Jev-LCT Adapters: Looped Calibration Transformer Layer Adapters & Mappers
Lightweight plug-and-play Looped Adapters (.pt) and post-hoc calibration mappers (.pkl) for base Qwen models.
《Jev-LCT 轻量级 Looped 层适配器与校准映射器合集，用于在原生 Qwen 基座上即插即用》
English Overview
This repository contains the standalone lightweight Looped Adapters and uncertainty calibration mappers for Jev-LCT (Looped Calibration Transformer).
Instead of downloading the entire merged model weights (which range from 2GB to 16GB), users who already have base Qwen models (Qwen2.5-0.5B, Qwen2.5-1.5B, Qwen3-8B) can directly mount these compact adapter weights (only 220MB ~ 1.5GB) onto the last $k=2$ layers of the base model.
Contents
| File Name | Target Base Model | File Size | Description |
|---|---|---|---|
lct_qwen05b.pt |
Qwen2.5-0.5B | 227.6 MB | Looped layer adapter weights for 0.5B |
lct_qwen15b.pt |
Qwen2.5-1.5B | 357.1 MB | Looped layer adapter weights for 1.5B |
lct_qwen3_8b.pt |
Qwen3-8B | 1.44 GB | Looped layer adapter weights for 8B |
lct_qwen7b.pt |
Qwen2.5-7B | 1.86 GB | Looped layer adapter weights for 7B |
lct_mapper_iso.pkl |
Universal | 910 B | Fitted Isotonic Regression calibration mapper |
lct_mapper_log.pkl |
Universal | 360 B | Multidimensional trajectory Logistic Regression mapper |
lct_mapper_temp.pkl |
Universal | 96 B | Temperature scaling calibration scalar |
Usage with Native Qwen Base Models
from lct_qwen_standalone import LCTQwen
# Mount adapter onto native Hugging Face Qwen base model
engine = LCTQwen.from_pretrained(
"Qwen/Qwen2.5-1.5B",
adapter_path="lct_qwen15b.pt",
device="cuda"
)
result = engine.predict_choice(
prompt="Determine triage urgency level: Patient exhibits severe acute chest pain.",
choices=["emergency", "urgent", "routine", "elective"]
)
print(result)
中文简介
本仓库托管 Jev-LCT (循环校准 Transformer) 的全套独立轻量级适配器权重（.pt）与保序校准映射器（.pkl）。
对于本地已有 Qwen 官方基座（如 Qwen/Qwen2.5-0.5B、Qwen/Qwen2.5-1.5B、Qwen3-8B）的开发者，无需重复下载数十 GB 的完整权重，仅需下载本仓库对应的轻量适配器（仅 220MB ~ 1.5GB），即可在原生基座上获得 50ms 级别系统一极速决策与内生轨迹校准置信度能力。
核心收益
- 极速下载与分发：1.5B 适配器仅 357MB，几秒即可下载加载完毕；
- 零显存与存储冗余：基座参数保持冻结，仅微调末端循环层；
- 即插即用：与
lct_qwen_standalone.py单文件推理引擎原生兼容。
Citation
@article{cao2026lct,
title={Looped Calibration Transformer: Free Calibrated Confidence from Recurrent Computation Trajectories for Small Decision Models},
author={Cao, Haowei},
year={2026},
publisher={GitHub},
journal={GitHub repository},
howpublished={\url{https://github.com/gitchw/LCT}}
}