Model Overview

HWERI/llama2-exams-orca-sharegpt is a 7 billion parameter language model built upon the Llama2 architecture. It has been specifically fine-tuned using a curated dataset that combines ShareGPT, the exams dataset, and a subset of the Orca dataset. This strategic blend of training data aims to enhance the model's capabilities in both general conversational understanding and performance on knowledge-intensive tasks.

Key Capabilities

• Instruction Following: Fine-tuned on diverse instruction datasets, making it proficient at understanding and executing user commands.
• Conversational AI: Leverages ShareGPT data to improve natural dialogue and response generation.
• Knowledge-Based Reasoning: Incorporates the exams dataset to bolster its ability to answer factual and reasoning-based questions.
• Orca Dataset Integration: Utilizes a subset of the Orca dataset, known for its high-quality, complex instruction-following examples, to further refine its reasoning and problem-solving skills.

Training Details

The model underwent three epochs of fine-tuning using the DeepSpeed Chat toolkit (specifically, the SFT step). A cosine learning rate scheduler was employed, starting with an initial learning rate of 2e-5, until performance plateaued on the validation dataset.

Good For

• Applications requiring robust instruction-following.
• Building conversational agents and chatbots.
• Tasks that benefit from improved reasoning and factual recall, particularly those similar to exam-style questions.
• General-purpose language generation where diverse conversational and instructional capabilities are beneficial.