← Back to Home

Train Custom Model

🎯 What is this?
Train a custom embedding model from your domain-specific text. The model will learn terminology and synonyms specific to your field (medical, legal, finance, etc.).

📋 Requirements:

Minimum: 1000 words (3+ paragraphs)
Recommended: 5000-10000 words (50-100 paragraphs)
File format: .txt or .md
Max file size: 10 MB

⚠️ Important: Training takes 5-60 seconds depending on corpus size. The page will redirect when complete. Larger corpora (5000+ words) produce better synonym learning.

Upload Training Corpus

Model Name *

Use alphanumeric characters and underscores only (e.g., medical_uti, finance_basics)

Training File *

Upload a .txt or .md file with your domain text (1000+ words recommended)

⚙️ Advanced Settings (Optional)

Embedding Dimension

Output vector size (64-256). Default: 128. Higher = more capacity but larger model.

Max Features

TF-IDF vocabulary size (1000-10000). Default: 5000. Higher = more terms but slower.

💡 Tips for Best Results

Use domain-specific text: Medical records, legal documents, financial reports, etc.
Include synonyms: Use both technical and common terms (pyrexia/fever, renal/kidney)
More is better: 50-100 paragraphs produce better models than 10-20
Consistent formatting: Separate paragraphs with blank lines
Clean text: Remove excessive formatting, headers, footers

📚 Example Use Cases

Medical: Train on medical textbooks to understand pyrexia↔fever, renal↔kidney, UTI↔bladder infection
Legal: Train on legal documents to understand plaintiff↔claimant, liability↔responsibility
Finance: Train on financial reports to understand liquidity↔cash flow, equity↔stock
Technical: Train on documentation to understand domain-specific jargon