Skip to content
Computer vision

Qwen3.8-27B OCR Finance Next R1

документный VQA по вьетнамским финотчётам

1 min read

Файнтюн Qwen3.8-27B от unsloth/Qwen3.8-27B, обученный читать вьетнамские аудированные финансовые отчёты. Отвечает на вопросы по одной странице или по целому отчёту, считает изменения, проценты, доли, коэффициенты, направления, суммы и проверки, извлекает данные в JSON и сообщает, если на странице нет ответа. Размер — 27,36 млрд параметров.

Ключевые особенности

  • Два варианта поставки: merged-веса bf16 (12 шардов, 51 ГБ) или LoRA-адаптер на 467 МБ поверх базовой модели.
  • LoRA с r=16, alpha=16, без dropout, применена к attention, linear-attention (GDN) и MLP-проекциям языковой модели.
  • Понимает вопросы по странице, по всему отчёту до 100 страниц в контексте и по нескольким страницам с метками вида «Ảnh 1».
  • Умеет извлекать ответы в JSON, считать арифметику и проверять суммы.
  • Рекомендуемый режим: thinking off, temperature 0, вопрос на вьетнамском в формате обучения.

Пример

vllm serve paulaan86/qwen38-27b-ocr-finance-next-r1 --max-model-len 32768
# для целых отчётов: --max-model-len 262144
ПодробнееANLS: страницы — 0,860 (база 0,809), целые отчёты — 0,846 (0,769); held-out accuracy — 0,826 (0,253).
Merged-веса: bf16, 12 шардов, 51 ГБ; LoRA-адаптер — 467 МБ.
Ограничения: проценты — 57 % точных ответов; только вьетнамские финотчёты; лицензия CC BY-NC 4.0, non-commercial.
Discuss Telegram 0 2

Will it run?

Runs in 4-bit

Full precision does not fit, a quantised version does: quality barely changes down to 8 bits, more noticeably at 4 bits.

  • BF16weights 51 GB · ready files57 GB
  • FP8 / INT8recommendedweights 25 GB29 GB
  • 4-bitweights 14 GB16 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.

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.

    # paulaan86/qwen38-27b-ocr-finance-next-r1
    # ===== 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 paulaan86/qwen38-27b-ocr-finance-next-r1 --local-dir ~/models/qwen38-27b-ocr-finance-next-r1
    # The model is in ~/models. How to run it is in the model card (link above).
  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. Qwen3.8-27Bfamily root
  2. Qwen3.8-27B
  3. qwen38-27b-ocr-finance-next-r1

New posts about versions of Qwen3.8-27B (quants, finetunes, LoRAs) will show up under the bell.

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

About the model

paulaan86/qwen38-27b-ocr-finance-next-r1
Author
paulaan86
Category
Computer vision
Released
Posted

On NeyroLearn

Hugging Face

20

Ask AI about this model

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

Comments