DCAgent/a1-nebius_swe_agent
The DCAgent/a1-nebius_swe_agent is an 8 billion parameter language model fine-tuned from Qwen/Qwen3-8B. This model is specifically optimized for software engineering tasks, leveraging a specialized dataset of agent trajectories and sandboxes. With a context length of 32768 tokens, it is designed to assist in complex code-related problem-solving and development workflows.
Loading preview...
Overview
This model, DCAgent/a1-nebius_swe_agent, is an 8 billion parameter language model built upon the Qwen/Qwen3-8B architecture. It has been fine-tuned using a unique dataset derived from DCAgent/neulab-nebius-swe-agent-trajectories-sandboxes_glm_4.7_traces_jupiter, which suggests a specialization in software engineering tasks and agent-based problem-solving.
Key Capabilities
- Software Engineering Focus: Fine-tuned on a dataset of agent trajectories and sandboxes, indicating an optimization for automated software development and problem-solving.
- Base Model: Leverages the robust capabilities of the Qwen3-8B model as its foundation.
- Context Length: Supports a substantial context window of 32768 tokens, beneficial for handling larger codebases or complex problem descriptions.
Training Details
The model was trained with a learning rate of 4e-05, a batch size of 1 per device across 16 GPUs (total batch size 16), and for 7 epochs. It utilized the AdamW_TORCH_FUSED optimizer with a cosine learning rate scheduler and a warmup ratio of 0.1. The training environment included Transformers 4.57.6 and Pytorch 2.9.1+cu130.