test / gpu_training /eval /summary.txt
tanglumy
initial commit of model weights
1a6c70c
FR-Ponder Evaluation Results
==============================
Dataset: gsm8k
--------------------
FR-Ponder:
Accuracy: 0.000
Avg FLOPs: 1368148992
Avg Time: 3.324s
Avg Steps: 104.3
Baseline (α=0.4):
Accuracy: 0.250
Avg FLOPs: 1259616826016
Avg Time: 5.597s
Improvements:
Accuracy: +-0.250
FLOPs reduction: 99.9%
Speedup: 1.68x