agurung/cobalt-seeded-rl-base-ramp25-stoppen-gen4k-ep2-ncp5-q4v2-iid16-s2

TEXT GENERATIONPricing:Input $0.4 / Cached $0.08 / Output $0.8Concurrent Unit Cost:1Model Size:4BQuant:BF16Context Size:32kTool Calling:SupportedPublished:Sep 14, 2026Architecture:Transformer Featherless Exclusive Cold

The agurung/cobalt-seeded-rl-base-ramp25-stoppen-gen4k-ep2-ncp5-q4v2-iid16-s2 is a 4 billion parameter reinforcement learning checkpoint based on Qwen3-4B-Instruct-2507, developed by agurung. This model was trained using OpenRLHF GRPO with a focus on binary code-correctness as its reward signal. It is specifically optimized for solving coding problems, particularly those where the base model struggled, and is noted as the best checkpoint by pass@8 in its development run.

Loading preview...

Overview

This model, agurung/cobalt-seeded-rl-base-ramp25-stoppen-gen4k-ep2-ncp5-q4v2-iid16-s2, is a 4 billion parameter reinforcement learning (RL) checkpoint derived from the Qwen/Qwen3-4B-Instruct-2507 base model. It was developed by agurung using the OpenRLHF GRPO algorithm, with RL applied directly to the base Qwen3-4B without an initial Supervised Fine-Tuning (SFT) seed. This checkpoint represents global step 36 of its RL training run.

Key Capabilities

  • Code Generation Optimization: Specifically trained to improve performance on coding problems, particularly those where the base model initially struggled.
  • Reinforcement Learning: Utilizes a binary code-correctness reward signal (1.0 for passing tests, 0.0 otherwise) to refine its code generation abilities.
  • GRPO Algorithm: Employs the Group-Normalized Advantages (GRPO) algorithm without a KL penalty for its RL training.
  • Anti-Truncation Shaping: Incorporates a stop-properly penalty where truncated samples receive a -1.0 reward, and an overlong penalty for responses nearing the token cap.

Good for

  • Code Generation Tasks: Ideal for scenarios requiring accurate code generation, especially for problems that are challenging for standard instruction-tuned models.
  • Improving Base Model Performance: Useful for developers looking to enhance the code-solving capabilities of the Qwen3-4B model family through RL.
  • Research in RL for Code: Provides a checkpoint for studying the effects of GRPO and specific reward shaping techniques on code generation performance.

Finetunes
1 models