|
--- |
|
license: mit |
|
language: multilingual |
|
tags: |
|
- conversational |
|
- healthcare |
|
- conversation-ending |
|
- LLM |
|
--- |
|
|
|
# ProdocAI/EndConvo-health-1b-GGUF-v1 |
|
|
|
`EndConvo-health-1b-GGUF-v1` is a fine-tuned version of the `Llama3.2-1B` model, trained on a dataset of healthcare-related conversations with the purpose of identifying whether a conversation has ended. This model helps to avoid unnecessary responses from larger language models by detecting closing statements. |
|
|
|
## Model Details |
|
|
|
- **Model Name**: `EndConvo-health-1b-GGUF-v1` |
|
- **Base Model**: `Llama3.2-1B` |
|
- **Number of Parameters**: 1 Billion |
|
- **Dataset**: Custom dataset of 4,000 rows focused on healthcare conversations |
|
- **Training Data Statistics**: |
|
- **Total Conversations**: 11,798 |
|
- **Chat Count**: 94,472 |
|
- **Average Chats per Conversation**: ~8 |
|
- **Languages**: Includes `en`, `mr`, `te`, `hi`, `bn`, among others (detailed in Language Map section) |
|
|
|
## Model Objective |
|
|
|
The model identifies if a healthcare-related conversation has reached a natural conclusion to prevent unnecessary responses from a larger LLM. The model is trained to output: |
|
- **True**: Conversation has ended. |
|
- **False**: Conversation is still active. |
|
|
|
## Dataset Overview |
|
|
|
This healthcare-focused conversational dataset includes 11,798 unique conversations, with an average of 8 messages per conversation. The dataset consists of conversations in a variety of languages with the following breakdown: |
|
- **English** (`en`): 78,404 messages |
|
- **Marathi** (`mr`): 2,092 messages |
|
- **Hindi** (`hi`): 2,857 messages |
|
- ... and others as per the Language Map section. |
|
|
|
### Example Input Format |
|
|
|
Input to the model should be provided in the following format: |
|
|
|
```plaintext |
|
"Below is the conversation between the bot and user: |
|
|
|
user: Please send me one bottle |
|
bot: Hi, I am Vaidushi and how can I help you today regarding your interest to buy Madhavprash? |
|
bot: Here is the link to purchase your Madhavprash https://madhavprash.store |
|
|
|
user: 👆COD not possible here |
|
bot: Currently, we do not support Cash on Delivery (COD) for purchases. You can complete your purchase using other available payment methods on our website. |
|
|
|
bot: Thanks for your order, it will be delivered to you within 2-3 working days. Dosage Guidelines... |
|
user: Thanks 🙏🤝 madam..... Kailas Varsekar ..." |
|
|
