v0.2.0

What's Changed

• Deps: Upgraded slime support to v0.3.0.
• Release: Bumped version to v0.2.0.

Full Changelog: v0.1.3...v0.2.0

v0.1.3

What's Changed

• Deps: Upgraded slime support to v0.2.4.
• Release: Bumped version to v0.1.3.

Full Changelog: v0.1.2...v0.1.3

v0.1.2

What's Changed

• Deps: Upgraded slime support to v0.2.3.
• Qwen3.5 Support: Added support for Qwen3.5 model series
• MCP Reconnection Mechanism: Implemented automatic reconnection for MCP connections to improve stability and reliability
• Release: Bumped version to v0.1.2.

Full Changelog: v0.1.1...v0.1.2

v0.1.1

What's Changed

• Deps: Upgraded slime support to v0.2.2.
• Release: Bumped version to v0.1.1.

Full Changelog: v0.1.0...v0.1.1

v0.1.0

Initial Release - ArenaRL & MCP Support

v0.1.0 is the first public release of qqr (a.k.a. hilichurl), a lightweight extension for slime enabling open-ended agent evolution via ArenaRL.

🌟 Key Features

• ArenaRL Algorithm: Full implementation of tournament-based topologies (Anchor-Based, Round-Robin, Swiss-System, etc.) to tackle discriminative collapse.
• MCP Support: Seamless integration with the Model Context Protocol to decouple LLM inference from tool environments.
• High Performance: Leverages slime for high-throughput, distributed rollout generation and training.

📋 Compatibility

Package	Version	Note
slime	v0.2.1	✅ Tested

Note: While newer versions of slime might work, v0.2.1 is the verified version for this release.

🚀 Quick Start

Run the example travel agent experiment:

bash scripts/travel/run-qwen3-8B.sh

📖 Further Reading

• Paper: ArenaRL: Scaling RL for Open-Ended Agents via Tournament-based Relative Ranking
• Blog: ArenaRL Blog Post