This article discusses the author’s progress in contributing to Google Summer of Code 2024 at Red Hen Lab. The author shares their journey with the TV News Chat LLM project, focusing on community bonding, data extraction, cleaning, and filtering processes. They also describe the dataset creation using the Self-Instruct framework and fine-tuning the Large Language Model (LLM) on English context. The author details the training process, which includes Supervised Fine-Tuning (SFT) and Parameter-Efficient Fine-Tuning (PEFT) using the LoRA configuration. They also mention using vLLM for inference and merging LORA adapters with base model weights. The article concludes with plans for the next phase of making the model adapt to multilingual questions. Special mentions are given to the mentor and others who supported the author during the project. The author’s code and PR can be found on GitHub. The author expresses gratitude to their mentor and colleagues for their support and insightful discussions throughout the project.
Source link
Source link: https://medium.com/@jaintarun7/google-summer-of-code-2024-mid-term-evaluations-5df8b9291b19?source=rss——llm-5
Google Summer of Code 2024 Midterm Evaluations by Tarun Jain #GSoC2024
