Overview

This model, laion/rl_pymethods2test-r2egym_terminus-structured, is an 8 billion parameter Qwen3-based language model that has undergone extensive Reinforcement Learning (RL) training. It is specifically designed to handle structured tool calls, utilizing a terminus-structured agent equipped with bash, view, edit, create, and search tools. The training pipeline involved multiple stages, starting with Supervised Fine-Tuning (SFT) on datasets like r2egym, nl2bash, and swesmith, followed by progressive RL training steps, culminating in 156 steps on the pymethods2test dataset.

Key Capabilities

• Automated Code Repair: Achieves 37-42% pass@3 on SWEBench-100, indicating strong ability in fixing code.
• Test-Writing: Demonstrates high proficiency in generating tests, with 91-97% pass@8 on Pymethods2test.
• Tool Integration: Leverages a terminus-structured agent for interacting with environments via bash, view, edit, create, and search commands.
• Context Handling: Supports a substantial 32k token context length (24k input + 8k output), beneficial for complex coding tasks.

Good For

This model is particularly well-suited for use cases requiring automated code generation, editing, and test-writing. Its specialized RL training makes it effective for tasks involving code repair and the creation of robust test suites, while maintaining general code-editing capabilities.