Birbal: An efficient 7B instruct-model fine-tuned with curated datasets

TL;DR

Birbal is a fine-tuned 7B instruction model that achieves significant performance gains using a cost-effective, single-GPU setup and curated datasets, demonstrating efficient and transparent large language model adaptation.

Contribution

We introduce Birbal, a 7B instruction-tuned model trained on curated datasets with minimal resources, showcasing improved performance and transparency in LLM fine-tuning.

Findings

35% performance improvement over Qwen-14B

Fine-tuned on a single GPU within 16 hours

Curated diverse high-quality instruction datasets

Abstract

LLMOps incur significant costs due to hardware requirements, hindering their widespread accessibility. Additionally, a lack of transparency in model training methods and data contributes to the majority of models being non-reproducible. To tackle these challenges, the LLM Efficiency Challenge was introduced at NeurIPS Workshop, aiming to adapt foundation models on a diverse set of tasks via fine-tuning on a single GPU (RTX 4090 or A100 with 40GB) within a 24-hour timeframe. In this system description paper, we introduce Birbal, our Mistral-7B based winning model, fine-tuned on a single RTX 4090 for 16 hours. Birbal's success lies in curating high-quality instructions covering diverse tasks, resulting in a 35% performance improvement over second-best Qwen-14B based submission.