Overview

DCAgent/a1-synatra is an 8 billion parameter language model, fine-tuned from the base Qwen/Qwen3-8B architecture. It was trained on a specialized dataset, /e/scratch/jureap59/raoof1/sft_data/hf_hub/datasets--DCAgent--neulab-synatra-sandboxes_glm_4.7_traces_jupiter/snapshots/501f4b499d348087011837642b6842e8b5c29d54_thinking_preprocessed, indicating a focus on specific trace data processing.

Key Training Details

• Base Model: Qwen/Qwen3-8B
• Learning Rate: 4e-05
• Optimizer: AdamW_Torch_Fused with betas=(0.9, 0.98) and epsilon=1e-08
• Scheduler: Cosine learning rate scheduler with 0.1 warmup ratio
• Epochs: 7.0
• Batch Size: 1 (train), 8 (eval) across 16 devices, resulting in a total batch size of 16 (train) and 128 (eval).

Intended Use Cases

While specific intended uses are not detailed in the provided README, the model's fine-tuning on a specialized trace dataset suggests its utility in applications requiring analysis, generation, or understanding of similar data structures. Developers should consider its specialized training for tasks within the domain of neulab-synatra-sandboxes_glm_4.7_traces_jupiter.