pretrain/SFT/GRPO, Modal-пайплайн данных и Colab для RTX PRO 6000.

Пример

modal run modal_smoke_test.py
python train_pretrain.py --data-dir ./data --ckpt-dir ./checkpoints --batch-size 32 --grad-accum 2 --seq-len 2048 --max-steps 35000
ПодробнееБенчмарков в карточке нет: заявлены 25 млрд токенов предобучения и 352,3 млн параметров.
Контекст — 2048 токенов; требования к VRAM не указаны.
GRPO использует GSM8K и MBPP с жёсткой проверкой ответов и Python-тестов.
Ограничение: модель обучена с нуля и не приводит сравнений с другими LLM.
Discuss Telegram 0

Will it run?

Runs in BF16

Fits without quantisation.

  • BF16weights 3.0 GB · ready files3.7 GB
  • FP8 / INT8recommendedweights 1.5 GB2.1 GB
  • 4-bitweights 0.8 GB1.4 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.

    # SSSASDA4353535/idk325-ckpt
    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 SSSASDA4353535/idk325-ckpt --local-dir ~/models/idk325-ckpt
    # The model is in ~/models. How to run it is in the model card (link above).
    Add more LoRAs to idk325-ckpt
  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

New posts about versions of idk325-ckpt (quants, finetunes, LoRAs) will show up under the bell.

No known relatives of this model yet. All derivatives on Hugging Face

About the model

SSSASDA4353535/idk325-ckpt
Author
SSSASDA4353535
Category
Other
Released
Posted

On NeyroLearn

Hugging Face

00

Ask AI about this model

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

Comments

Other

Night3R

Night3R
Other

BruNet

BruNet