qijiashen/e2e-terminal-rl-qwen35-4b-miles-harbor-sb-pDsMUHN7IBXz9iGYo10Wh5
The qijiashen/e2e-terminal-rl-qwen35-4b-miles-harbor-sb-pDsMUHN7IBXz9iGYo10Wh5 model is a 4.5 billion parameter model, derived from Qwen3.5, specifically developed by qijiashen as an auto-pushed agentrl-bench submission. This model represents a final selected two-update GRPO model, demonstrating performance in a benchmark environment with specific validation scores. Its primary use case is within reinforcement learning benchmarks, serving as an agent submission for evaluation.
Loading preview...
Overview
This model, qijiashen/e2e-terminal-rl-qwen35-4b-miles-harbor-sb-pDsMUHN7IBXz9iGYo10Wh5, is a 4.5 billion parameter variant of Qwen3.5, developed by qijiashen. It functions as an auto-pushed submission for the agentrl-bench benchmark, specifically a final selected two-update GRPO (Generalized Reinforcement Policy Optimization) model. The model's development involved an elapsed time of 28306 seconds, with its weights identified by 624d63447064.
Key Capabilities
- Reinforcement Learning Agent: Designed to operate as an agent within a reinforcement learning benchmark environment.
- GRPO Optimization: Utilizes a two-update GRPO model, indicating a specific approach to policy optimization.
- Benchmark Submission: Represents a submission for the
agentrl-benchbenchmark, with performance metrics recorded.
Performance Metrics
- Development Validation: Achieved 55/106 on development validation, compared to a base score of 50/106.
- Frozen Fresh Validation: Scored 78/160 on frozen fresh validation, against a base score of 83/160.
Good for
- Reinforcement Learning Research: Suitable for researchers and developers exploring GRPO models and their application in benchmark settings.
- Agent Benchmarking: Ideal for evaluating and comparing reinforcement learning agents within the
agentrl-benchframework. - Understanding RL Submissions: Provides an example of a benchmark agent's submission, complete with logs, provenance, and OCR audit disclosure.