agurung/cobalt-seeded-rl-base-ramp25-stoppen-gen4k-ep2-ncp10-vs8

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

agurung/cobalt-seeded-rl-base-ramp25-stoppen-gen4k-ep2-ncp10-vs8 is a 4 billion parameter Qwen3-4B-Instruct-based language model, fine-tuned using OpenRLHF GRPO for improved code generation. This model specializes in solving programming problems by optimizing for binary code-correctness, making it particularly effective for tasks requiring functional code output. It was seeded directly from the base Qwen3-4B model and represents the best checkpoint by pass@8 in its training run, focusing on problems the base model struggled with.

Loading preview...

Model Overview

This model, agurung/cobalt-seeded-rl-base-ramp25-stoppen-gen4k-ep2-ncp10-vs8, is a 4 billion parameter Qwen3-4B-Instruct-based language model. It has been fine-tuned using the OpenRLHF GRPO (Group-Normalized Advantages, No KL Penalty) reinforcement learning algorithm. Notably, it was seeded directly from the base Qwen3-4B model without an initial Supervised Fine-Tuning (SFT) step, applying RL directly to enhance its capabilities.

Key Capabilities

  • Code Generation Optimization: The model's primary strength lies in generating correct code, having been trained with a reward signal based on binary code-correctness (1.0 for passing tests, 0.0 otherwise).
  • Targeted Problem Solving: It was specifically trained and validated on a frontier of problems that the base Qwen3-4B model struggled with, aiming to improve performance on challenging coding tasks.
  • RL Training Methodology: Utilizes advanced RL techniques including a stop-properly penalty for truncated samples and an overlong penalty for responses nearing the token limit, ensuring more robust and complete outputs.

Good For

  • Automated Code Generation: Ideal for applications requiring the generation of functional and test-passing code snippets.
  • Improving Base Model Weaknesses: Suitable for scenarios where a base model's performance on specific, harder coding problems needs significant enhancement.
  • Research in RL for Code: Provides a practical example of applying GRPO with specific penalty mechanisms for code generation tasks.

Finetunes
3 models