|
|
| --- |
| base_model: Qwen/Qwen2.5-3B-Instruct |
| license: apache-2.0 |
| tags: |
| - qwen |
| - qwen-2.5 |
| - text-generation |
| - conversational |
| - chat |
| - mezaya |
| - colab |
| - gradio |
| widget: |
| - text: "Hello! How can I assist you with project management today?" |
| --- |
| |
| # MezayaAI - Qwen2.5-3B-Instruct (Fine-tuned) |
|
|
| This repository hosts a fine-tuned version of the Qwen/Qwen2.5-3B-Instruct model, specifically adapted for **MezayaAI**, a role-aware AI assistant designed for project/product/agile delivery management. |
|
|
| MezayaAI aims to assist in various project delivery contexts, such as PI Planning, Backlog Refinement, and Governance meetings, by extracting key information, answering questions, and generating role-specific outputs based on a Single Source of Truth (SSOT). |
|
|
| ## Model Details |
|
|
| - **Base Model**: [Qwen/Qwen2.5-3B-Instruct](https://huggingface.co/Qwen/Qwen2.5-3B-Instruct) |
| - **Fine-tuning**: This model has been fine-tuned to incorporate deterministic routing logic, refined SSOT handling for partial decisions and versioning, and strict hallucination guardrails. |
| - **Languages**: English |
|
|
| ## Usage |
|
|
| This model is designed to be integrated into applications that manage project delivery lifecycle. It functions best when provided with context (e.g., meeting transcripts) and clear instructions for extracting structured data or generating role-specific reports. |
|
|
| ### Core Functionality: |
| - **INGEST**: Extracts decisions, actions, risks, dependencies, open questions, and story candidates from unstructured text (e.g., meeting notes). |
| - **QNA**: Answers questions based *only* on the ingested Single Source of Truth (SSOT), stating when information is missing. |
| - **GENERATE**: Produces role-specific outputs (e.g., executive summaries, action plans, user stories) based on the SSOT, adhering to strict output contracts and hallucination guardrails. |
|
|
| ### Example Gradio Demo (`app.py`) |
|
|
| The accompanying `app.py` provides a basic Gradio interface for interacting with the model. It demonstrates how to send user messages and receive responses, integrating the `hf_chat` function to leverage the loaded model. |
|
|
| To run the Gradio demo locally: |
|
|
| 1. Ensure you have `gradio`, `torch`, `transformers`, `accelerate`, and `sentencepiece` installed (`pip install -r requirements.txt`). |
| 2. Run `python app.py` in your terminal. |
|
|
| ## `requirements.txt` |
|
|
| The `requirements.txt` file lists the Python dependencies required to run this model and the Gradio demo: |
|
|
| ``` |
| torch |
| transformers |
| accelerate |
| sentencepiece |
| huggingface_hub |
| gradio |
| |
