step=32 loss=0.7803519368171692

step=320 loss=0.5616167783737183

step=288 loss=1.5417312383651733

step=256 loss=0.8266531825065613

step=224 loss=1.5066956281661987

step=192 loss=1.1840404272079468

step=160 loss=0.9588625431060791

step=128 loss=1.2087572813034058

step=96 loss=1.340285301208496

step=64 loss=1.172095775604248

step=32 loss=1.1201095581054688

32 batches; loss=1.12

step=176 loss=1.1972198486328125

step=160 loss=1.2805689573287964

step=144 loss=1.4531817436218262

step=128 loss=1.2950183153152466

step=112 loss=1.7769852876663208

step=96 loss=1.256285548210144

step=80 loss=1.703926682472229

step=64 loss=1.73197340965271

step=48 loss=1.0865358114242554

step=32 loss=1.59828782081604

step=16 loss=1.092730164527893

step=16 loss=1.1869220733642578

batch=421 loss=2.16 (1.17)

training interrupted and continued with new epoch. batch=331 loss=2.32 (1.84)

batch 512, batchsize=19 loss=2.27 (1.52)

batch 333, batchsize=19 loss=2.13 (1.49)

batch 202, batchsize=19 loss=2.38

batch 182, batchsize=19 loss=2.76

new tokenizer. batch 132, batchsize=19 loss=2.81

new tokenizer. batch 132, batchsize=19 loss=2.81

batch 49, batchsize=19 loss=1.67

batch 43, batchsize=19 loss=1.45

batch 29, batchsize=19 loss=1.71

batch 24, batchsize=19 loss=2.04

batch 21, batchsize=19 loss=1.63

batch 11, batchsize=19 loss=1.74

2.34