Skip to content
Language models

egemma-td-v2

маленький локальный фильтр доказательств для веб-поиска

2 min read

egemma-td-v2 от nekomimi-llc — компактная 272M-параметровая модель для feature-extraction. Она отвечает yes/no на вопросы о результатах веб-поиска и абзацах страниц: стоит ли открывать результат, есть ли в абзаце доказательство, нет ли попытки дать инструкции ассистенту. Это файнтюн mrm8488/embeddinggemma-2-ft-typed-decision, обученный повторять ответы более крупной remote-модели Jev 1.13.

Ключевые особенности

  • Для search result оценивает useful: вероятно ли, что страница содержит доказательства по запросу.
  • Для абзацев отвечает на relevant, evidence, contradiction и injection; абзацы с injection отбрасываются при score ≥ 0.85.
  • В продакшене локально решает useful и evidence, если вероятность ≤ 0.1 или ≥ 0.9; неопределённые случаи уходят в remote Jev 1.13.
  • В v2 добавили 6 092 внешних prompt-injection окна с весом 2x; обучение — 2 эпохи.
  • Обучена на 44 915 строках из 6 711 remote calls; оценка — на 2 589 held-out строках.

Пример

from transformers import AutoModel
model = AutoModel.from_pretrained("nekomimi-llc/egemma-td-v2", trust_remote_code=True).eval()
model.config.max_seq_length = 256
state = {"query": "how does a B-tree stay balanced", "passage": {"title": "B-tree - Wikipedia", "heading": "Insertion and deletion", "text": "A B-tree is kept balanced after insertion by splitting a would-be overfilled node into two d-key siblings."}}
questions = {"evidence": {"type": "noul", "instructions": "Does the passage explicitly answer AT LEAST ONE part of the query?"}}
answers = model.decide(state, questions, workflow="web_search_evidence")
print({k: v.to_dict() for k, v in answers.items()})
ПодробнееAUC v2: useful 0.940, evidence 0.897, relevant 0.894.
Injection: AUC 0.993 на InjecAgent и 0.97 на AgentDojo, но recall на пороге 0.85 — всего 3–5% вне распределения, поэтому проверку оставляют remote-модели.
Ограничения: вход обрезается до 256 токенов, обучение шло на английском веб-поиске, средние вероятности 0.1–0.9 менее надёжны.
Модель наследует границы решений учителя Jev 1.13.
Discuss Telegram 2 3

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.

    # nekomimi-llc/egemma-td-v2
    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 nekomimi-llc/egemma-td-v2 --local-dir ~/models/egemma-td-v2
    # The model is in ~/models. How to run it is in the model card (link above).
    Add more LoRAs to egemma-td-v2
  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 egemma-td-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

nekomimi-llc/egemma-td-v2
Author
nekomimi-llc
Category
Language models
Released
Posted

On NeyroLearn

Hugging Face

32

Ask AI about this model

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

Comments