learning_rate: 0.0002
train_batch_size: 2
eval_batch_size: 2
seed: 42
distributed_type: multi-GPU
num_devices: 16
gradient_accumulation_steps: 8
total_train_batch_size: 256
total_eval_batch_size: 32
optimizer: Use OptimizerNames.ADAMW_TORCH 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
2.1927	0.0582	10	2.0806
0.1655	0.1164	20	0.1102
0.0307	0.1745	30	0.0408
0.0305	0.2327	40	0.0329
0.0291	0.2909	50	0.0311
0.0255	0.3491	60	0.0300
0.0325	0.4073	70	0.0292
0.0259	0.4655	80	0.0285
0.0305	0.5236	90	0.0276
0.0275	0.5818	100	0.0267
0.0274	0.64	110	0.0263
0.0253	0.6982	120	0.0256
0.0196	0.7564	130	0.0252
0.0223	0.8145	140	0.0248
0.0259	0.8727	150	0.0246
0.0238	0.9309	160	0.0242
0.0228	0.9891	170	0.0241
0.0163	1.0465	180	0.0243
0.0155	1.1047	190	0.0240
0.0238	1.1629	200	0.0239
0.0186	1.2211	210	0.0241
0.022	1.2793	220	0.0239
0.0174	1.3375	230	0.0239
0.0212	1.3956	240	0.0236
0.0194	1.4538	250	0.0235
0.0176	1.512	260	0.0233
0.0223	1.5702	270	0.0234
0.019	1.6284	280	0.0236
0.0174	1.6865	290	0.0234
0.0166	1.7447	300	0.0233
0.0153	1.8029	310	0.0234
0.0138	1.8611	320	0.0234
0.0212	1.9193	330	0.0231
0.0225	1.9775	340	0.0230
0.0104	2.0349	350	0.0232
0.0102	2.0931	360	0.0247
0.0116	2.1513	370	0.0251
0.0091	2.2095	380	0.0251
0.0139	2.2676	390	0.0253
0.0088	2.3258	400	0.0254
0.0099	2.384	410	0.0253
0.0089	2.4422	420	0.0254
0.0128	2.5004	430	0.0256
0.0101	2.5585	440	0.0254
0.0115	2.6167	450	0.0254
0.0113	2.6749	460	0.0256
0.0105	2.7331	470	0.0257
0.0086	2.7913	480	0.0258
0.0081	2.8495	490	0.0258
0.0099	2.9076	500	0.0258
0.0134	2.9658	510	0.0257

airesearch/Qwen3-32B-medqa

Qwen3-32B-medqa