resume-builder-coverletter

resume-builder-coverletter (GGUF)

A fine-tune of Qwen2.5-3B-Instruct that writes job cover letters in the exact prompt format used by the resume-builder desktop app. It is conditioned on cover-letter templates — it follows whatever template appears in its prompt, so new templates work without retraining — and is grounded on retrieved résumé bullets (RAG) to avoid fabricating experience.

Built with Qwen.

• Base model: Qwen/Qwen2.5-3B-Instruct
• Format: GGUF, Q4_K_M quantization (~1.8 GB; ~3 GB RSS at runtime)
• Runtime: CPU via llama.cpp — roughly 1–2 min per letter on an 8 GB machine
• Prompt format: ChatML (the model's embedded chat template)

Usage

In the resume-builder app (recommended)

Open Settings → Local (GGUF) and click Download the tuned cover-letter model. The app fetches this file into its app-data directory and wires it up automatically — no manual paths.

Directly with llama.cpp

llama-cli -m coverletter-qwen2.5-3b-Q4_K_M.gguf -p "your ChatML prompt"

The app builds its own system/user prompt (RAG bullets + job description + optional template) and applies the model's chat template; see src-tauri/src/llm/prompt.rs for the exact shape.

Training

Distilled and QLoRA-fine-tuned entirely on synthetic, non-personal data — real public job descriptions paired with fictional candidate profiles, so the model generalizes at inference instead of memorizing anyone's résumé. Full pipeline: training/.

License

Non-commercial use only. This is a derivative of Qwen2.5-3B-Instruct and is distributed under the Qwen Research License Agreement (see LICENSE and NOTICE), which permits research and evaluation use only.

Qwen is licensed under the Qwen RESEARCH LICENSE AGREEMENT, Copyright (c) Alibaba Cloud. All Rights Reserved.