---
library_name: peft
license: apache-2.0
base_model: meta-llama/Llama-2-7b-hf
tags:
- resume-screening
- hr-tech
- llama2
- lora
- peft
- fine-tuned
---

# Advanced Resume Screening Model

## Model Description

This is a LoRA (Low-Rank Adaptation) fine-tuned version of Llama-2-7B specifically optimized for resume screening and candidate evaluation tasks. The model can analyze resumes, extract key information, and provide structured assessments of candidate qualifications.

- **Developed by:** kiritps
- **Model type:** Causal Language Model (LoRA Fine-tuned)
- **Language(s):** English
- **License:** Apache 2.0
- **Finetuned from model:** meta-llama/Llama-2-7b-hf

## Model Sources

- **Repository:** https://huggingface.co/kiritps/Advanced-resume-screening

## Uses

### Direct Use

This model is designed for HR professionals and recruitment systems to:
- Analyze and screen resumes automatically
- Extract key qualifications and skills
- Provide structured candidate assessments
- Filter candidates based on specific criteria
- Generate summaries of candidate profiles

### Downstream Use

The model can be integrated into:
- Applicant Tracking Systems (ATS)
- HR management platforms
- Recruitment automation tools
- Candidate matching systems

### Out-of-Scope Use

- Should not be used as the sole decision-maker in hiring processes
- Not intended for discriminatory screening based on protected characteristics
- Not suitable for general-purpose text generation outside of resume/HR context

## How to Get Started with the Model

from transformers import AutoTokenizer, AutoModelForCausalLM
from peft import PeftModel

Load base model and tokenizer
base_model = AutoModelForCausalLM.from_pretrained("meta-llama/Llama-2-7b-hf")
tokenizer = AutoTokenizer.from_pretrained("meta-llama/Llama-2-7b-hf")

Load LoRA adapter
model = PeftModel.from_pretrained(base_model, "kiritps/Advanced-resume-screening")

Example usage
prompt = "Analyze this resume and provide key qualifications: [RESUME TEXT HERE]"
inputs = tokenizer(prompt, return_tensors="pt")
outputs = model.generate(**inputs, max_length=512, temperature=0.7)
response = tokenizer.decode(outputs, skip_special_tokens=True)

text

## Training Details

### Training Data

The model was fine-tuned on a curated dataset of resume-response pairs, designed to teach the model how to:
- Extract relevant information from resumes
- Provide structured analysis of candidate qualifications
- Generate appropriate screening responses

### Training Procedure

#### Training Hyperparameters

- **Training regime:** 4-bit quantization with bfloat16 mixed precision
- **LoRA rank:** 64
- **LoRA alpha:** 16
- **Learning rate:** 2e-4
- **Batch size:** 4
- **Gradient accumulation steps:** 4
- **Training epochs:** Multiple checkpoints saved (3840, 4320, 4800, 5280, 5760 steps)

#### Quantization Configuration

- **Quantization method:** bitsandbytes
- **Load in 4bit:** True
- **Quantization type:** nf4
- **Double quantization:** True
- **Compute dtype:** bfloat16

## Bias, Risks, and Limitations

### Limitations

- Model responses should be reviewed by human recruiters
- May exhibit biases present in training data
- Performance may vary across different industries or job types
- Requires careful prompt engineering for optimal results

### Recommendations

- Use as a screening aid, not a replacement for human judgment
- Regularly audit outputs for potential bias
- Combine with diverse evaluation methods
- Ensure compliance with local employment laws and regulations

## Technical Specifications

### Model Architecture

- **Parameter Count:** ~7B parameters (base) + LoRA adapters
- **Quantization:** 4-bit NF4 quantization

### Compute Infrastructure

#### Hardware
- GPU training environment
- Compatible with consumer and enterprise GPUs

#### Software
- **Framework:** PyTorch
- **PEFT Version:** 0.6.2
- **Transformers:** Latest compatible version
- **Quantization:** bitsandbytes

## Training Procedure

The following `bitsandbytes` quantization config was used during training:
- quant_method: bitsandbytes
- load_in_8bit: False
- load_in_4bit: True
- llm_int8_threshold: 6.0
- llm_int8_skip_modules: None
- llm_int8_enable_fp32_cpu_offload: False
- llm_int8_has_fp16_weight: False
- bnb_4bit_quant_type: nf4
- bnb_4bit_use_double_quant: True
- bnb_4bit_compute_dtype: bfloat16

### Framework Versions
- PEFT 0.6.2
- Transformers (compatible version)
- PyTorch (latest stable)
- bitsandbytes (for quantization)

## Model Card Authors

kiritps

## Model Card Contact

For questions or issues regarding this model, please open an issue in the model repository.