Skip to content
Other

Florence-2 estate antiques v2

LoRA для опознания антиквариата

2 min read

Это LoRA-адаптер для Florence-2-large от ThreeSixNine: он помогает описывать и опознавать антикварные предметы по изображениям. v2 — не новый файнтюн, а продолжение обучения v1 с тем же конфигом LoRA r16 (4 718 592 обучаемых параметра) на расширенном наборе данных.

Ключевые особенности

  • База — Florence-2-large; адаптер v1 загружен без переинициализации и дообучен ещё 1 эпоху.
  • 6 937 train-строк: 3 500 строк по фарфору из v1 и 3 437 новых RedCaps-строк по антиквариату, керамике, гончарке, винтажу, thriftstorehauls и castiron.
  • На том же 386-строчном subset CiQi-Bench: generation token-F1 ≥ 0.5 — 56.0% → 66.5%, exact match — 0.0% → 3.5%, MC в JSON — 20.98% → 22.02%.
  • 1 эпоха, 868 шагов, batch 8, lr 1e-4 cosine, bf16, gradient checkpointing, около 14 минут на a10g-small.
  • Smoke gate: 288/288 adapter-тензоров изменились, сохранение/перезагрузка адаптера проверены, утечки ответов не обнаружено.

Пример

Сценарий из карточки: Space estate-antique-identifier сейчас отдаёт v1. Чтобы подключить v2, замените ADAPTER_REPO в app.py на ThreeSixNine/florence2-estate-antiques-v2 — структура adapter/ и MCP-инструмент не меняются.

ПодробнееГлавный эффект v2 — качество генерации: token-F1 ≥ 0.5 вырос на 10.5 п.п., exact match сдвинулся с нуля.
Средний train loss за эпоху — 0.7338; обучение заняло около 14 минут на a10g-small.
Ограничения: нет данных о ценах, атрибуция китайского фарфора на held-out MCQ ниже 25% случайного уровня, confidence-метки эвристические, возможны артефакты словарного перевода; лицензия CC-BY-NC-4.0.
Для v2 нужно вручную переключить Space, который пока обслуживает v1.
Discuss Telegram 0

Will it run?

Runs in F16

Fits without quantisation.

  • F16weights 1.5 GB · ready files2.1 GB
  • FP8 / INT8recommendedweights 0.7 GB1.3 GB
  • 4-bitweights 0.4 GB1.0 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 · F16 — from RTX 2060

Massed Compute hourly prices, November 2026. The discounted price applies with the coupon at checkout.

This is a LoRA: counted together with its base model Florence-2-large.

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.

    # ThreeSixNine/florence2-estate-antiques-v2
    # ===== 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 florence-community/Florence-2-large --local-dir ~/models/Florence-2-large
    hf download ThreeSixNine/florence2-estate-antiques-v2 --local-dir ~/models/florence2-estate-antiques-v2
    # The model is in ~/models. How to run it is in the model card (link above).
    Add more LoRAs to Florence-2-large
  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

Requires the base model

florence-community/Florence-2-large

This model is a LoRA / adapter. Download the base model first, then load this one on top.

New posts about versions of Florence-2-large (quants, finetunes, LoRAs) will show up under the bell.

LoRAs and adapters 2All on Hugging Face

About the model

ThreeSixNine/florence2-estate-antiques-v2
Author
ThreeSixNine
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