README.md

metadata

datasets:
  - UKPLab/DARA-Agentbench

DARA: Decomposition-Alignment-Reasoning Autonomous Language Agent for Question Answering over Knowledge Graphs

Model Information

This model is a fine-tuned semantic parsing LLM agent for KGQA. We fine-tune the llama-2-7B on our curated reasoning trajectory in the Agentbench format: https://huggingface.co/datasets/UKPLab/dara-agentbench.

Model Usage

from transformers import AutoModelForCausalLM

model = AutoModelForCausalLM.from_pretrained( "UKPLab/agentbench-7b", torch_dtype=torch.float16, device_map="auto", cache_dir = "cache")

For more information, please check the repository https://github.com/UKPLab/acl2024-DARA

Hyperparameters

• Learning rate: 2e-5
• Batch size: 4
• Training epochs: 10