模型微调异常 #362
Comments
您好！我也遇到这个问题，训练acc很低，请问您解决了吗？ |
|
我也遇到这样的问题，你们都解决了吗？acc一直很低 |
{{ message }}
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2024-10-10,14:19:42 | INFO | Rank 0 | train LMDB file contains 4000 images and 4000 pairs.
2024-10-10,14:19:42 | INFO | Rank 0 | val LMDB file contains 500 images and 500 pairs.
2024-10-10,14:19:42 | INFO | Rank 0 | Params:
2024-10-10,14:19:42 | INFO | Rank 0 | accum_freq: 1
2024-10-10,14:19:42 | INFO | Rank 0 | aggregate: True
2024-10-10,14:19:42 | INFO | Rank 0 | batch_size: 128
2024-10-10,14:19:42 | INFO | Rank 0 | bert_weight_path: None
2024-10-10,14:19:42 | INFO | Rank 0 | beta1: 0.9
2024-10-10,14:19:42 | INFO | Rank 0 | beta2: 0.98
2024-10-10,14:19:42 | INFO | Rank 0 | checkpoint_path: /workspace/code/experiments/demo_data_finetune_vit-b-16_roberta-base_bs128_8gpu/checkpoints
2024-10-10,14:19:42 | INFO | Rank 0 | clip_weight_path: None
2024-10-10,14:19:42 | INFO | Rank 0 | context_length: 52
2024-10-10,14:19:42 | INFO | Rank 0 | debug: False
2024-10-10,14:19:42 | INFO | Rank 0 | device: cuda:0
2024-10-10,14:19:42 | INFO | Rank 0 | distillation: False
2024-10-10,14:19:42 | INFO | Rank 0 | eps: 1e-06
2024-10-10,14:19:42 | INFO | Rank 0 | freeze_vision: False
2024-10-10,14:19:42 | INFO | Rank 0 | gather_with_grad: False
2024-10-10,14:19:42 | INFO | Rank 0 | grad_checkpointing: False
2024-10-10,14:19:42 | INFO | Rank 0 | kd_loss_weight: 0.5
2024-10-10,14:19:42 | INFO | Rank 0 | local_device_rank: 0
2024-10-10,14:19:42 | INFO | Rank 0 | log_interval: 1
2024-10-10,14:19:42 | INFO | Rank 0 | log_level: 20
2024-10-10,14:19:42 | INFO | Rank 0 | log_path: /workspace/code/experiments/demo_data_finetune_vit-b-16_roberta-base_bs128_8gpu/out_2024-10-10-06-19-39.log
2024-10-10,14:19:42 | INFO | Rank 0 | logs: /workspace/code/experiments/
2024-10-10,14:19:42 | INFO | Rank 0 | lr: 0.0005
2024-10-10,14:19:42 | INFO | Rank 0 | mask_ratio: 0
2024-10-10,14:19:42 | INFO | Rank 0 | max_epochs: 100
2024-10-10,14:19:42 | INFO | Rank 0 | max_steps: 3200
2024-10-10,14:19:42 | INFO | Rank 0 | name: demo_data_finetune_vit-b-16_roberta-base_bs128_8gpu
2024-10-10,14:19:42 | INFO | Rank 0 | num_workers: 4
2024-10-10,14:19:42 | INFO | Rank 0 | precision: amp
2024-10-10,14:19:42 | INFO | Rank 0 | rank: 0
2024-10-10,14:19:42 | INFO | Rank 0 | report_training_batch_acc: True
2024-10-10,14:19:42 | INFO | Rank 0 | reset_data_offset: True
2024-10-10,14:19:42 | INFO | Rank 0 | reset_optimizer: True
2024-10-10,14:19:42 | INFO | Rank 0 | resume: /code/xx/Chinese-CLIP/pretrained_weights/clip_cn_vit-b-16.pt
2024-10-10,14:19:42 | INFO | Rank 0 | save_epoch_frequency: 1
2024-10-10,14:19:42 | INFO | Rank 0 | save_step_frequency: 999999
2024-10-10,14:19:42 | INFO | Rank 0 | seed: 123
2024-10-10,14:19:42 | INFO | Rank 0 | skip_aggregate: False
2024-10-10,14:19:42 | INFO | Rank 0 | skip_scheduler: False
2024-10-10,14:19:42 | INFO | Rank 0 | teacher_model_name: None
2024-10-10,14:19:42 | INFO | Rank 0 | text_model: RoBERTa-wwm-ext-base-chinese
2024-10-10,14:19:42 | INFO | Rank 0 | train_data: /workspace/code/demo_data/lmdb/train
2024-10-10,14:19:42 | INFO | Rank 0 | use_augment: True
2024-10-10,14:19:42 | INFO | Rank 0 | use_bn_sync: False
2024-10-10,14:19:42 | INFO | Rank 0 | use_flash_attention: False
2024-10-10,14:19:42 | INFO | Rank 0 | val_data: /workspace/code/demo_data/lmdb/valid
2024-10-10,14:19:42 | INFO | Rank 0 | valid_batch_size: 128
2024-10-10,14:19:42 | INFO | Rank 0 | valid_epoch_interval: 1
2024-10-10,14:19:42 | INFO | Rank 0 | valid_num_workers: 1
2024-10-10,14:19:42 | INFO | Rank 0 | valid_step_interval: 150
2024-10-10,14:19:42 | INFO | Rank 0 | vision_model: ViT-B-16
2024-10-10,14:19:42 | INFO | Rank 0 | warmup: 100
2024-10-10,14:19:42 | INFO | Rank 0 | wd: 0.001
2024-10-10,14:19:42 | INFO | Rank 0 | world_size: 1
2024-10-10,14:19:42 | INFO | Rank 0 | Use GPU: 0 for training
2024-10-10,14:19:42 | INFO | Rank 0 | => begin to load checkpoint '/code/xx/Chinese-CLIP/pretrained_weights/clip_cn_vit-b-16.pt'
2024-10-10,14:20:15 | INFO | Rank 0 | => loaded checkpoint '/code/xx/Chinese-CLIP/pretrained_weights/clip_cn_vit-b-16.pt' (epoch 15 @ 0 steps)
2024-10-10,14:20:23 | INFO | Rank 0 | Global Steps: 1/3200 | Train Epoch: 1 [128/4096 (3%)] | Loss: 5.140202 | Image2Text Acc: 5.47 | Text2Image Acc: 3.91 | Data Time: 6.993s | Batch Time: 8.429s | LR: 0.000005 | logit_scale: 4.605 | Global Batch Size: 128
2024-10-10,14:20:24 | INFO | Rank 0 | Global Steps: 2/3200 | Train Epoch: 1 [256/4096 (6%)] | Loss: 5.020943 | Image2Text Acc: 6.25 | Text2Image Acc: 5.47 | Data Time: 0.503s | Batch Time: 0.813s | LR: 0.000010 | logit_scale: 4.605 | Global Batch Size: 128
2024-10-10,14:20:25 | INFO | Rank 0 | Global Steps: 3/3200 | Train Epoch: 1 [384/4096 (9%)] | Loss: 4.862580 | Image2Text Acc: 6.25 | Text2Image Acc: 3.91 | Data Time: 0.025s | Batch Time: 0.385s | LR: 0.000015 | logit_scale: 4.605 | Global Batch Size: 128
2024-10-10,14:20:25 | INFO | Rank 0 | Global Steps: 4/3200 | Train Epoch: 1 [512/4096 (12%)] | Loss: 5.083204 | Image2Text Acc: 4.69 | Text2Image Acc: 3.12 | Data Time: 0.036s | Batch Time: 0.330s | LR: 0.000020 | logit_scale: 4.605 | Global Batch Size: 128
2024-10-10,14:20:29 | INFO | Rank 0 | Global Steps: 5/3200 | Train Epoch: 1 [640/4096 (16%)] | Loss: 4.380543 | Image2Text Acc: 4.69 | Text2Image Acc: 9.38 | Data Time: 4.093s | Batch Time: 4.395s | LR: 0.000025 | logit_scale: 4.605 | Global Batch Size: 128
2024-10-10,14:20:31 | INFO | Rank 0 | Global Steps: 6/3200 | Train Epoch: 1 [768/4096 (19%)] | Loss: 4.561520 | Image2Text Acc: 3.91 | Text2Image Acc: 6.25 | Data Time: 1.478s | Batch Time: 1.772s | LR: 0.000030 | logit_scale: 4.605 | Global Batch Size: 128
2024-10-10,14:20:31 | INFO | Rank 0 | Global Steps: 7/3200 | Train Epoch: 1 [896/4096 (22%)] | Loss: 4.347610 | Image2Text Acc: 3.91 | Text2Image Acc: 7.81 | Data Time: 0.036s | Batch Time: 0.330s | LR: 0.000035 | logit_scale: 4.605 | Global Batch Size: 128
2024-10-10,14:20:32 | INFO | Rank 0 | Global Steps: 8/3200 | Train Epoch: 1 [1024/4096 (25%)] | Loss: 4.256195 | Image2Text Acc: 7.03 | Text2Image Acc: 3.91 | Data Time: 0.040s | Batch Time: 0.334s | LR: 0.000040 | logit_scale: 4.605 | Global Batch Size: 128
2024-10-10,14:20:37 | INFO | Rank 0 | Global Steps: 9/3200 | Train Epoch: 1 [1152/4096 (28%)] | Loss: 4.305431 | Image2Text Acc: 2.34 | Text2Image Acc: 2.34 | Data Time: 4.561s | Batch Time: 4.863s | LR: 0.000045 | logit_scale: 4.605 | Global Batch Size: 128
2024-10-10,14:20:39 | INFO | Rank 0 | Global Steps: 10/3200 | Train Epoch: 1 [1280/4096 (31%)] | Loss: 4.286503 | Image2Text Acc: 3.91 | Text2Image Acc: 7.81 | Data Time: 1.666s | Batch Time: 1.964s | LR: 0.000050 | logit_scale: 4.605 | Global Batch Size: 128
2024-10-10,14:20:39 | INFO | Rank 0 | Global Steps: 11/3200 | Train Epoch: 1 [1408/4096 (34%)] | Loss: 4.256180 | Image2Text Acc: 5.47 | Text2Image Acc: 3.12 | Data Time: 0.041s | Batch Time: 0.338s | LR: 0.000055 | logit_scale: 4.605 | Global Batch Size: 128
2024-10-10,14:20:39 | INFO | Rank 0 | Global Steps: 12/3200 | Train Epoch: 1 [1536/4096 (38%)] | Loss: 4.268936 | Image2Text Acc: 5.47 | Text2Image Acc: 6.25 | Data Time: 0.036s | Batch Time: 0.330s | LR: 0.000060 | logit_scale: 4.605 | Global Batch Size: 128
2024-10-10,14:20:44 | INFO | Rank 0 | Global Steps: 13/3200 | Train Epoch: 1 [1664/4096 (41%)] | Loss: 4.263233 | Image2Text Acc: 6.25 | Text2Image Acc: 6.25 | Data Time: 3.959s | Batch Time: 4.263s | LR: 0.000065 | logit_scale: 4.605 | Global Batch Size: 128
2024-10-10,14:20:46 | INFO | Rank 0 | Global Steps: 14/3200 | Train Epoch: 1 [1792/4096 (44%)] | Loss: 4.249680 | Image2Text Acc: 7.03 | Text2Image Acc: 4.69 | Data Time: 2.528s | Batch Time: 2.829s | LR: 0.000070 | logit_scale: 4.605 | Global Batch Size: 128
2024-10-10,14:20:47 | INFO | Rank 0 | Global Steps: 15/3200 | Train Epoch: 1 [1920/4096 (47%)] | Loss: 4.208305 | Image2Text Acc: 6.25 | Text2Image Acc: 5.47 | Data Time: 0.040s | Batch Time: 0.339s | LR: 0.000075 | logit_scale: 4.605 | Global Batch Size: 128
2024-10-10,14:20:47 | INFO | Rank 0 | Global Steps: 16/3200 | Train Epoch: 1 [2048/4096 (50%)] | Loss: 4.351048 | Image2Text Acc: 5.47 | Text2Image Acc: 6.25 | Data Time: 0.038s | Batch Time: 0.333s | LR: 0.000080 | logit_scale: 4.605 | Global Batch Size: 128
2024-10-10,14:20:50 | INFO | Rank 0 | Global Steps: 17/3200 | Train Epoch: 1 [2176/4096 (53%)] | Loss: 4.289299 | Image2Text Acc: 3.91 | Text2Image Acc: 4.69 | Data Time: 2.945s | Batch Time: 3.242s | LR: 0.000085 | logit_scale: 4.605 | Global Batch Size: 128
2024-10-10,14:20:53 | INFO | Rank 0 | Global Steps: 18/3200 | Train Epoch: 1 [2304/4096 (56%)] | Loss: 4.244534 | Image2Text Acc: 3.91 | Text2Image Acc: 5.47 | Data Time: 2.589s | Batch Time: 2.889s | LR: 0.000090 | logit_scale: 4.605 | Global Batch Size: 128
2024-10-10,14:20:54 | INFO | Rank 0 | Global Steps: 19/3200 | Train Epoch: 1 [2432/4096 (59%)] | Loss: 4.298996 | Image2Text Acc: 5.47 | Text2Image Acc: 4.69 | Data Time: 0.036s | Batch Time: 0.334s | LR: 0.000095 | logit_scale: 4.605 | Global Batch Size: 128
2024-10-10,14:20:54 | INFO | Rank 0 | Global Steps: 20/3200 | Train Epoch: 1 [2560/4096 (62%)] | Loss: 4.175068 | Image2Text Acc: 7.03 | Text2Image Acc: 2.34 | Data Time: 0.038s | Batch Time: 0.332s | LR: 0.000100 | logit_scale: 4.605 | Global Batch Size: 128
2024-10-10,14:20:57 | INFO | Rank 0 | Global Steps: 21/3200 | Train Epoch: 1 [2688/4096 (66%)] | Loss: 4.202049 | Image2Text Acc: 6.25 | Text2Image Acc: 5.47 | Data Time: 2.378s | Batch Time: 2.680s | LR: 0.000105 | logit_scale: 4.605 | Global Batch Size: 128
2024-10-10,14:21:00 | INFO | Rank 0 | Global Steps: 22/3200 | Train Epoch: 1 [2816/4096 (69%)] | Loss: 4.255169 | Image2Text Acc: 5.47 | Text2Image Acc: 6.25 | Data Time: 3.118s | Batch Time: 3.419s | LR: 0.000110 | logit_scale: 4.605 | Global Batch Size: 128
2024-10-10,14:21:00 | INFO | Rank 0 | Global Steps: 23/3200 | Train Epoch: 1 [2944/4096 (72%)] | Loss: 4.340736 | Image2Text Acc: 6.25 | Text2Image Acc: 6.25 | Data Time: 0.044s | Batch Time: 0.343s | LR: 0.000115 | logit_scale: 4.605 | Global Batch Size: 128
2024-10-10,14:21:01 | INFO | Rank 0 | Global Steps: 24/3200 | Train Epoch: 1 [3072/4096 (75%)] | Loss: 4.433716 | Image2Text Acc: 1.56 | Text2Image Acc: 6.25 | Data Time: 0.041s | Batch Time: 0.340s | LR: 0.000120 | logit_scale: 4.605 | Global Batch Size: 128
2024-10-10,14:21:03 | INFO | Rank 0 | Global Steps: 25/3200 | Train Epoch: 1 [3200/4096 (78%)] | Loss: 4.339813 | Image2Text Acc: 3.91 | Text2Image Acc: 6.25 | Data Time: 1.788s | Batch Time: 2.085s | LR: 0.000125 | logit_scale: 4.605 | Global Batch Size: 128
2024-10-10,14:21:06 | INFO | Rank 0 | Global Steps: 26/3200 | Train Epoch: 1 [3328/4096 (81%)] | Loss: 4.351143 | Image2Text Acc: 2.34 | Text2Image Acc: 3.12 | Data Time: 2.790s | Batch Time: 3.092s | LR: 0.000130 | logit_scale: 4.605 | Global Batch Size: 128
2024-10-10,14:21:06 | INFO | Rank 0 | Global Steps: 27/3200 | Train Epoch: 1 [3456/4096 (84%)] | Loss: 4.369926 | Image2Text Acc: 2.34 | Text2Image Acc: 4.69 | Data Time: 0.043s | Batch Time: 0.338s | LR: 0.000135 | logit_scale: 4.605 | Global Batch Size: 128
2024-10-10,14:21:07 | INFO | Rank 0 | Global Steps: 28/3200 | Train Epoch: 1 [3584/4096 (88%)] | Loss: 4.199516 | Image2Text Acc: 3.12 | Text2Image Acc: 3.12 | Data Time: 0.037s | Batch Time: 0.335s | LR: 0.000140 | logit_scale: 4.605 | Global Batch Size: 128
2024-10-10,14:21:10 | INFO | Rank 0 | Global Steps: 29/3200 | Train Epoch: 1 [3712/4096 (91%)] | Loss: 4.327763 | Image2Text Acc: 3.91 | Text2Image Acc: 5.47 | Data Time: 3.056s | Batch Time: 3.354s | LR: 0.000145 | logit_scale: 4.605 | Global Batch Size: 128
2024-10-10,14:21:12 | INFO | Rank 0 | Global Steps: 30/3200 | Train Epoch: 1 [3840/4096 (94%)] | Loss: 4.432281 | Image2Text Acc: 2.34 | Text2Image Acc: 4.69 | Data Time: 1.928s | Batch Time: 2.226s | LR: 0.000150 | logit_scale: 4.605 | Global Batch Size: 128
2024-10-10,14:21:12 | INFO | Rank 0 | Global Steps: 31/3200 | Train Epoch: 1 [3968/4096 (97%)] | Loss: 4.358601 | Image2Text Acc: 6.25 | Text2Image Acc: 5.47 | Data Time: 0.037s | Batch Time: 0.332s | LR: 0.000155 | logit_scale: 4.604 | Global Batch Size: 128
2024-10-10,14:21:13 | INFO | Rank 0 | Global Steps: 32/3200 | Train Epoch: 1 [4096/4096 (100%)] | Loss: 4.322407 | Image2Text Acc: 4.69 | Text2Image Acc: 5.47 | Data Time: 0.037s | Batch Time: 0.332s | LR: 0.000160 | logit_scale: 4.604 | Global Batch Size: 128
2024-10-10,14:21:13 | INFO | Rank 0 | Begin to eval on validation set (epoch 1 @ 32 steps)...
2024-10-10,14:21:34 | INFO | Rank 0 | Validation Result (epoch 1 @ 32 steps) | Valid Loss: 4.217743 | Image2Text Acc: 3.91 | Text2Image Acc: 5.86 | logit_scale: 4.604 | Valid Batch Size: 128
2024-10-10,14:21:34 | INFO | Rank 0 | train LMDB file contains 4000 images and 4000 pairs.
2024-10-10,14:21:34 | INFO | Rank 0 | val LMDB file contains 500 images and 500 pairs.
2024-10-10,14:21:51 | INFO | Rank 0 | Saved checkpoint /workspace/code/experiments/demo_data_finetune_vit-b-16_roberta-base_bs128_8gpu/checkpoints/epoch1.pt (epoch 1 @ 32 steps) (writing took 17.00214672088623 seconds)
2024-10-10,14:22:08 | INFO | Rank 0 | Saved checkpoint /workspace/code/experiments/demo_data_finetune_vit-b-16_roberta-base_bs128_8gpu/checkpoints/epoch_latest.pt (epoch 1 @ 32 steps) (writing took 16.851421356201172 seconds)
2024-10-10,14:22:16 | INFO | Rank 0 | Global Steps: 33/3200 | Train Epoch: 2 [128/4096 (3%)] | Loss: 4.297310 | Image2Text Acc: 3.91 | Text2Image Acc: 5.47 | Data Time: 6.647s | Batch Time: 6.951s | LR: 0.000165 | logit_scale: 4.604 | Global Batch Size: 128
2024-10-10,14:22:16 | INFO | Rank 0 | Global Steps: 34/3200 | Train Epoch: 2 [256/4096 (6%)] | Loss: 4.284641 | Image2Text Acc: 4.69 | Text2Image Acc: 1.56 | Data Time: 0.042s | Batch Time: 0.339s | LR: 0.000170 | logit_scale: 4.604 | Global Batch Size: 128
2024-10-10,14:22:17 | INFO | Rank 0 | Global Steps: 35/3200 | Train Epoch: 2 [384/4096 (9%)] | Loss: 4.414612 | Image2Text Acc: 3.12 | Text2Image Acc: 4.69 | Data Time: 0.604s | Batch Time: 0.902s | LR: 0.000175 | logit_scale: 4.604 | Global Batch Size: 128
2024-10-10,14:22:17 | INFO | Rank 0 | Global Steps: 36/3200 | Train Epoch: 2 [512/4096 (12%)] | Loss: 4.766368 | Image2Text Acc: 3.12 | Text2Image Acc: 2.34 | Data Time: 0.036s | Batch Time: 0.331s | LR: 0.000180 | logit_scale: 4.604 | Global Batch Size: 128
2024-10-10,14:22:22 | INFO | Rank 0 | Global Steps: 37/3200 | Train Epoch: 2 [640/4096 (16%)] | Loss: 4.304634 | Image2Text Acc: 3.12 | Text2Image Acc: 2.34 | Data Time: 4.513s | Batch Time: 4.818s | LR: 0.000185 | logit_scale: 4.604 | Global Batch Size: 128
2024-10-10,14:22:22 | INFO | Rank 0 | Global Steps: 38/3200 | Train Epoch: 2 [768/4096 (19%)] | Loss: 4.475136 | Image2Text Acc: 3.12 | Text2Image Acc: 3.91 | Data Time: 0.038s | Batch Time: 0.334s | LR: 0.000190 | logit_scale: 4.604 | Global Batch Size: 128
2024-10-10,14:22:24 | INFO | Rank 0 | Global Steps: 39/3200 | Train Epoch: 2 [896/4096 (22%)] | Loss: 4.387520 | Image2Text Acc: 3.91 | Text2Image Acc: 3.91 | Data Time: 1.873s | Batch Time: 2.171s | LR: 0.000195 | logit_scale: 4.604 | Global Batch Size: 128
2024-10-10,14:22:25 | INFO | Rank 0 | Global Steps: 40/3200 | Train Epoch: 2 [1024/4096 (25%)] | Loss: 4.462852 | Image2Text Acc: 3.12 | Text2Image Acc: 3.91 | Data Time: 0.040s | Batch Time: 0.335s | LR: 0.000200 | logit_scale: 4.604 | Global Batch Size: 128
2024-10-10,14:22:29 | INFO | Rank 0 | Global Steps: 41/3200 | Train Epoch: 2 [1152/4096 (28%)] | Loss: 4.601013 | Image2Text Acc: 2.34 | Text2Image Acc: 1.56 | Data Time: 3.525s | Batch Time: 3.828s | LR: 0.000205 | logit_scale: 4.604 | Global Batch Size: 128
2024-10-10,14:22:29 | INFO | Rank 0 | Global Steps: 42/3200 | Train Epoch: 2 [1280/4096 (31%)] | Loss: 4.392643 | Image2Text Acc: 3.12 | Text2Image Acc: 3.91 | Data Time: 0.040s | Batch Time: 0.338s | LR: 0.000210 | logit_scale: 4.604 | Global Batch Size: 128
2024-10-10,14:22:30 | INFO | Rank 0 | Global Steps: 43/3200 | Train Epoch: 2 [1408/4096 (34%)] | Loss: 4.625900 | Image2Text Acc: 0.78 | Text2Image Acc: 3.91 | Data Time: 1.126s | Batch Time: 1.420s | LR: 0.000215 | logit_scale: 4.604 | Global Batch Size: 128
2024-10-10,14:22:31 | INFO | Rank 0 | Global Steps: 44/3200 | Train Epoch: 2 [1536/4096 (38%)] | Loss: 4.499672 | Image2Text Acc: 0.00 | Text2Image Acc: 2.34 | Data Time: 0.036s | Batch Time: 0.331s | LR: 0.000220 | logit_scale: 4.604 | Global Batch Size: 128
2024-10-10,14:22:35 | INFO | Rank 0 | Global Steps: 45/3200 | Train Epoch: 2 [1664/4096 (41%)] | Loss: 4.556229 | Image2Text Acc: 3.12 | Text2Image Acc: 4.69 | Data Time: 3.978s | Batch Time: 4.282s | LR: 0.000225 | logit_scale: 4.604 | Global Batch Size: 128
2024-10-10,14:22:35 | INFO | Rank 0 | Global Steps: 46/3200 | Train Epoch: 2 [1792/4096 (44%)] | Loss: 4.524231 | Image2Text Acc: 3.91 | Text2Image Acc: 3.12 | Data Time: 0.035s | Batch Time: 0.335s | LR: 0.000230 | logit_scale: 4.604 | Global Batch Size: 128
2024-10-10,14:22:37 | INFO | Rank 0 | Global Steps: 47/3200 | Train Epoch: 2 [1920/4096 (47%)] | Loss: 4.618797 | Image2Text Acc: 0.78 | Text2Image Acc: 3.12 | Data Time: 1.077s | Batch Time: 1.372s | LR: 0.000235 | logit_scale: 4.604 | Global Batch Size: 128
2024-10-10,14:22:37 | INFO | Rank 0 | Global Steps: 48/3200 | Train Epoch: 2 [2048/4096 (50%)] | Loss: 4.657127 | Image2Text Acc: 1.56 | Text2Image Acc: 2.34 | Data Time: 0.036s | Batch Time: 0.335s | LR: 0.000240 | logit_scale: 4.604 | Global Batch Size: 128
2024-10-10,14:22:41 | INFO | Rank 0 | Global Steps: 49/3200 | Train Epoch: 2 [2176/4096 (53%)] | Loss: 4.595833 | Image2Text Acc: 2.34 | Text2Image Acc: 1.56 | Data Time: 3.846s | Batch Time: 4.150s | LR: 0.000245 | logit_scale: 4.604 | Global Batch Size: 128
2024-10-10,14:22:42 | INFO | Rank 0 | Global Steps: 50/3200 | Train Epoch: 2 [2304/4096 (56%)] | Loss: 4.590302 | Image2Text Acc: 2.34 | Text2Image Acc: 2.34 | Data Time: 0.038s | Batch Time: 0.337s | LR: 0.000250 | logit_scale: 4.603 | Global Batch Size: 128
2024-10-10,14:22:43 | INFO | Rank 0 | Global Steps: 51/3200 | Train Epoch: 2 [2432/4096 (59%)] | Loss: 4.545975 | Image2Text Acc: 3.91 | Text2Image Acc: 1.56 | Data Time: 1.587s | Batch Time: 1.887s | LR: 0.000255 | logit_scale: 4.603 | Global Batch Size: 128
2024-10-10,14:22:44 | INFO | Rank 0 | Global Steps: 52/3200 | Train Epoch: 2 [2560/4096 (62%)] | Loss: 4.431618 | Image2Text Acc: 3.12 | Text2Image Acc: 3.91 | Data Time: 0.040s | Batch Time: 0.340s | LR: 0.000260 | logit_scale: 4.603 | Global Batch Size: 128
2024-10-10,14:22:47 | INFO | Rank 0 | Global Steps: 53/3200 | Train Epoch: 2 [2688/4096 (66%)] | Loss: 4.410482 | Image2Text Acc: 3.91 | Text2Image Acc: 3.12 | Data Time: 3.403s | Batch Time: 3.703s | LR: 0.000265 | logit_scale: 4.603 | Global Batch Size: 128
2024-10-10,14:22:48 | INFO | Rank 0 | Global Steps: 54/3200 | Train Epoch: 2 [2816/4096 (69%)] | Loss: 4.380985 | Image2Text Acc: 2.34 | Text2Image Acc: 5.47 | Data Time: 0.042s | Batch Time: 0.340s | LR: 0.000270 | logit_scale: 4.603 | Global Batch Size: 128
2024-10-10,14:22:50 | INFO | Rank 0 | Global Steps: 55/3200 | Train Epoch: 2 [2944/4096 (72%)] | Loss: 4.415821 | Image2Text Acc: 3.91 | Text2Image Acc: 3.12 | Data Time: 1.674s | Batch Time: 1.969s | LR: 0.000275 | logit_scale: 4.603 | Global Batch Size: 128
2024-10-10,14:22:50 | INFO | Rank 0 | Global Steps: 56/3200 | Train Epoch: 2 [3072/4096 (75%)] | Loss: 4.612484 | Image2Text Acc: 2.34 | Text2Image Acc: 3.12 | Data Time: 0.038s | Batch Time: 0.334s | LR: 0.000280 | logit_scale: 4.603 | Global Batch Size: 128
2024-10-10,14:22:54 | INFO | Rank 0 | Global Steps: 57/3200 | Train Epoch: 2 [3200/4096 (78%)] | Loss: 5.101124 | Image2Text Acc: 0.78 | Text2Image Acc: 3.91 | Data Time: 3.150s | Batch Time: 3.456s | LR: 0.000285 | logit_scale: 4.603 | Global Batch Size: 128
2024-10-10,14:22:54 | INFO | Rank 0 | Global Steps: 58/3200 | Train Epoch: 2 [3328/4096 (81%)] | Loss: 5.060188 | Image2Text Acc: 1.56 | Text2Image Acc: 0.78 | Data Time: 0.039s | Batch Time: 0.335s | LR: 0.000290 | logit_scale: 4.603 | Global Batch Size: 128
2024-10-10,14:22:56 | INFO | Rank 0 | Global Steps: 59/3200 | Train Epoch: 2 [3456/4096 (84%)] | Loss: 4.785370 | Image2Text Acc: 1.56 | Text2Image Acc: 3.91 | Data Time: 1.727s | Batch Time: 2.025s | LR: 0.000295 | logit_scale: 4.603 | Global Batch Size: 128
2024-10-10,14:22:56 | INFO | Rank 0 | Global Steps: 60/3200 | Train Epoch: 2 [3584/4096 (88%)] | Loss: 4.811279 | Image2Text Acc: 1.56 | Text2Image Acc: 2.34 | Data Time: 0.036s | Batch Time: 0.337s | LR: 0.000300 | logit_scale: 4.603 | Global Batch Size: 128
2024-10-10,14:23:00 | INFO | Rank 0 | Global Steps: 61/3200 | Train Epoch: 2 [3712/4096 (91%)] | Loss: 4.825073 | Image2Text Acc: 1.56 | Text2Image Acc: 0.78 | Data Time: 3.007s | Batch Time: 3.313s | LR: 0.000305 | logit_scale: 4.602 | Global Batch Size: 128
2024-10-10,14:23:00 | INFO | Rank 0 | Global Steps: 62/3200 | Train Epoch: 2 [3840/4096 (94%)] | Loss: 4.803360 | Image2Text Acc: 0.00 | Text2Image Acc: 0.78 | Data Time: 0.039s | Batch Time: 0.334s | LR: 0.000310 | logit_scale: 4.602 | Global Batch Size: 128
2024-10-10,14:23:01 | INFO | Rank 0 | Global Steps: 63/3200 | Train Epoch: 2 [3968/4096 (97%)] | Loss: 4.794815 | Image2Text Acc: 2.34 | Text2Image Acc: 1.56 | Data Time: 1.087s | Batch Time: 1.382s | LR: 0.000315 | logit_scale: 4.602 | Global Batch Size: 128
2024-10-10,14:23:02 | INFO | Rank 0 | Global Steps: 64/3200 | Train Epoch: 2 [4096/4096 (100%)] | Loss: 4.777557 | Image2Text Acc: 0.00 | Text2Image Acc: 2.34 | Data Time: 0.036s | Batch Time: 0.332s | LR: 0.000320 | logit_scale: 4.602 | Global Batch Size: 128
2024-10-10,14:23:02 | INFO | Rank 0 | Begin to eval on validation set (epoch 2 @ 64 steps)...
2024-10-10,14:23:18 | INFO | Rank 0 | Validation Result (epoch 2 @ 64 steps) | Valid Loss: 4.740685 | Image2Text Acc: 1.37 | Text2Image Acc: 1.37 | logit_scale: 4.602 | Valid Batch Size: 128
2024-10-10,14:23:19 | INFO | Rank 0 | train LMDB file contains 4000 images and 4000 pairs.
2024-10-10,14:23:19 | INFO | Rank 0 | val LMDB file contains 500 images and 500 pairs.
The text was updated successfully, but these errors were encountered: