Skip to content

Reports Index

The reports/ directory is the formal report entry point for the reconstructed project. Its role is to keep the Markdown source files and the final PDF outputs organized in one place.

The current export path is:

Markdown -> HTML -> Chromium PDF printing

The script that maintains this mapping is:

reports/build_reports.py

Current Status

The PDFs in this repository are not treated as permanently stable artifacts. They change often and can grow the repository size quickly, so the final PDF set will be cleaned up again in a later push.

At the current stage:

  1. some PDFs already exist,
  2. some PDFs may be temporarily missing,
  3. the Markdown source files still remain the formal report sources.

If the goal is to read report content, the Markdown files are the primary source. If the goal is to collect final deliverables, the later cleaned PDF set should be treated as the final version.

Step 1: Model Selection and Dataset Analysis

Open-Source Model Selection Report for the China Post Customer-Service Task

SFT Training Plan and Risk Control Report

CSDS Statistics and Keyword Extraction Report

Postal Conversation Filtering and Vector-Space Visualization Report

Classification Evaluation and Boundary-Case Analysis Report

Cluster Visualization and Label Optimization Report

Step 2: Dataset Collection and Framework Setup

Postal FAQ Crawler Training Sample Report

Postal Customer-Service LLM System Design Report

Step 3: LoRA Fine-Tuning

Full Apple MLX Qwen2.5 Postal Fine-Tuning Report

Qwen2.5-3B Postal LoRA Rank Sweep Report

Qwen2.5-7B Postal LoRA Rank Sweep Report

Qwen2.5 Postal SFT Final Comparison Report