Overview

DCAgent/a1-go_browse_wa is an 8 billion parameter language model, fine-tuned from the Qwen/Qwen3-8B architecture. This model was trained using a specific dataset, /e/scratch/jureap59/raoof1/sft_data/hf_hub/datasets--DCAgent--neulab-go-browse-wa-sandboxes_glm_4.7_traces_jupiter/snapshots/dd1b9192915cfa62681657f80a3c4cccece9c18f_thinking_preprocessed, suggesting a specialization in tasks related to its training data, potentially involving web browsing or agent-based interactions. It supports a substantial context length of 32768 tokens.

Training Details

The model was trained with the following key hyperparameters:

• Learning Rate: 4e-05
• Batch Size: 1 (train), 8 (eval)
• Optimizer: ADAMW_TORCH_FUSED with betas=(0.9, 0.98) and epsilon=1e-08
• Scheduler: Cosine learning rate scheduler with a 0.1 warmup ratio
• Epochs: 7.0
• Devices: 16 GPUs were used for distributed training, resulting in a total effective training batch size of 16 and evaluation batch size of 128.

Framework Versions

The training utilized:

• Transformers 4.57.6
• Pytorch 2.9.1+cu130
• Datasets 4.7.0
• Tokenizers 0.22.2