Skip to content
Computer vision

Joko-edited

точечная правка зацикливания thinking

2 min read

Экспериментальный чекпоинт на базе Qwen/Qwen3.5-2B (2,21 млрд параметров) — двуязычный помощник для индонезийского и английского. Это не новая тренировка: в полных BF16-весах изменена одна tied-строка embedding/output для токена </think> (248069), всего 2000 значений, без градиентов и обучения.

Ключевые особенности

  • Правка одной строки весов влияет и на чтение, и на генерацию токена </think>.
  • База — Qwen3.5-2B; исходный SFT-экспорт Joko epoch 2 с merged LoRA rank 16 / alpha 32.
  • Поддерживает индонезийский и английский, image-text-to-text, обучен в том числе на GUI/agent-данных.
  • В калибровочном тесте identity-промпт завершился за 15 токенов вместо 2048; все 10 базовых thinking-ответов дошли до EOS.
  • Задача со скриншотом и копированием иконки всё ещё зацикливалась до 512 токенов.
  • Inference по умолчанию работает в non-thinking режиме.

Пример

Калибровочный identity-промпт: вместо ухода в цикл до 2048 токенов ответ закончился за 15 токенов. На 10 простых thinking-запросах модель дошла до EOS, но сценарий со скриншотом и копированием иконки не исправился — цикл до 512 токенов.

ПодробнееБенчмарков нет: 15 токенов вместо 2048 и 10/10 EOS — результаты только на калибровочном наборе, не held-out тест.
Полные BF16-веса; нужен Transformers с Qwen3_5ForConditionalGeneration и AutoProcessor, PEFT не требуется.
Обобщение и побочные эффекты не проверены; улучшений в математике, коде и visual grounding не установлено.
Discuss Telegram 0 20

Will it run?

Runs in BF16

Fits without quantisation.

  • BF16weights 4.1 GB · ready files5.0 GB
  • FP8 / INT8recommendedweights 2.1 GB2.8 GB
  • 4-bitweights 1.2 GB1.8 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.

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.

    # Mesosfer/Joko-edited
    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 Mesosfer/Joko-edited --local-dir ~/models/Joko-edited
    # The model is in ~/models. How to run it is in the model card (link above).
    Add more LoRAs to Joko-edited
  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. Qwen3.5-2B-Basefamily root
  2. Qwen3.5-2B
  3. Joko-edited

New posts about versions of Qwen3.5-2B-Base (quants, finetunes, LoRAs) will show up under the bell.

Quantised versions 12All on Hugging Face
Fine-tunes 12All on Hugging Face
LoRAs and adapters 10All on Hugging Face
Merges 2All on Hugging Face

About the model

Mesosfer/Joko-edited
Author
Mesosfer
Category
Computer vision
Released
Posted

On NeyroLearn

Hugging Face

200

Ask AI about this model

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

Comments

Language models

Flint

Flint