Sheila-Assistant

Sheila-Assistant

Model Description

Sheila-Assistant is a conversational AI assistant developed by Sheila Wafula. It was created by fine-tuning Qwen2.5-1.5B-Instruct using QLoRA (Low-Rank Adaptation) and the PEFT library on a custom conversational instruction dataset.

Features

• Conversational AI
• Python programming assistance
• Code debugging
• Code explanation
• Professional email writing
• Essay and report generation
• Text summarization
• Mathematics problem solving
• Machine Learning explanations
• Career guidance
• Study assistance

Base Model

• Qwen2.5-1.5B-Instruct

Training Details

• Fine-tuning Method: QLoRA (PEFT)
• Quantization: 4-bit BitsAndBytes
• Frameworks: Transformers, PEFT, TRL
• Hardware: NVIDIA Tesla T4 GPU
• Training Steps: Approximately 964 optimization steps
• Context Length: Up to 2048 tokens

Intended Use

Sheila-Assistant is designed to help users with:

• Answering general knowledge questions
• Writing and explaining code
• Generating professional emails
• Writing reports and essays
• Summarizing text
• Explaining technical concepts
• Providing study support
• General conversational assistance

Limitations

This model was trained for approximately 964 optimization steps due to computational resource and time constraints. While it performs well on many instruction-following tasks, additional training could further improve its conversational behavior and specialization.

Usage

This repository contains a LoRA adapter for Qwen2.5-1.5B-Instruct.

Load the base Qwen2.5-1.5B-Instruct model first, then load this adapter using the PEFT library.

Author

Sheila Wafula

Bachelor of Science in Economic

License

Apache-2.0