Datasets:
TMarcus
/
InternVL

Dataset card Data Studio Files
xet
Community
1
text
stringlengths
0
1.16k
"loss_scale": 0,
"initial_scale_power": 32,
"loss_scale_window": 1000,
"hysteresis": 2,
"min_loss_scale": 1
},
"bf16": {
"enabled": true
},
"optimizer": {
"type": "AdamW",
"params": {
"lr": 4e-05,
"betas": [0.9, 0.999],
"eps": 1e-08,
"weight_decay": 0.01
}
},
"gradient_accumulation_steps": 1,
"gradient_clipping": 1.0,
"steps_per_print": inf,
"train_batch_size": 16,
"train_micro_batch_size_per_gpu": 4,
"wall_clock_breakdown": true
}
[INFO|trainer.py:1721] 2024-10-22 17:18:27,733 >> ***** Running training *****
[INFO|trainer.py:1722] 2024-10-22 17:18:27,733 >> Num examples = 9,681
[INFO|trainer.py:1723] 2024-10-22 17:18:27,733 >> Num Epochs = 4
[INFO|trainer.py:1724] 2024-10-22 17:18:27,733 >> Instantaneous batch size per device = 4
[INFO|trainer.py:1727] 2024-10-22 17:18:27,733 >> Total train batch size (w. parallel, distributed & accumulation) = 16
[INFO|trainer.py:1728] 2024-10-22 17:18:27,733 >> Gradient Accumulation steps = 1
[INFO|trainer.py:1729] 2024-10-22 17:18:27,733 >> Total optimization steps = 2,424
[INFO|trainer.py:1730] 2024-10-22 17:18:27,736 >> Number of trainable parameters = 15,728,640
0%| | 0/2424 [00:00<?, ?it/s][2024-10-22 17:18:29,441] [INFO] [real_accelerator.py:133:get_accelerator] Setting ds_accelerator to cuda (auto detect)
[2024-10-22 17:18:29,449] [INFO] [real_accelerator.py:133:get_accelerator] Setting ds_accelerator to cuda (auto detect)
[2024-10-22 17:18:29,454] [INFO] [real_accelerator.py:133:get_accelerator] Setting ds_accelerator to cuda (auto detect)
[2024-10-22 17:18:29,468] [INFO] [real_accelerator.py:133:get_accelerator] Setting ds_accelerator to cuda (auto detect)
[2024-10-22 17:18:32,501] [INFO] [real_accelerator.py:133:get_accelerator] Setting ds_accelerator to cuda (auto detect)
[2024-10-22 17:18:32,776] [INFO] [real_accelerator.py:133:get_accelerator] Setting ds_accelerator to cuda (auto detect)
[2024-10-22 17:18:32,780] [INFO] [real_accelerator.py:133:get_accelerator] Setting ds_accelerator to cuda (auto detect)
[2024-10-22 17:18:32,796] [INFO] [real_accelerator.py:133:get_accelerator] Setting ds_accelerator to cuda (auto detect)
[2024-10-22 17:18:35,375] [INFO] [real_accelerator.py:133:get_accelerator] Setting ds_accelerator to cuda (auto detect)
[2024-10-22 17:18:35,680] [INFO] [real_accelerator.py:133:get_accelerator] Setting ds_accelerator to cuda (auto detect)
[2024-10-22 17:18:35,757] [INFO] [real_accelerator.py:133:get_accelerator] Setting ds_accelerator to cuda (auto detect)
[2024-10-22 17:18:35,790] [INFO] [real_accelerator.py:133:get_accelerator] Setting ds_accelerator to cuda (auto detect)
[2024-10-22 17:18:38,205] [INFO] [real_accelerator.py:133:get_accelerator] Setting ds_accelerator to cuda (auto detect)
[2024-10-22 17:18:38,495] [INFO] [real_accelerator.py:133:get_accelerator] Setting ds_accelerator to cuda (auto detect)
[2024-10-22 17:18:38,663] [INFO] [real_accelerator.py:133:get_accelerator] Setting ds_accelerator to cuda (auto detect)
[2024-10-22 17:18:38,720] [INFO] [real_accelerator.py:133:get_accelerator] Setting ds_accelerator to cuda (auto detect)
Registering VQA_lavis step
Registering EVAL step
Registering RESULT step
ANSWER0=VQA(image=RIGHT,question='Can you see the customers in the image?')
ANSWER1=EVAL(expr='{ANSWER0}')
FINAL_ANSWER=RESULT(var=ANSWER1)
torch.Size([1, 3, 448, 448])
Registering VQA_lavis step
Registering EVAL step
Registering RESULT step
ANSWER0=VQA(image=RIGHT,question='Is there apparent damage to the bus in the image?')
ANSWER1=EVAL(expr='not {ANSWER0}')
FINAL_ANSWER=RESULT(var=ANSWER1)
torch.Size([3, 3, 448, 448])
Registering VQA_lavis step
Registering EVAL step
Registering RESULT step
ANSWER0=VQA(image=RIGHT,question='How many animal species are in the image?')
ANSWER1=EVAL(expr='{ANSWER0} >= 2')
FINAL_ANSWER=RESULT(var=ANSWER1)
torch.Size([7, 3, 448, 448])
Registering VQA_lavis step
Registering EVAL step
Registering RESULT step
ANSWER0=VQA(image=LEFT,question='How many windows are on the left wall?')
ANSWER1=EVAL(expr='{ANSWER0} == 3')
FINAL_ANSWER=RESULT(var=ANSWER1)
torch.Size([7, 3, 448, 448])
question: ['Can you see the customers in the image?'], responses:['yes']
[WARNING|tokenization_utils_base.py:2697] 2024-10-22 17:18:41,401 >> Asking to truncate to max_length but no maximum length is provided and the model has no predefined maximum length. Default to no truncation.
question: ['Is there apparent damage to the bus in the image?'], responses:['yes']
[WARNING|tokenization_utils_base.py:2697] 2024-10-22 17:18:41,816 >> Asking to truncate to max_length but no maximum length is provided and the model has no predefined maximum length. Default to no truncation.
[('yes', 0.1298617250866936), ('congratulations', 0.12464161604141298), ('no', 0.12445222599225532), ('honey', 0.12437056445881921), ('solid', 0.12422595371654564), ('right', 0.12419889376311324), ('candle', 0.12414264780165109), ('chocolate', 0.12410637313950891)]
[['yes', 'congratulations', 'no', 'honey', 'solid', 'right', 'candle', 'chocolate']]
torch.Size([1, 3, 448, 448]) knan debug pixel values shape
question: ['How many animal species are in the image?'], responses:['1']
[WARNING|tokenization_utils_base.py:2697] 2024-10-22 17:18:42,162 >> Asking to truncate to max_length but no maximum length is provided and the model has no predefined maximum length. Default to no truncation.
[('yes', 0.1298617250866936), ('congratulations', 0.12464161604141298), ('no', 0.12445222599225532), ('honey', 0.12437056445881921), ('solid', 0.12422595371654564), ('right', 0.12419889376311324), ('candle', 0.12414264780165109), ('chocolate', 0.12410637313950891)]
[['yes', 'congratulations', 'no', 'honey', 'solid', 'right', 'candle', 'chocolate']]
tensor([6.9835e-01, 2.3419e-02, 2.7348e-01, 2.6040e-03, 1.8967e-04, 6.1416e-04,
1.0074e-04, 1.2463e-03], device='cuda:2', grad_fn=<SoftmaxBackward0>)
yes *************
['yes', 'congratulations', 'no', 'honey', 'solid', 'right', 'candle', 'chocolate'] tensor([6.9835e-01, 2.3419e-02, 2.7348e-01, 2.6040e-03, 1.8967e-04, 6.1416e-04,
1.0074e-04, 1.2463e-03], device='cuda:2', grad_fn=<SelectBackward0>)
最后的概率分布为: {True: tensor(0.6983, device='cuda:2', grad_fn=<DivBackward0>), False: tensor(0.2735, device='cuda:2', grad_fn=<DivBackward0>), 'Execute Error': tensor(0.0282, device='cuda:2', grad_fn=<DivBackward0>)}
ANSWER0=VQA(image=LEFT,question='Is liquid being poured into a cup?')
ANSWER1=EVAL(expr='{ANSWER0}')
FINAL_ANSWER=RESULT(var=ANSWER1)
torch.Size([3, 3, 448, 448]) knan debug pixel values shape
torch.Size([5, 3, 448, 448])