agurung/cobalt-seeded-rl-base-ramp25-stoppen-gen4k-ep2-ncp10-groot8
agurung/cobalt-seeded-rl-base-ramp25-stoppen-gen4k-ep2-ncp10-groot8 is a 4 billion parameter Qwen3-based model, fine-tuned using OpenRLHF GRPO for enhanced code generation. This model specializes in solving programming problems, particularly those where the base Qwen3-4B model struggled. It leverages a binary code-correctness reward signal and advanced RL techniques to improve performance on challenging coding tasks.
Loading preview...
Model Overview
agurung/cobalt-seeded-rl-base-ramp25-stoppen-gen4k-ep2-ncp10-groot8 is a 4 billion parameter model based on Qwen/Qwen3-4B-Instruct-2507, fine-tuned using the OpenRLHF GRPO (Group-Normalized Advantages, No KL Penalty) reinforcement learning algorithm. This checkpoint represents the best performing iteration by pass@8 in its training run, specifically targeting improved code generation capabilities.
Key Capabilities
- Enhanced Code Generation: The model is specifically trained to solve programming problems, particularly those that the base Qwen3-4B model found difficult.
- Reinforcement Learning Optimization: It utilizes a binary code-correctness reward signal, meaning it's rewarded for generating programs that pass problem-specific tests.
- Advanced RL Techniques: Incorporates a "stop-properly" penalty for truncated samples and a DAPO overlong penalty for responses nearing the token limit, designed to improve output quality and prevent premature truncation.
When to Use This Model
- Code Generation Tasks: Ideal for applications requiring the generation of correct and functional code, especially for problems where a high pass rate is critical.
- Improving Base Model Performance: This model serves as an RL-tuned version of Qwen3-4B, offering improved performance on specific coding challenges compared to its base.
- Research in RL for Code: Useful for researchers exploring the application of GRPO and similar RL techniques for code generation and problem-solving.