Views
No views yet
nv3[v0] (1700) | nv4[v1-2k] (4000) | nv4[v1-210k] (b1b2: 4000)8168 ***** train metrics *****
8169 epoch = 100.0
8170 total_flos = 334833095087GF
8171 train_loss = 0.0776
8172 train_runtime = 4:53:00.40
8173 train_samples_per_second = 56.955
8174 train_steps_per_second = 0.893
8176 ***** eval metrics *****
8177 epoch = 100.0
8178 eval_accuracy = 0.7487
8179 eval_loss = 1.9947
8180 eval_runtime = 0:00:12.56
8181 eval_samples_per_second = 140.622
8182 eval_steps_per_second = 2.9451BASE_MODEL = "facebook/dinov2-with-registers-large"
2DATASET = "distill-lab/COMBINE_nai-distill_00-01_eagle.library"
3TASK = "classification"
4# using single card to train it, so had to do higher batch size
5
6cmd = f"""python -m trainlib.hf_trainer.cli \
7 --model_name_or_path {BASE_MODEL} \
8 --dataset_name {DATASET} \
9 --output_dir distill-n4_00-01_combined_cls_v1b2-100e \
10 --remove_unused_columns False \
11 --label_column_name star \
12 --task {TASK} \
13 --do_train \
14 --do_eval \
15 --eval_strategy steps \
16 --eval_steps 100 \
17 --learning_rate 1e-5 \
18 --num_train_epochs 100 \
19 --per_device_train_batch_size 64 \
20 --per_device_eval_batch_size 48 \
21 --logging_strategy steps \
22 --logging_steps 2 \
23 --save_total_limit 1 \
24 --seed 1337 \
25 --lr_scheduler_type cosine \
26 --dataloader_num_workers 16 \
27 --ignore_mismatched_sizes True
28
29"""
30
31rest = f"""
32--push_to_hub: True \
33 --push_to_hub_organization distill-lab \
34 --hub_model_id nai-distill_00-01_combined_eagle_{TASK} \
35 --hub_strategy "end"""
36
37print(cmd)
38!{cmd}