learning_rate: 2e-05
train_batch_size: 8
eval_batch_size: 8
seed: 42
distributed_type: multi-GPU
gradient_accumulation_steps: 8
total_train_batch_size: 64
optimizer: Use OptimizerNames.ADAMW_TORCH_FUSED with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
lr_scheduler_type: cosine
lr_scheduler_warmup_ratio: 0.1
num_epochs: 3.0

Training Loss	Epoch	Step	Validation Loss
0.6842	0.1875	300	0.7154
0.7339	0.3751	600	0.7645
0.7369	0.5626	900	0.7523
0.7618	0.7501	1200	0.7474
0.7158	0.9376	1500	0.7349
0.6541	1.1250	1800	0.7247
0.603	1.3125	2100	0.7165
0.6339	1.5001	2400	0.6994
0.6052	1.6876	2700	0.6871
0.6189	1.8751	3000	0.6707
0.5176	2.0625	3300	0.6660
0.485	2.2500	3600	0.6554
0.4961	2.4376	3900	0.6486
0.4842	2.6251	4200	0.6442
0.5102	2.8126	4500	0.6411
0.4737	3.0	4800	0.6408

Gueule-d-ange/llama32-3b-redo-sft

llama32-3b-redo-sft