Context Traceback on NarrativeQA LongBench
96PrecisionTracLLM
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| TracLLMLLM=Llama-3.1-8B-Instruct2025.08 | 96 | 84 | 644.7 | |
| AttnTraceLLM=Llama-3.1-8B-Instruct2025.08 | 96 | 85 | 44 | |
| STCLLM=Llama-3.1-8B-Instruct2025.08 | 95 | 83 | 5.4 | |
| AT2LLM=Llama-3.1-8B-Instruct2025.08 | 86 | 76 | 4 | |
| LIME/CCLLM=Llama-3.1-8B-Instruct2025.08 | 78 | 69 | 648.3 | |
| DAALLM=Llama-3.1-8B-Instruct2025.08 | 72 | 64 | 3.3 | |
| ShapleyLLM=Llama-3.1-8B-Instruct2025.08 | 71 | 63 | 1,043.2 | |
| Self-CLLM=Llama-3.1-8B-Instruct2025.08 | 25 | 22 | 3.4 | |
| LOOLLM=Llama-3.1-8B-Instruct2025.08 | 21 | 18 | 464.4 | |
| GradientLLM=Llama-3.1-8B-Instruct2025.08 | 5 | 5 | 10.8 |