Skip to content
Other

nanochat RLM d24

SFT-чекпоинт с рекуррентной латентной памятью

1 min read

Это chat (SFT) чекпоинт исследовательской RLM-модели от rlm-exps из форка nanochat (ветка h100-rlm). 24-слойный GPT-бэкбон обрабатывает текст ограниченными чанками по 1023 токена и передаёт между ними 512 обученных векторов памяти.

Ключевые особенности

  • Архитектура: 24 слоя, 12 голов, ширина 1536; словарь 32 768 токенов (nanochat BPE).
  • Рекуррентная латентная память: вход цикла устроен как [memory vectors] [<EOM>] [text tokens] [summary slots].
  • SFT на nanochat chat mixture: SmolTalk, MMLU auxiliary_train x3 и GSM8K train x4 — всего 789 759 диалогов.
  • Веса в bfloat16 в формате PyTorch state_dict; в репозитории также токенизатор, конфиги и логи SFT.
  • Это не модель для Hugging Face transformers: она не загрузится через AutoModel, нужен форк nanochat и скрипт rlm_sample.

Пример

git clone --branch h100-rlm https://github.com/ifsheldon/nanochat.git
cd nanochat
hf download rlm-exps/nanochat-rlm --local-dir training_run/rlm-sft
Подробнее Претрен: 5 574 шага по 1 047 552 токена на ClimbMix — около 5,84 млрд текстовых токенов, одна H100 NVL 94 ГБ.
SFT: 504 обновления, held-out loss снизился с 1.511 до 0.867; обучение заняло 14,5 ч на одной H100 NVL.
Ограничения: язык — английский; чат-формат требует специальных токенов nanochat; это исследовательский чекпоинт, а не готовая HF-модель.
Discuss Telegram 0

Will it run?

Runs in BF16

Fits without quantisation.

  • BF16weights 3.9 GB · ready files4.8 GB
  • FP8 / INT8recommendedweights 2.0 GB2.7 GB
  • 4-bitweights 1.1 GB1.7 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.

    # rlm-exps/nanochat-rlm
    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 rlm-exps/nanochat-rlm --local-dir ~/models/nanochat-rlm
    # The model is in ~/models. How to run it is in the model card (link above).
    Add more LoRAs to nanochat-rlm
  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 nanochat-rlm (quants, finetunes, LoRAs) will show up under the bell.

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

About the model

rlm-exps/nanochat-rlm
Author
rlm-exps
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

S-PLM-TM

S-PLM-TM