kacperwikiel's picture
Update tokenizer, configuration and training reports
d6a86af verified
Raw History Blame Contribute Delete
174 Bytes
seed = 1337
block = 2048
micro_batch = 16
accumulation = 2
adam_lr = 0.0015
muon_lr = 0.02
weight_decay = 0.1
clip = 1.0
checkpoint_every = 500
log_every = 20
compile = true