Long-context Question Answering on LooGLE 900 (test)
82.8Success RateCVT-RL
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| CVT-RLDescription=per-seed CVT-RL success, Split (train/dev/test)=4300/650/900, Individual Seeds=81.9/83.5/82.6/83.1/82.72026.06 | 82.8 | 4.2 | |
| IMDescription=information-matched counterfactual-process RL, Split (train/dev/test)=4300/650/9002026.06 | 78.6 | — | |
| CMDescription=compute-matched non-causal RL, Split (train/dev/test)=4300/650/9002026.06 | 75.3 | — |