ReMCTS

Developers and researchers can use ReMCTS to search for and evaluate Python or C++17 programs using an OpenAI-compatible model API.

Availability

Installable from the public GitHub repository

Pricing model

Not established

Deployment

Runs locally; models are called through an OpenAI-compatible Chat Completions API

Privacy

Not established

Licence

MIT

• ReMCTS searches a tree of candidate programs using selection, expansion, execution, reflection, and backpropagation. ^[1]

  <p align="center"><img src="assets/overview.png" width="100%"><br><em>ReMCTS: selection, expansion, execution, reflection and backpropagation over a tree of candidate programs.</em></p>

• Candidate code can be executed in Python or C++17 subprocesses. ^[1]
• ReMCTS supports searches with visible tests and a no-tests setting. ^[1]

  On HumanEval and MBPP-Sanitized, ReMCTS with visible tests beats direct generation in 8 of 10 model–dataset pairs; controlled sampling, repair, tree-search and memory ablations show where the gains come from and where they stop.

• The repository provides CLI commands and a Python API. ^[1]
• Models are called through any OpenAI-compatible Chat Completions API, so no GPU is needed. ^[1]

  Python 3.10+ is enough; models are called through any OpenAI-compatible Chat Completions API, so no GPU is needed.

• The executors add timeouts and process isolation but are not a security sandbox. ^[1]

  The executors add timeouts and process isolation, but they are not a security sandbox.