Skip to content
Other

Qwen3.8-27B Jacobian Lens

двуязычная линза для интерпретации LLM

2 min readupdated

Предрасчитанная full-layer Jacobian Lens (J-lens) для Qwen/Qwen3.8-27B: 64 слоя, d_model = 5120, гибридная архитектура из 48 Gated DeltaNet linear-attention и 16 full-attention слоёв. Чекпоинт обучен официальным оценщиком anthropics/jacobian-lens и позволяет проецировать residual stream промежуточных слоёв в пространство финального слоя.

Ключевые особенности

  • Двуязычная подгонка строго 1:1: 50 пассажей WikiText-103 на английском + 50 из китайской Wikipedia, всего N = 100.
  • Сохранены 63 полные матрицы 5120 × 5120 для слоёв 0..62 в float16; целевой слой — 63.
  • Формат совместим с torch.load и jacobian-lens; словарь J отображает номер слоя в тензор.
  • В карточке приведены бенчмарки Multihop и Multilingual и примеры bilingual readout.
  • Это не готовая LLM, а исследовательский артефакт для анализа внутренних представлений Qwen3.8-27B.

Пример

import torch
lens = torch.load("Qwen3.8-27B_jacobian_lens.pt", map_location="cpu")
J_38 = lens["J"][38].float()
# h_proj = h_38 @ J_38.T
# logits = lm_head(final_rms_norm(h_proj))
ПодробнееФайл: 3.30 GB / 3.08 GiB, SHA256 e9b11582f2023b7934ed591ab1134f7a3a9b6d64782665d3400a3fc56f532402.
На N=100: Multihop pass@1 — 36.31%, pass@5 — 56.35%; Multilingual pass@1 — 25.08%, pass@5 — 40.58%.
Vanilla Logit Lens даёт 33.93% / 53.97% и 21.26% / 38.86% соответственно.
В примере readout на промпте про «страну в форме сапога» топ-5 токенов на слое 41: ['意大利', ' Italy', ' Italian', 'Italy', ' Italia'].
Оценка проводилась на двух NVIDIA DGX Spark GB10; всего 256 000 полных обратных проходов.
Discuss Telegram 1

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.

    # lmxxf/Qwen3.8-27B-jacobian-lens
    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 lmxxf/Qwen3.8-27B-jacobian-lens --local-dir ~/models/Qwen3.8-27B-jacobian-lens
    # 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

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

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

About the model

lmxxf/Qwen3.8-27B-jacobian-lens
Author
lmxxf
Category
Other
Released
Posted

On NeyroLearn

Hugging Face

01

Ask AI about this model

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

Comments