NhatCuong22/qwen2.5-7b-proofdag-sft
NhatCuong22/qwen2.5-7b-proofdag-sft is a 7.6 billion parameter language model, fine-tuned from Qwen/Qwen2.5-7B-Instruct. This model is specifically optimized for evaluating logical statements, classifying conclusions as True, False, or Uncertain based on provided premises. It excels in tasks requiring proof verification and logical reasoning within a multi-turn chat context.
Loading preview...
Overview
This model, NhatCuong22/qwen2.5-7b-proofdag-sft, is a specialized fine-tune of the Qwen/Qwen2.5-7B-Instruct base model, featuring 7.6 billion parameters and a 32K context length. Its primary purpose is to assess the validity of proposed conclusions given a set of premises, categorizing them as True, False, or Uncertain. This makes it particularly adept at tasks involving logical deduction and proof verification.
Key Capabilities
- Logical Reasoning: Specialized in evaluating the truthfulness of conclusions based on provided premises.
- Proof Verification: Designed to determine if a conclusion logically follows from given statements.
- Multi-turn Chat: Trained on multi-turn chat data, enabling interactive logical assessment.
- Classification: Outputs classifications of "True", "False", or "Uncertain" for logical propositions.
Training Details
The model was fine-tuned on the ProofDAG dataset, utilizing 5640 training examples and 330 validation examples, all structured as multi-turn chats. The training involved 3 epochs with a global batch size of 128 and a maximum sequence length of 4096 tokens, achieving a final training loss of 0.207 and an evaluation loss of 0.251.
Good for
- Automated logical proof checking.
- Applications requiring assessment of statement validity.
- Educational tools for logic and critical thinking.
- Systems needing to classify conclusions based on explicit premises.