Stage 2 Final: Best loss=1.4674 at step 17200 (early stopped)

Stage 2 Final: Best loss=1.4674 at step 17200 (early stopped)

Best model metadata (loss=1.4469)

Best loss checkpoint (step 23200, loss=1.4469)

Best model metadata (loss=1.4522)

Best loss checkpoint (step 21200, loss=1.4522)

Best model metadata (loss=1.4608)

Best loss checkpoint (step 19200, loss=1.4608)

Best model metadata (loss=1.4674)

Best loss checkpoint (step 17200, loss=1.4674)

Best model metadata (loss=1.5020)

Best loss checkpoint (step 15200, loss=1.5020)

Best model metadata (loss=1.5283)

Best loss checkpoint (step 13200, loss=1.5283)

Best model metadata (loss=1.5561)

Best loss checkpoint (step 11200, loss=1.5561)

Best model metadata (loss=1.6002)

Best loss checkpoint (step 9200, loss=1.6002)

Best model metadata (loss=1.6772)

Best loss checkpoint (step 7200, loss=1.6772)

Best model metadata (loss=1.8523)

Best loss checkpoint (step 5200, loss=1.8523)

Best model metadata (loss=2.4044)

Best loss checkpoint (step 3200, loss=2.4044)

Best Stage 2 model (step 1300, score=0.000000)

Best Stage 2 model (step 1300, score=0.000000)

Best Stage 2 model (step 1300, score=0.000000)

Best Stage 2 model (step 1300, score=0.000000)

Best Stage 2 model (step 1300, score=0.000000)

Best Stage 2 model (step 1300, score=0.000000)

Best Stage 2 FP model (step 1300, score=0.000000) - Stage-3 compatible

Best Stage 2 FP model (step 1300, score=0.000000) - Stage-3 compatible

Best Stage 2 FP model (step 1300, score=0.000000) - Stage-3 compatible

Best Stage 2 FP model (step 1300, score=0.000000) - Stage-3 compatible

Best Stage 2 FP model (step 1300, score=0.000000) - Stage-3 compatible

Best Stage 2 FP model (step 1300, score=0.000000) - Stage-3 compatible

Best Stage 2 1.58-bit quantized (step 1300, score=0.000000) - Deployment only

Best Stage 2 3-bit quantized (step 1300, score=0.000000) - Deployment only

Best Stage 2 4-bit quantized (step 1300, score=0.000000) - Deployment only

Best Stage 2 FP model (step 1300, score=0.000000) - Stage-3 compatible

Best Stage 2 4-bit quantized (step 1300, score=0.000000) - Deployment only

Best Stage 2 FP model (step 1300, score=0.000000) - Stage-3 compatible

Best Stage 2 model (step 1300, score=0.000000)

Best Stage 2 model (step 1300, score=0.000000)