Model Card: Resume Information Extractor (LLM-based)

Overview

This model is a distilled, instruction-tuned version of the DeepSeek-R1-Distill-Llama-8B language model, optimized for extracting structured information from resumes in English. It was built using the Unsloth library for efficient fine-tuning and inference.

Given a raw resume text, the model outputs structured JSON containing:

skills: list of skills mentioned
education: simplified school-degree-major format
experience: list of job roles

Intended Uses

This model is designed for:

HR software to parse applicant resumes automatically
Applicant tracking systems (ATS)
AI assistants helping with recruiting and screening
EdTech or job board platforms classifying user profiles

Example Input Prompt:

You are an experienced HR and now you will review a resume then extract key information from it.

# Input
Here is the resume text:
[PASTE RESUME TEXT HERE]

### Response
<think>

Expected Output:

{
  "skills": [...],
  "education": [...],
  "experience": [...]
}

Training & Technical Details

Base model: unsloth/DeepSeek-R1-Distill-Llama-8B
Library: Unsloth with support for 4-bit quantization (bitsandbytes)
Fine-tuning style: Instruction-tuning using formatted HR task prompts
Max sequence length: 8096 tokens
Hardware requirements: ~16GB GPU RAM (with 4-bit loading)

Limitations

Performance may degrade with non-English or poorly formatted resumes
Only extracts roles (not company names or dates)
Cannot handle multi-lingual documents
Does not validate output schema; use external validators if needed

Citation

If you use this model, please cite the following components:

Unsloth: https://github.com/unslothai/unsloth
DeepSeek LLM: https://github.com/deepseek-ai

License

Apache 2.0