CodeSteer: Symbolic Augmented Language Models via Code/Text Guidance SymBench comprises 37 symbolic tasks related to the following papers. The specific description of each task is in page 16 19 of the paper'CodeSteer: Symbolic Augmented Language Models via Code/Text Guidance'. This dataset comprises the dataset for finetuning CodeSteerLLM with SFT and DPO datasets, the SymBench with 37 tested tasks, the code scripts to synthesize the SymBench samples. CodeSteer: Symbolic Augmented Language Models via Code/Text Guidance Steering Large Language Models between Code Execution and Textual Reasoning (ICLR'2025) Code    Huggingface🤗    Model Weights Finetune Datasets    SymBench Datasets    SymBench Synthesis Scripts Contents Framework Inspirations Performance Environment Setup LLM API Key Setup Train and Test Models Assistance Citation Framework Figure: CodeSteer on guiding LLM code/text generation to integrate symbolic computing. At each interaction with TaskLLM, it reviews current and previous answers, then provides guidance for the next round. Inspirations Figure: The cases that GPT 4o makes simple mistakes by direct textual reasoning but can r…
We use cookies for essential functionality and analytics. You can accept or reject analytics cookies.Cookie policy