Saadhyam Context-Aware Email Generator
A fine-tuned FLAN-T5 model for generating professional emails from context-aware user requests.
Model
Base model:
google/flan-t5-small
The model was fine-tuned using context-aware email-generation data.
Context Information
The model uses:
- Audience
- Context
- Email type
- User request
Supported audiences include:
- candidate
- customer
- employees
- general
- students
Supported contexts include:
- customer_service
- education
- general
- recruitment
- workplace
Training
Training samples: 960
Validation samples: 120
Test samples: 120
Training epochs: 5
Learning rate: 2e-5
Batch size: 8
Evaluation
Test loss: 0.248608
Test perplexity: 1.2822
Template accuracy on test set: 100%
Context compatibility on test set: 94.17%
Real-world generation test:
- JSON validity: 100%
- Output completeness: 100%
- Template accuracy: 100%
- Context compatibility: 100%
Important
The reported accuracy values are based on the Saadhyam evaluation dataset and real-world test cases used during development.
Additional evaluation on unseen production data is recommended.
Repository
Hugging Face repository:
https://huggingface.co/likhitha7274/saadhyam-context-email-generator