|
|
--- |
|
|
language: en |
|
|
license: apache-2.0 |
|
|
tags: |
|
|
- gpt2 |
|
|
- distilgpt2 |
|
|
- knowledge-distillation |
|
|
- tally |
|
|
- accounting |
|
|
- conversational |
|
|
- business |
|
|
- transformer |
|
|
- language-model |
|
|
- safetensors |
|
|
model_type: gpt2 |
|
|
library_name: transformers |
|
|
datasets: custom |
|
|
pipeline_tag: text-generation |
|
|
base_model: |
|
|
- openai-community/gpt2-large |
|
|
--- |
|
|
|
|
|
# 💼 TallyPrimeAssistant — Distilled GPT-2 Model |
|
|
|
|
|
This is a distilled GPT-2-based conversational model fine-tuned on FAQs and navigation instructions from **TallyPrime**, a leading business accounting software used widely in India. The model is designed to help users get quick and accurate answers about using features in TallyPrime like GST, e-invoicing, payroll, and more. |
|
|
|
|
|
--- |
|
|
|
|
|
## 🧠 Model Summary |
|
|
|
|
|
- **Teacher Model**: `gpt2-large` |
|
|
- **Student Model**: `distilgpt2` |
|
|
- **Distillation Method**: Knowledge Distillation using Hugging Face's Transformers and custom training pipeline |
|
|
- **Training Dataset**: Internal dataset of Q&A pairs and system navigation steps from TallyPrime documentation and usage |
|
|
- **Format**: `safetensors` (secure and fast) |
|
|
- **Tokenizer**: Byte-Pair Encoding (BPE), same as GPT-2 |
|
|
|
|
|
--- |
|
|
|
|
|
## 🚀 Example Usage |
|
|
|
|
|
```python |
|
|
from transformers import AutoTokenizer, AutoModelForCausalLM |
|
|
|
|
|
model = AutoModelForCausalLM.from_pretrained("Jayanthram/TallyPrimeAssistant") |
|
|
tokenizer = AutoTokenizer.from_pretrained("Jayanthram/TallyPrimeAssistant") |
|
|
|
|
|
prompt = "How to enable GST in Tally Prime?" |
|
|
inputs = tokenizer(prompt, return_tensors="pt") |
|
|
output = model.generate(**inputs, max_new_tokens=60) |
|
|
print(tokenizer.decode(output[0], skip_special_tokens=True)) |
