Question Answering on PubMedQA (Context Influence, ROUGE-L, BERTScore, FactKB)
115.78Context InfluenceCID (LLaMA 3 8B IT)
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| CID (LLaMA 3 8B IT)Decoding lambda=1.52024.10 | 115.78 | 20.88 | 75.21 | 63.08 | |
| CID (Gemma 2 9B IT)Decoding lambda=1.52024.10 | 111.07 | 18.52 | 74.03 | 35.6 | |
| CID (OPT 1.3B)Decoding lambda=1.52024.10 | 97.95 | 16.96 | 72.88 | 48.81 | |
| CID (GPT-Neo 1.3B)Decoding lambda=1.52024.10 | 77.91 | 18.91 | 74.08 | 68.54 | |
| CID (LLaMA 3 8B)Decoding lambda=1.52024.10 | 70.91 | 18.79 | 74.41 | 56.76 | |
| CID (Gemma 2 9B IT)Decoding lambda=1.02024.10 | 70.1 | 20.05 | 74.97 | 41.6 | |
| CID (LLaMA 3 8B IT)Decoding lambda=1.02024.10 | 66.39 | 21.47 | 75.47 | 56.64 | |
| CID (OPT 1.3B)Decoding lambda=1.02024.10 | 45.66 | 16.51 | 72.81 | 37.38 | |
| CID (GPT-Neo 1.3B)Decoding lambda=1.02024.10 | 38.79 | 18.47 | 73.65 | 52.36 | |
| CID (LLaMA 3 8B)Decoding lambda=1.02024.10 | 37.01 | 19.2 | 74.66 | 49.63 | |
| CID (Gemma 2 9B IT)Decoding lambda=0.52024.10 | 26.68 | 18.52 | 74.03 | 35.6 | |
| CID (LLaMA 3 8B IT)Decoding lambda=0.52024.10 | 17.26 | 20.17 | 74.51 | 51.64 | |
| CID (LLaMA 3 8B)Decoding lambda=0.52024.10 | 16.69 | 17.73 | 73.33 | 44.71 | |
| CID (OPT 1.3B)Decoding lambda=0.52024.10 | 13.2 | 15.41 | 72.13 | 31.4 | |
| CID (GPT-Neo 1.3B)Decoding lambda=0.52024.10 | 11.2 | 16.26 | 72.32 | 35.66 |