Model Overview

DCAgent/a1-codeactinstruct is an 8 billion parameter instruction-tuned model, fine-tuned from the Qwen/Qwen3-8B base architecture. It was developed by DCAgent and specifically trained on the neulab-codeactinstruct-sandboxes_glm_4.7_traces_jupiter dataset. This specialized training suggests an emphasis on tasks requiring code generation, execution, and interactive agentic reasoning within sandboxed environments.

Key Characteristics

• Base Model: Qwen/Qwen3-8B
• Parameter Count: 8 billion parameters
• Context Length: 32768 tokens, enabling processing of substantial codebases or complex instructions.
• Training Data: Fine-tuned on a dataset focused on code-action instructions and traces, indicating a specialization in agentic code-related tasks.

Training Details

The model underwent 7 epochs of training using a learning rate of 4e-05 and an AdamW optimizer. It was trained across 16 devices with a total batch size of 16, utilizing a cosine learning rate scheduler with a 0.1 warmup ratio. The training leveraged Transformers 4.57.6 and Pytorch 2.9.1+cu130.