An LLM-driven pipeline automatically creates a 5,404-task dataset of data-driven scientific coding problems, and fine-tuning Qwen2.5-Coder on it improves performance on ScienceAgentBench and DiscoveryBench.
pred_results/pred_[code_file_name].[extension]
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.LG 1years
2025 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
AutoSDT: Scaling Data-Driven Discovery Tasks Toward Open Co-Scientists
An LLM-driven pipeline automatically creates a 5,404-task dataset of data-driven scientific coding problems, and fine-tuning Qwen2.5-Coder on it improves performance on ScienceAgentBench and DiscoveryBench.