Model Overview

The laion/swesmith-nl2bash-stack-bugsseq is an 8 billion parameter language model, fine-tuned from the Qwen/Qwen3-8B architecture. This model has been specifically trained on a combination of specialized datasets to enhance its capabilities in several key areas.

Key Capabilities

• Natural Language to Bash Translation: Fine-tuned on penfever/GLM-4.6-nl2bash-verified-32ep-32k-reasoning, indicating a strong focus on converting natural language instructions into executable bash commands.
• Code Reasoning: Training on penfever/GLM-4.6-swesmith-32ep-131k-nosumm-reasoning and penfever/GLM-4.6-stackexchange-overflow-sandboxes-32eps-65k-reasoning suggests proficiency in understanding and processing code-related logic and discussions.
• Bug Detection and Inference: The inclusion of penfever/GLM-4.6-inferredbugs-32ep-65k-reasoning dataset points to an ability to identify and reason about potential bugs or errors in code.

Training Details

The model was trained with a learning rate of 4e-05, using an AdamW optimizer with specific beta and epsilon values. It underwent 7 epochs of training with a total batch size of 16 across 8 GPUs, utilizing a cosine learning rate scheduler with a 0.1 warmup ratio. The training leveraged Transformers 4.57.3 and PyTorch 2.9.0+cu128.