Skip to content
Language models

GrimmCoder_v2

компактная модель на 107M параметров для Python-кода

2 min readupdated

GrimmCoder_v2 — ультралёгкая языковая модель архитектуры LLaMA на 107M параметров от автора Bayhakl1, обученная с нуля исключительно на исходном коде Python. Она решает задачи обычного автодополнения кода слева направо (Left-to-Right) и вставки недостающего фрагмента в середину функции — FIM (Fill-in-the-Middle, генерация кода между известным началом и концом). Вторая версия дообучена на датасете Avelina/python-edu-cleaned поверх базового чекпоинта GrimmCoder, благодаря чему результат на бенчмарке HumanEval вырос почти вдвое по сравнению с первой версией. Модель подойдёт разработчикам и исследователям для экспериментов с локальным сверхбыстрым автодополнением Python без серьёзных требований к железу.

Ключевые особенности

  • Компактный размер ~106.98M параметров, 16 слоёв и контекстное окно 2048 токенов с групповым механизмом внимания GQA и активацией SwiGLU.
  • Специализированный байтовый BPE-токенизатор на 8192 токена, обученный именно под синтаксис Python.
  • Поддержка двух режимов генерации: классического продолжения кода и FIM в форматах PSM и SPM (50/50 при обучении).
  • Заметный прирост точности: 9.15% (15 из 164 задач) на HumanEval против 5.49% у базовой версии GrimmCoder.

Что можно сделать

  • Встроить через библиотеку transformers в локальный плагин или скрипт для подсказок и дописывания Python-функций на лету.
  • Использовать режим Fill-in-the-Middle для автоматического заполнения тела функции, когда уже написаны её сигнатура и последующий вызов.
  • Взять за лёгкую основу для дальнейшего дообучения (файнтюна) на собственных репозиториях или учебных задачах по Python.

Пример

from transformers import AutoModelForCausalLM, AutoTokenizer
import torch

tokenizer = AutoTokenizer.from_pretrained("Bayhakl1/GrimmCoder_v2")
model = AutoModelForCausalLM.from_pretrained("Bayhakl1/GrimmCoder_v2", dtype=torch.bfloat16, device_map="auto")
prompt = "<|fim_prefix|>def add(a: int, b: int) -> int:\n    <|fim_suffix|>\n\nprint(add(2, 3))\n<|fim_middle|>"
inputs = tokenizer(prompt, return_tensors="pt", add_special_tokens=False).to(model.device)
print(tokenizer.decode(model.generate(**inputs, max_new_tokens=64)[0]))
ПодробнееHumanEval pass@1: 9.15% (15/164 задач в режимах L2R greedy, L2R t=0.2 и FIM greedy), всего решено 20 уникальных задач из 164; синтаксическая валидность до 96.3%.
Обучение: 8.70B токенов суммарно (7.93B на претрейне bigcode/starcoderdata + 774.1M на дообучении Avelina/python-edu-cleaned, 6000 шагов, val loss 1.2260).
Ограничения: работает только с кодом на Python в формате автодополнения и FIM, не предназначена для диалогов и инструкций на естественном языке.
Discuss Telegram 1 88

Will it run?

Runs in BF16

Fits without quantisation.

  • BF16weights 0.2 GB · ready files0.9 GB
  • FP8 / INT8recommendedweights 0.1 GB0.8 GB
  • 4-bitweights 0.1 GB0.8 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.

Language models: with a 8192-token context.

Estimate: weights plus working memory. Long videos, high resolutions and long contexts need more.

Run it on Massed Compute

A ready script for the machine we advise above: paste it and the model downloads and starts.

  1. 1. Rent A30

    $0.26 an hour with the coupon. Pick the Ubuntu image when you create the machine.

    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.

    # Bayhakl1/GrimmCoder_v2 — A30, FP8 / INT8
    # NVIDIA driver: CUDA 13 needs 580+. If older or missing, the current nvidia-open is installed (keep yours: KEEP_DRIVER=1)
    DRV=$(nvidia-smi --query-gpu=driver_version --format=csv,noheader 2>/dev/null | head -1 | cut -d. -f1)
    case "$DRV" in ""|*[!0-9]*) DRV=0 ;; esac
    if [ "$DRV" -lt 580 ] && [ -z "${KEEP_DRIVER:-}" ]; then
      . /etc/os-release && \
      curl -fsSLO "https://developer.download.nvidia.com/compute/cuda/repos/ubuntu${VERSION_ID//./}/x86_64/cuda-keyring_1.1-1_all.deb" && \
      sudo dpkg -i cuda-keyring_1.1-1_all.deb && rm -f cuda-keyring_1.1-1_all.deb && \
      sudo apt-get update && \
      sudo DEBIAN_FRONTEND=noninteractive apt-get install -y nvidia-open || \
      { echo "Could not install the NVIDIA driver: see the error above. To keep the current one, run with KEEP_DRIVER=1."; exit 1; }
      echo "NVIDIA driver updated. Reboot the machine (sudo reboot) and run this same script again."
      exit 0
    fi
    curl -LsSf https://astral.sh/uv/install.sh | sh
    source "$HOME/.local/bin/env"
    uv venv ~/vllm --python 3.12 && source ~/vllm/bin/activate
    uv pip install vllm
    vllm serve Bayhakl1/GrimmCoder_v2 --max-model-len 8192 --quantization fp8
    Add more LoRAs to GrimmCoder_v2
  3. 3. Use it

    The server listens on port 8000 and speaks the OpenAI API: http://localhost:8000/v1 on the machine. From your computer: ssh -L 8000:localhost:8000 user@machine-IP.

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.

  • MTP (multi-token prediction)0.34.0from November 16, 2026new release

    1.5–2× generation speed

    If the model has MTP layers (DeepSeek-V3/R1, Qwen3-Next, GLM, MiMo and others), it predicts several tokens per step itself.

    How to enable: vLLM / SGLang: --speculative-config '{"method": "mtp", "num_speculative_tokens": 1}'. Check the model card.

    Earlier generations (2)
    • EAGLE-3 — Speculative decoding with a small draft head. For models without their own MTP layers.
    • EAGLE-2 / Medusa — Earlier draft-head generations. Still work if that is all you find.
  • FP8 KV cache0.34.0from November 16, 2026new release

    2× more context in the same memory

    Keeps the attention cache in 8 bits, so long contexts fit where they did not.

    How to enable: vLLM: --kv-cache-dtype fp8; llama.cpp: -ctk q8_0 -ctv q8_0.

  • vLLM / SGLang0.34.0from November 16, 2026new release

    many times more tokens/s on a server

    Server engines: batching, FlashAttention, FP8 and speculative decoding out of the box.

    How to enable: vllm serve <model> — an OpenAI-compatible API.

    Earlier generations (1)
    • llama.cpp — GGUF models on any GPU, Mac or even CPU; some layers can stay in RAM.

Model family

New posts about versions of GrimmCoder_v2 (quants, finetunes, LoRAs) will show up under the bell.

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

About the model

Bayhakl1/GrimmCoder_v2
Author
Bayhakl1
Category
Language models
Released
Posted

On NeyroLearn

Hugging Face

881

Ask AI about this model

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

Comments