Overview
cris177/llama-2-7b-Arguments is a Llama 2-7B model specifically fine-tuned to understand and process logical arguments. Developed by cris177, this model's primary function is to analyze text and extract the core components of an argument, including premises, conclusions, argument type, and validity. The training involved a unique dataset generation process focusing on logical structures like modus ponens and modus tollens.
Key Capabilities
- Argument Detection: Identifies and extracts arguments from natural language.
- Component Analysis: Pinpoints premises and conclusions within an argument.
- Argument Type Classification: Distinguishes between argument types such as modus ponens and modus tollens.
- Validity Assessment: Evaluates the logical validity of identified arguments.
- Structured Output: Provides a structured breakdown of argument analysis, as demonstrated in the example template.
Training Methodology
The model was fine-tuned on a custom dataset, cris177/Arguments, which was systematically built through several steps:
- Statement and Negation Generation: LLMs were used to create diverse statements and their negations.
- Modus Ponens and Modus Tollens Construction: Arguments following these logical forms were generated using the statements.
- Labeled Dataset Creation: A comprehensive dataset was compiled, labeling each argument with its premises, conclusion, type, and validity.
Evaluation
The fine-tuned model is intended for evaluation using the Open-LLM-Leaderboard, which includes benchmarks like AI2 Reasoning Challenge, HellaSwag, MMLU, TruthfulQA, Winogrande, and GSM8k. Evaluation results are currently pending.
Good for
- Logical Reasoning Tasks: Ideal for applications requiring the identification and analysis of logical structures.
- Educational Tools: Can be used in tools designed to teach logic and critical thinking.
- Text Analysis: Useful for extracting structured argumentative content from various texts.