README.md

metadata

title: Resume Normalizer Trainer
emoji: 📝
colorFrom: blue
colorTo: green
sdk: gradio
sdk_version: 5.6.0
app_file: app.py
pinned: false
license: apache-2.0
hardware: 4xL4

Resume Normalizer Trainer

Fine-tune a Flan-T5 model for resume entity normalization and deduplication.

Features

• Company Name Normalization: Handle mergers, acquisitions, and rebranding (e.g., "Facebook" → "Meta Platforms Inc.")
• Job Title Standardization: Recognize equivalent roles and seniority levels (e.g., "SWE" → "Software Engineer")
• Skills Normalization: Standardize technology names and abbreviations (e.g., "JS" → "JavaScript")
• Binary Equivalency Detection: Determine if two entities refer to the same thing

Model Details

• Base Model: Google Flan-T5 (instruction-tuned for better zero-shot performance)
• Fine-tuning Method: LoRA (Low-Rank Adaptation) for efficient training
• Parameters: 250M (T5-Base) or 770M (T5-Large)
• Training Data: 9,302 high-quality examples (478 manual + 8,824 synthetic)

Usage

1. Check that training data is available using the "Check Data" tab
2. Enter your HuggingFace token and username
3. Select model size and training epochs
4. Click "Start Training" and monitor progress in the "Training Status" tab
5. Once complete, your model will be available on HuggingFace Hub

Expected Performance

• Inference Speed: <100ms per query
• Accuracy: >90% on entity normalization tasks
• Memory Usage: ~1GB (T5-Base) or ~3GB (T5-Large)

Hardware Requirements

This Space runs on 4xL4 GPUs (96GB total VRAM) for efficient distributed training.