laion/r2egym-bugsseq

TEXT GENERATIONConcurrency Cost:1Model Size:8BQuant:FP8Ctx Length:32kPublished:Dec 6, 2025License:apache-2.0Architecture:Transformer Open Weights Cold

The laion/r2egym-bugsseq model is an 8 billion parameter language model fine-tuned from Qwen/Qwen3-8B. It was specifically trained on the penfever/glm-4.6-r2egym-32ep-32k and penfever/GLM-4.6-inferredbugs-32eps-65k datasets. This model is optimized for tasks related to bug detection and resolution within the R2E (Reasoning to Execution) Gym environment, leveraging its 32768 token context length for comprehensive analysis.

Loading preview...

Model Overview

laion/r2egym-bugsseq is an 8 billion parameter language model developed by laion, building upon the robust architecture of Qwen/Qwen3-8B. This model has been specifically fine-tuned to address challenges within the R2E (Reasoning to Execution) Gym, focusing on bug-related tasks.

Key Capabilities

  • Specialized Fine-tuning: The model underwent fine-tuning on two distinct datasets: penfever/glm-4.6-r2egym-32ep-32k and penfever/GLM-4.6-inferredbugs-32eps-65k. This targeted training suggests a strong capability in identifying, analyzing, and potentially resolving bugs or issues within code or reasoning sequences.
  • Extended Context Window: With a context length of 32768 tokens, the model can process and understand extensive code snippets or problem descriptions, crucial for complex bug analysis.
  • Foundation Model: Leveraging the Qwen3-8B base, it inherits strong general language understanding and generation capabilities, which are then specialized for its bug-related domain.

Training Details

The model was trained with a learning rate of 4e-05, using AdamW_TORCH_FUSED optimizer, and a cosine learning rate scheduler with a 0.1 warmup ratio over 7 epochs. The training utilized a distributed setup across 8 GPUs with a total batch size of 16.

Potential Use Cases

This model is particularly suited for applications requiring:

  • Automated bug detection in code.
  • Assisting in debugging processes by identifying potential issues.
  • Analyzing reasoning traces for errors or inconsistencies, especially within environments like R2E Gym.
  • Code quality assurance and vulnerability assessment.

Finetunes
1 models