DCAgent/a1-agenttuning_mind2web
DCAgent/a1-agenttuning_mind2web is an 8 billion parameter language model fine-tuned from Qwen/Qwen3-8B. This model is specifically optimized for agentic tasks, leveraging the neulab-agenttuning-mind2web-sandboxes_glm_4.7_traces_jupiter dataset. It is designed to enhance performance in web-based agent interactions and automated task execution.
Loading preview...
Overview
This model, DCAgent/a1-agenttuning_mind2web, is an 8 billion parameter language model derived from the Qwen/Qwen3-8B architecture. It has undergone fine-tuning on a specialized dataset, /e/scratch/jureap59/raoof1/sft_data/hf_hub/datasets--DCAgent--neulab-agenttuning-mind2web-sandboxes_glm_4.7_traces_jupiter/snapshots/18a00618fba76dd32bdea57571d69b0a5ee386ad_thinking_preprocessed, which suggests a focus on agentic capabilities, particularly within web environments.
Key Capabilities
- Agentic Task Optimization: Fine-tuned on a dataset related to 'agenttuning' and 'mind2web', indicating a specialization in tasks requiring autonomous interaction with web interfaces or complex multi-step processes.
- Foundation Model: Built upon the robust Qwen3-8B base, inheriting its general language understanding and generation capabilities.
Training Details
The model was trained with the following hyperparameters:
- Learning Rate: 4e-05
- Batch Size: 1 (train), 8 (eval) with a total distributed batch size of 16 (train) and 128 (eval) across 16 GPUs.
- 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
Good For
- Developing and deploying AI agents for web automation.
- Tasks requiring understanding and execution within web-based sandboxes.
- Research into agentic AI and fine-tuning large language models for specific interactive environments.