Skip to content
Other

Spark1-14M

крошечная LLM, собранная редактированием весов

1 min readupdated

Spark1-14M — английская decoder-only модель на 14,31 млн параметров в стиле Llama. Её не обучали: автор получил веса прямым редактированием и заявляет, что модель обходит Pythia-14M и Ettin-17M на всех проверенных бенчмарках. Контекст — 1024 токена, словарь — 8192 BPE.

Ключевые особенности

  • 14,31M параметров, tied embeddings; hidden 288, 12 слоёв, 6 голов, GQA с 3 KV-головами.
  • Словарь уменьшен с 50 304 до 8192: больше параметров остаётся трансформеру.
  • Без обучения: pruning словаря, depth-scaled init, bias в SwiGLU, буст частых английских токенов и GQA 6→3.
  • Заявлен обход Pythia-14M и Ettin-17M на всех 11 бенчмарках, суммарно 36,6% против 27,0% и 31,9%.
  • Есть GGUF Q8_0 (15,2 МБ) и Q4_0 (8,1 МБ) для llama.cpp/ollama.

Пример

llama-cli -m Spark1-14M-Q8_0.gguf -p 'The man opened the door and' -n 50
ПодробнееБенчмарки: ARC-Challenge 28,4%, HellaSwag 32,8%, PIQA 61,5%, LAMBADA 26,5%.
Ограничения: только английский, контекст 1024, base-модель без instruction tuning; не для фактов и мультиязычности.
Формат: GGUF Q8_0 — точнее, Q4_0 — меньше; запуск через llama.cpp.
Discuss Telegram 1 28

Will it run?

Working out the memory needs

The bot is still collecting this model's weight sizes; the estimate will appear here in a few minutes. Meanwhile you can rent a GPU with headroom.

Massed ComputeOpen Massed Compute

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.

    # Kuantum-ai1/Spark1-14M
    # ===== Tokens: paste yours between the quotes =====
    export HF_TOKEN=""        # huggingface.co/settings/tokens, for gated models
    # ==========================================
    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 Kuantum-ai1/Spark1-14M --local-dir ~/models/Spark1-14M
    # The model is in ~/models. How to run it is in the model card (link above).
    Add more LoRAs to Spark1-14M
  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 Spark1-14M (quants, finetunes, LoRAs) will show up under the bell.

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

About the model

Kuantum-ai1/Spark1-14M
Author
Kuantum-ai1
Category
Other
Released
Posted

On NeyroLearn

Hugging Face

281

Ask AI about this model

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

Comments