AgentTuning is an instruction-tuning method developed at Tsinghua University that enhances LLMs with agent capabilities while preserving their general language abilities. Introduced by Zeng et al. (2023), it produces AgentLM models where the 70B variant achieves performance comparable to GPT-3.5-turbo on unseen agent tasks.¹⁾²⁾³⁾

Fine-tuning LLMs exclusively on agent-specific data risks catastrophic forgetting of general capabilities. AgentTuning solves this through a hybrid instruction-tuning strategy: mixing agent interaction trajectories (AgentInstruct) with general-domain instructions during training. This is the first systematic attempt at instruction-tuning LLMs across multiple agent task types.

The process has two main components:

1. AgentInstruct Dataset

A curated dataset of 1,866 verified interaction trajectories created in three stages:

• Instruction Generation: Diverse task instructions across agent domains
• Trajectory Interaction: LLMs execute tasks, producing thought-action-observation chains
• Trajectory Filtering: Only high-quality, successful trajectories with valid CoT reasoning are retained

2. Hybrid Instruction-Tuning

Agent and general-domain data are mixed at a controlled ratio <latex>\alpha</latex>:

<latex>\mathcal{D}_{\text{train}} = \alpha \cdot \mathcal{D}_{\text{agent}} + (1 - \alpha) \cdot \mathcal{D}_{\text{general}}</latex>

The training loss is standard supervised fine-tuning:

<latex>\mathcal{L} = -\sum_{(x,y) \in \mathcal{D}_{\text{train}}} \sum_{t=1}^{|y|} \log P_\theta(y_t | x, y_{<t})</latex>

This hybrid approach prevents overfitting to agent patterns while building robust planning, reasoning, and tool-use capabilities.

The dataset spans multiple agent task domains:

• Web browsing: Navigating and interacting with web pages
• Database operations: Querying and manipulating structured data
• Tool use: Invoking APIs and external tools
• Operating system tasks: File manipulation, command execution
• Knowledge-grounded QA: Multi-step reasoning with retrieval

AgentTuning was applied to the Llama 2 series, producing AgentLM models:

• Generalization: Strong performance on both held-in and held-out (unseen) agent tasks
• Preserved general abilities: No significant degradation on standard NLP benchmarks
• Error reduction: Significant decrease in formatting errors, duplicated generation, and refusal to answer
• Bridges the gap between open-source and commercial LLMs for agent applications
• Only 1,866 trajectories needed – demonstrating data efficiency

# AgentTuning-style hybrid instruction tuning
from transformers import AutoModelForCausalLM, Trainer, TrainingArguments
from datasets import concatenate_datasets, load_dataset
 
# Load agent and general instruction datasets
agent_data = load_dataset('THUDM/AgentInstruct')
general_data = load_dataset('general_instructions', split='train')
 
# Hybrid mixing at ratio alpha
alpha = 0.3  # 30% agent data, 70% general
n_agent = int(alpha * len(general_data))
agent_subset = agent_data['train'].select(range(min(n_agent, len(agent_data['train']))))
mixed_data = concatenate_datasets([agent_subset, general_data])
mixed_data = mixed_data.shuffle(seed=42)
 
# Fine-tune Llama 2
model = AutoModelForCausalLM.from_pretrained('meta-llama/Llama-2-70b-hf')
trainer = Trainer(
    model=model,
    args=TrainingArguments(
        output_dir='./agentlm-70b',
        num_train_epochs=3,
        per_device_train_batch_size=4,
        learning_rate=2e-5,
    ),
    train_dataset=mixed_data,
)
trainer.train()

• Zeng et al. (2023) - AgentTuning: Enabling Generalized Agent Capabilities in LLMs
• AgentTuning Project Page (Tsinghua)
• AgentTuning GitHub Repository
• ACL 2024 Findings Paper

• FireAct: Agent Fine-tuning
• LLM Agent Survey: Brain/Perception/Action Framework
• Retroformer: Policy Gradient Agent Optimization
• Agent Benchmarks and Evaluation