cakra84/Agrease-Chatbot

license: apache-2.0 tags:

text-generation

conversational

mistral

fine-tuned

chatbot

bangkit widget:

text: "Hello! I'm looking for recommendations on agricultural products."

Fine-Tuned Mistral Model for Agrease Application This repository contains a fine-tuned version of a Mistral Large Language Model, specifically adapted for the "Agrease" application. The model was developed as part of a capstone project for Bangkit Academy 2024 Batch 2.

The primary goal of this project was to create a specialized conversational AI capable of assisting users within the Agrease application, likely by providing information and recommendations based on data from various marketplaces.

Model Description Base Model: The model is a fine-tuned version of a Mistral v3 Large Language Model.

Fine-tuning Task: The model was fine-tuned for conversational question-answering and recommendations.

Training Data: The training data was collected by scraping various online marketplaces using Python libraries such as BeautifulSoup and Scrapy.

Performance: The fine-tuning process achieved a final training loss of 11%.

Intended Use This model is intended to be used as a chatbot or a conversational agent within a larger application. It can answer user queries, provide product recommendations, and engage in domain-specific conversations related to the "Agrease" application's scope.

How to Use You can use this model with the transformers library for text generation.

from transformers import pipeline

Load the text generation pipeline from the Hugging Face Hub

Replace "your-username/model-name" with the actual model path

generator = pipeline('text-generation', model='your-username/model-name')

Example prompt

prompt = "What are the best fertilizers for rice paddies in a tropical climate?"

Generate a response

response = generator(prompt, max_length=150, num_return_sequences=1)

print(response[0]['generated_text'])

Training Data The dataset used for fine-tuning was created by scraping publicly available data from various e-commerce and marketplace websites. The scraping was performed using custom Python scripts with BeautifulSoup and Scrapy. The collected data was then processed and formatted into a conversational format suitable for training a large language model.

Training Procedure The fine-tuning was performed using the PyTorch framework on the collected dataset. The training focused on minimizing the cross-entropy loss to improve the model's ability to generate relevant and coherent responses in a conversational context. The final model achieved a training loss of 0.11