Skip to content
Language models

kev-4b LUX Places

LoRA против ошибочных слияний мест

2 min read

Это LoRA-адаптер от Yale Cultural Heritage AI для проверки LUX place records: указывают ли внешние авторитетные записи (Wikidata, GeoNames, Who's On First, Getty TGN, Library of Congress, GND), объявленные эквивалентными, на одно и то же реальное место. Модель отвечает на Noul-вопросы вероятностью true/false и умеет проверять как весь набор эквивалентов, так и отдельные пары.

Ключевые особенности

  • Проверяет Noul-утверждения и возвращает true/false с вероятностью: все ли внешние эквиваленты записи LUX указывают на одно место и совпадают ли отдельные пары.
  • PEFT LoRA r16 / α32 на Qwen/Qwen3.5-4B-Base; продолжает адаптер kev-4b-europeana-persons и затрагивает attention, MLP и DeltaNet projections.
  • Обучен на выборке 1000 из 143 559 LUX places с двумя и более внешними эквивалентами; использованы 2063 согласованных решения Gemini 3.1 Flash Lite и TypeSafe's Jev.
  • На held-out тесте: balanced accuracy 0.940, AUROC 0.989, Brier 0.044 против 0.795 / 0.944 / 0.115 у базового адаптера.
  • В chai-engine подключается как pointer-модель: ~135 MB адаптера плюс 4B базовая модель; около 0.5 с на решение на M4 Max после загрузки.
  • Метки — машинный консенсус двух моделей, не проверены людьми; метрики показывают согласие с этим консенсусом, а не точность.

Пример

Запуск workflow, который ищет over-merge:

python flag_overmerged_places.py places.lmdb --sample 1000 \
    --model pointer/yale-cultural-heritage/kev-4b-lux-places-equivalence \
    --device mps --workers 1 --pair-workers 1
ПодробнееТест: 499 решений по 150 местам; balanced accuracy 0.940, AUROC 0.989, Brier 0.044 (базовый адаптер — 0.795 / 0.944 / 0.115).
Для пар: 0.947 balanced accuracy и 0.992 AUROC; для whole-place: 0.852 и 0.972.
Нужны ~135 MB адаптера и 4B базовая Qwen; в chai-engine около 0.5 с на решение на M4 Max после загрузки.
Ограничение: метки — согласие двух моделей, без проверки людьми; 29 ошибок на тесте в основном спорные слияния.
Discuss Telegram 0 26

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.

    # yale-cultural-heritage/kev-4b-lux-places-equivalence
    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 yale-cultural-heritage/kev-4b-lux-places-equivalence --local-dir ~/models/kev-4b-lux-places-equivalence
    # 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.

  • 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 kev-4b-lux-places-equivalence (quants, finetunes, LoRAs) will show up under the bell.

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

About the model

yale-cultural-heritage/kev-4b-lux-places-equivalence
Author
yale-cultural-heritage
Category
Language models
Released
Posted

On NeyroLearn

Hugging Face

260

Ask AI about this model

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

Comments