agurung/cobalt-seeded-rl-base-ramp25-stoppen-gen4k-ep2-ncp10-mgroot16

TEXT GENERATIONPricing:Input $0.4 / Cached $0.08 / Output $0.8Concurrent Unit Cost:1Model Size:4BQuant:BF16Context Size:32kTool Calling:SupportedPublished:Sep 3, 2026Architecture:Transformer Featherless Exclusive Cold

The agurung/cobalt-seeded-rl-base-ramp25-stoppen-gen4k-ep2-ncp10-mgroot16 is a 4 billion parameter Qwen3-based model, fine-tuned using OpenRLHF GRPO. This checkpoint is specifically optimized for code generation tasks, leveraging a binary code-correctness reward signal. It excels at solving programming problems that the base model struggled with, making it suitable for applications requiring robust code output.

Loading preview...

Model Overview

The agurung/cobalt-seeded-rl-base-ramp25-stoppen-gen4k-ep2-ncp10-mgroot16 is a specialized reinforcement learning (RL) checkpoint based on the Qwen3-4B architecture. Developed by agurung, this model was fine-tuned using the OpenRLHF GRPO algorithm, directly applied to the base Qwen3-4B model without an initial Supervised Fine-Tuning (SFT) seed. It represents the best performing checkpoint by pass@8 in its training run.

Key Capabilities and Training

  • Code Generation Focus: The model is specifically trained and validated on the cobalt-train dataset, targeting programming problems that the base model could only solve on a limited number of samples.
  • Reward Signal: Training utilizes a binary code-correctness reward, meaning the model is rewarded 1.0 if its generated program passes problem tests and 0.0 otherwise, directly optimizing for functional code.
  • RL Algorithm: Employs GRPO (group-normalized advantages, no KL penalty) with specific anti-truncation and overlong response penalties to guide generation.
  • Context Length: Supports a maximum of 4096 new tokens per rollout during training, indicating its capacity for generating substantial code blocks.

Use Cases

This model is particularly well-suited for:

  • Automated Code Generation: Generating solutions for programming challenges or specific coding tasks.
  • Code Completion and Repair: Assisting developers by producing correct and functional code snippets.
  • Educational Tools: Creating systems that can solve or demonstrate solutions to coding problems.