agurung/cobalt-seeded-rl-base-ramp25-stoppen-gen4k-ep2-ncp10-mvs16
agurung/cobalt-seeded-rl-base-ramp25-stoppen-gen4k-ep2-ncp10-mvs16 is a 4 billion parameter Qwen3-4B-Instruct-based model fine-tuned using OpenRLHF GRPO for code generation. This model was seeded directly from the base Qwen3-4B model without an initial Supervised Fine-Tuning (SFT) step. It is specifically optimized for solving coding problems, demonstrating strong performance on the cobalt-train frontier by achieving the best pass@8 score in its training run. The model's training utilized a binary code-correctness reward signal, making it highly effective for tasks requiring functional code output.
Loading preview...
Model Overview
agurung/cobalt-seeded-rl-base-ramp25-stoppen-gen4k-ep2-ncp10-mvs16 is a 4 billion parameter language model built upon the Qwen/Qwen3-4B-Instruct-2507 architecture. It has been fine-tuned using the OpenRLHF GRPO (Group-Normalized Advantages, no KL penalty) algorithm, directly applying Reinforcement Learning (RL) to the base Qwen3-4B model without an initial Supervised Fine-Tuning (SFT) phase.
Key Capabilities & Training
- Code Generation Optimization: This model is specifically trained for code generation tasks, with its primary reward signal being binary code-correctness (1.0 for passing tests, 0.0 otherwise).
- Performance: It achieved the best pass@8 score in its training run, indicating strong performance in generating correct code solutions.
- Training Data: Validated on the
cobalt-train <=2/64 frontier, which includes 1833 training problems and 112 held-out validation problems that the base model solved on at most 2 of 64 samples. - RL Recipe: Incorporates a stop-properly penalty (truncated samples receive -1.0 reward) and a DAPO overlong penalty for responses nearing the token cap.
- Context Length: Supports a context length of 32768 tokens, allowing for processing longer code prompts and generating more extensive solutions.
Use Cases
This model is particularly well-suited for:
- Automated Code Generation: Generating functional code snippets or complete programs based on problem descriptions.
- Competitive Programming Assistance: Solving coding challenges where correctness is paramount.
- Code Completion & Refinement: Assisting developers by suggesting correct code implementations.