AgentThink is a systematic evaluation framework that automatically identifies failure patterns in large language models.

First, clone the repository and install the required packages:

cd ThinkingAgent
pip install -r requirements.txt

The framework consists of two main components:

1. format_message.py: Processes and formats interaction logs into a standardized format
2. analyze_agent_think.py: Analyzes the formatted interactions and produces overthinking scores

The framework uses a config.toml file to configure the LLM settings:

[llm]
model = "claude-3-5-sonnet-20241022"
api_key = ""  # Set via environment variable LLM_API_KEY
temperature = 0.0
max_output_tokens = 4096
num_retries = 3
retry_min_wait = 4
retry_max_wait = 10
retry_multiplier = 2

The evaluation process follows these steps:

1. Data Collection: Gather interaction logs from models performing agentic tasks
2. Message Formatting: Use format_message.py to standardize the interaction format
3. Analysis: Run analyze_overthinking.py to evaluate overthinking behaviors
4. Scoring: Generate scores (0-10) for each interaction based on:
   • 0-3: Always interacting with the environment
   • 4-7: Sometimes relies on internal reasoning
   • 8-10: Completely relies on internal reasoning

To analyze a set of interactions:

# Load configuration and initialize LLM
config = load_config()
llm = LLM(config)

# Analyze responses
analyze_responses("path/to/interactions", iteration_number=None)