Question Answering on ASQA, HotpotQA, NQ, TriviaQA, POPQA, and BIOASQ
83ASQA ScoreOSCAR-llama
Evaluation Results
| Method | Links | |||||||
|---|---|---|---|---|---|---|---|---|
| OSCAR-llamaSetting=standalone, Decoder=Mistral-7B2025.03 | 83 | 64 | 80 | 91 | 66 | 80 | 77 | |
| RECOMPBackbone=Mistral-7B2025.03 | 83 | 61 | 78 | 89 | 58 | 83 | 75 | |
| OSCAR-8-LayersSetting=standalone, Decoder=Mistral-7B2025.03 | 82 | 64 | 81 | 91 | 64 | 79 | 77 | |
| OSCAR-10-LayersSetting=standalone, Decoder=Mistral-7B2025.03 | 82 | 64 | 81 | 91 | 64 | 80 | 77 | |
| No compressionBackbone=Mistral-7B2025.03 | 82 | 62 | 80 | 90 | 61 | 82 | 76 | |
| ProvenceBackbone=Mistral-7B2025.03 | 82 | 60 | 80 | 89 | 58 | 85 | 76 | |
| OSCAR-llamaBackbone=Mistral-7B2025.03 | 82 | 64 | 81 | 91 | 65 | 81 | 77 | |
| OSCAR-5-LayersBackbone=Mistral-7B2025.03 | 82 | 62 | 79 | 90 | 63 | 77 | 76 | |
| OSCAR-8-LayersBackbone=Mistral-7B2025.03 | 82 | 64 | 80 | 91 | 64 | 79 | 77 | |
| OSCAR-llamaBackbone=Qwen-7B2025.03 | 82 | 62 | 79 | 90 | 65 | 81 | 76 | |
| No compressionBackbone=Mistral-24B2025.03 | 82 | 71 | 80 | 92 | 70 | 85 | 80 | |
| OSCAR-llamaBackbone=Mistral-24B2025.03 | 82 | 65 | 82 | 92 | 67 | 84 | 79 | |
| OSCAR-llamaSetting=e2e, Decoder=Mistral-7B2025.03 | 81 | 63 | 79 | 91 | 66 | 80 | 77 | |
| OSCAR-8-LayersSetting=e2e, Decoder=Mistral-7B2025.03 | 81 | 63 | 79 | 90 | 64 | 78 | 76 | |
| OSCAR-10-LayersSetting=e2e, Decoder=Mistral-7B2025.03 | 81 | 65 | 82 | 91 | 66 | 78 | 77 | |
| PISCOBackbone=Mistral-7B2025.03 | 81 | 60 | 78 | 89 | 57 | 79 | 74 | |
| OSCAR-8-LayersBackbone=Qwen-7B2025.03 | 81 | 61 | 78 | 90 | 64 | 80 | 76 | |
| No compressionBackbone=Qwen-7B2025.03 | 80 | 67 | 78 | 91 | 65 | 84 | 78 | |
| OSCAR-5-LayersBackbone=Llama-1B2025.03 | 71 | 53 | 70 | 85 | 55 | 72 | 68 | |
| No compressionBackbone=Llama-1B2025.03 | 69 | 48 | 66 | 81 | 52 | 76 | 65 |