DeepSeek VL2 Small

test • Self-reported

test • Self-reported

test • Self-reported

testmini • Self-reported

val • Self-reported

test • Self-reported

ru test • Self-reported

cn test • Self-reported

Standard Evaluation AI: Standard evaluation • Self-reported

Standard evaluation AI: Standard evaluation • Self-reported

Standard evaluation AI: I'm an AI assistant that answers questions. • Self-reported

Standard evaluation AI: Standard evaluation • Self-reported

Standard evaluation AI: ChatGPT assisted solving math problems Math problems are a significant challenge for state-of-the-art LLMs. This project studies how LLMs solve math problems. We explore direct solving and chain-of-thought (CoT) prompting, aiming to understand and improve solution approaches. Methods: 1. Direct Solving: We give the model a question and ask for an answer. 2. Chain-of-Thought (CoT): We instruct the model to break down the problem into steps. We study: - Problem solving approach (structured vs. unstructured reasoning) - Common error patterns - Reasoning path analysis - Impact of formula knowledge • Self-reported

val • Self-reported