DCFA: Dual-view Causal-inspired Attribution for Failure Reasoning in LLM-based Multi-agent Systems. EMNLP 2026

Datasets

We use the datasets in the Who&When benchmark, where the datasets are publicated on Hugging Face

Requirements

To install requirements:

pip install -r requirements.txt

Inference

Please ensure that you specify the AutoFA method (--method) in the corresponding sections of the code before executing it.

• Models DCFA supports two inference modes:
• Cloud API mode for large models (e.g., DeepSeek-R1-671B, GPT-5)
• vLLM or Local deployment for smaller models (e.g., Qwen3-Coder-30B)

Run

Used for models that can be loaded locally:

python Automated_FA/DCFA_failure_attribution.py --model #MODEL --api_key #API_KEY --base_url #BASE_URL --local_llm_type "local" --local_model_path #LOCAL_MODEL_PATH 

Used when the model is hosted via a vLLM inference server:

python Automated_FA/DCFA_failure_attribution.py --model #MODEL --api_key #API_KEY --base_url #BASE_URL --local_llm_type "vllm" --vllm_api_key #VLLM_API_KEY --vllm_base_url #VLLM_BASE_URL