This repo contains core algorithm implementations of ExploraCoder and the Torchdata-Manual benchmark we constructed for evaluating ExploraCoder and its baselines' performance on multi-API invocations. Some case studies from our paper regarding different approaches solving one example task are also provided at example.md.

In the provided Torchdata-Manual, Torchdata-Github, MonkBeat-Eval benchmarks, the entries represent:

• task requirement (entry 'task')
• full programming problem. including task requirement and code context (entry 'prompt')
• canonical solution for the problem (entry 'canonical_solution')
• test csaes for problem (entry 'test')

In the data files, we currently provide partial data samples. Full data will be released after paper being accepted. Other Implementation details can be found in our paper.