text
stringlengths
0
1.16k
Registering VQA_lavis stepRegistering VQA_lavis step
Registering EVAL step
Registering RESULT step
Registering EVAL step
Registering RESULT step
ANSWER0=VQA(image=RIGHT,question='How many white capped bottles are in the image?')
ANSWER1=EVAL(expr='{ANSWER0} >= 16')
FINAL_ANSWER=RESULT(var=ANSWER1)
torch.Size([1, 3, 448, 448])
ANSWER0=VQA(image=RIGHT,question='Is the dispenser tall and round?')
FINAL_ANSWER=RESULT(var=ANSWER0)
torch.Size([1, 3, 448, 448])
ANSWER0=VQA(image=RIGHT,question='How many chimpanzees are in the image?')
ANSWER1=EVAL(expr='{ANSWER0} >= 4')
FINAL_ANSWER=RESULT(var=ANSWER1)
ANSWER0=VQA(image=LEFT,question='How many veggies are shown in the image?')
ANSWER1=EVAL(expr='{ANSWER0} == 2')
FINAL_ANSWER=RESULT(var=ANSWER1)
torch.Size([3, 3, 448, 448])
torch.Size([7, 3, 448, 448])
question: ['How many white capped bottles are in the image?'], responses:['100']
question: ['Is the dispenser tall and round?'], responses:['no']
[('100', 0.1277092174007614), ('120', 0.12519936731884676), ('88', 0.12483671971182599), ('80', 0.12474858811112934), ('60', 0.12457749608485191), ('99', 0.1243465850330014), ('90', 0.12430147627057883), ('101', 0.12428055006900451)]
[['100', '120', '88', '80', '60', '99', '90', '101']]
torch.Size([1, 3, 448, 448]) knan debug pixel values shape
[('no', 0.1313955057270409), ('yes', 0.12592208734904367), ('no smoking', 0.12472972590078177), ('gone', 0.12376514658020793), ('man', 0.12367833016285167), ('meow', 0.1235796378467502), ('kia', 0.12347643720898455), ('no clock', 0.12345312922433942)]
[['no', 'yes', 'no smoking', 'gone', 'man', 'meow', 'kia', 'no clock']]
dynamic ViT batch size: 1, images per sample: 1.0, dynamic token length: 326
torch.Size([1, 3, 448, 448]) knan debug pixel values shape
dynamic ViT batch size: 1, images per sample: 1.0, dynamic token length: 326
dynamic ViT batch size: 1, images per sample: 1.0, dynamic token length: 326
dynamic ViT batch size: 1, images per sample: 1.0, dynamic token length: 326
question: ['How many chimpanzees are in the image?'], responses:['1']
dynamic ViT batch size: 1, images per sample: 1.0, dynamic token length: 326
[('1', 0.12829009354978346), ('3', 0.12529928082343206), ('4', 0.12464806219229535), ('8', 0.12460015878893425), ('6', 0.12451220062887247), ('12', 0.124338487048427), ('2', 0.12420459433498025), ('47', 0.12410712263327517)]
[['1', '3', '4', '8', '6', '12', '2', '47']]
dynamic ViT batch size: 1, images per sample: 1.0, dynamic token length: 326
torch.Size([3, 3, 448, 448]) knan debug pixel values shape
dynamic ViT batch size: 1, images per sample: 1.0, dynamic token length: 326
dynamic ViT batch size: 1, images per sample: 1.0, dynamic token length: 326
tensor([9.4209e-01, 1.6975e-02, 1.6071e-04, 4.4430e-03, 1.1240e-02, 2.0205e-03,
1.3516e-02, 9.5507e-03], device='cuda:0', grad_fn=<SoftmaxBackward0>)
100 *************
['100', '120', '88', '80', '60', '99', '90', '101'] tensor([9.4209e-01, 1.6975e-02, 1.6071e-04, 4.4430e-03, 1.1240e-02, 2.0205e-03,
1.3516e-02, 9.5507e-03], device='cuda:0', grad_fn=<SelectBackward0>)
最后的概率分布为: {True: tensor(1., device='cuda:0', grad_fn=<DivBackward0>), False: tensor(0., device='cuda:0', grad_fn=<MulBackward0>), 'Execute Error': tensor(0., device='cuda:0', grad_fn=<DivBackward0>)}
tensor([1.0000e+00, 2.0612e-09, 1.7647e-07, 4.7251e-12, 2.0264e-12, 7.0240e-10,
1.7027e-10, 5.1928e-08], device='cuda:2', grad_fn=<SoftmaxBackward0>)
no *************
['no', 'yes', 'no smoking', 'gone', 'man', 'meow', 'kia', 'no clock'] tensor([1.0000e+00, 2.0612e-09, 1.7647e-07, 4.7251e-12, 2.0264e-12, 7.0240e-10,
1.7027e-10, 5.1928e-08], device='cuda:2', grad_fn=<SelectBackward0>)
ANSWER0=VQA(image=RIGHT,question='Which direction is the animal facing?')
ANSWER1=EVAL(expr='{ANSWER0} == "left"')
FINAL_ANSWER=RESULT(var=ANSWER1)
最后的概率分布为: {True: tensor(2.0612e-09, device='cuda:2', grad_fn=<UnbindBackward0>), False: tensor(1.0000, device='cuda:2', grad_fn=<UnbindBackward0>), 'Execute Error': tensor(1.1921e-07, device='cuda:2', grad_fn=<SubBackward0>)}
ANSWER0=VQA(image=LEFT,question='How many warthogs are in the image?')
ANSWER1=EVAL(expr='{ANSWER0} == 5')
FINAL_ANSWER=RESULT(var=ANSWER1)
torch.Size([13, 3, 448, 448])
torch.Size([13, 3, 448, 448])
question: ['How many veggies are shown in the image?'], responses:['25']
[('25', 0.12829254095279924), ('24', 0.12483820084904945), ('22', 0.12475987746692119), ('27', 0.12456245392996547), ('29', 0.12448525150609151), ('26', 0.12439374413507664), ('20', 0.12436562107615377), ('21', 0.1243023100839428)]
[['25', '24', '22', '27', '29', '26', '20', '21']]
torch.Size([7, 3, 448, 448]) knan debug pixel values shape
tensor([1.0000e+00, 1.3917e-10, 2.2193e-11, 5.4035e-11, 2.1147e-11, 9.7195e-09,
4.5468e-08, 1.4774e-10], device='cuda:1', grad_fn=<SoftmaxBackward0>)
1 *************
['1', '3', '4', '8', '6', '12', '2', '47'] tensor([1.0000e+00, 1.3917e-10, 2.2193e-11, 5.4035e-11, 2.1147e-11, 9.7195e-09,
4.5468e-08, 1.4774e-10], device='cuda:1', grad_fn=<SelectBackward0>)
最后的概率分布为: {True: tensor(9.9646e-09, device='cuda:1', grad_fn=<DivBackward0>), False: tensor(1., device='cuda:1', grad_fn=<DivBackward0>), 'Execute Error': tensor(0., device='cuda:1', grad_fn=<DivBackward0>)}
ANSWER0=VQA(image=RIGHT,question='How many wolves are in the image?')
ANSWER1=EVAL(expr='{ANSWER0} <= 2')
FINAL_ANSWER=RESULT(var=ANSWER1)
torch.Size([7, 3, 448, 448])
question: ['Which direction is the animal facing?'], responses:['right']
question: ['How many warthogs are in the image?'], responses:['1']
[('right', 0.12743553739412528), ('right 1', 0.12490968573275477), ('straight', 0.12485251094891832), ('floating', 0.12468075392646753), ('flip', 0.12467791878738273), ('backwards', 0.12452118816110067), ('serious', 0.12447626064603681), ('working', 0.12444614440321403)]
[['right', 'right 1', 'straight', 'floating', 'flip', 'backwards', 'serious', 'working']]
[('1', 0.12829009354978346), ('3', 0.12529928082343206), ('4', 0.12464806219229535), ('8', 0.12460015878893425), ('6', 0.12451220062887247), ('12', 0.124338487048427), ('2', 0.12420459433498025), ('47', 0.12410712263327517)]
[['1', '3', '4', '8', '6', '12', '2', '47']]
question: ['How many wolves are in the image?'], responses:['三']
[('biking', 0.12639990046765587), ('geese', 0.1262789403477572), ('cushion', 0.1253965842661667), ('bulldog', 0.1252365705078606), ('striped', 0.12499404846420245), ('floral', 0.12444127054742124), ('stove', 0.12381223353082338), ('dodgers', 0.12344045186811266)]
[['biking', 'geese', 'cushion', 'bulldog', 'striped', 'floral', 'stove', 'dodgers']]
torch.Size([7, 3, 448, 448]) knan debug pixel values shape
torch.Size([13, 3, 448, 448]) knan debug pixel values shape
dynamic ViT batch size: 13, images per sample: 13.0, dynamic token length: 3395
torch.Size([13, 3, 448, 448]) knan debug pixel values shape
tensor([0.6140, 0.0828, 0.0961, 0.0443, 0.0074, 0.0274, 0.0842, 0.0438],
device='cuda:3', grad_fn=<SoftmaxBackward0>)
25 *************
['25', '24', '22', '27', '29', '26', '20', '21'] tensor([0.6140, 0.0828, 0.0961, 0.0443, 0.0074, 0.0274, 0.0842, 0.0438],
device='cuda:3', grad_fn=<SelectBackward0>)
dynamic ViT batch size: 13, images per sample: 13.0, dynamic token length: 3397
最后的概率分布为: {True: tensor(0., device='cuda:3', grad_fn=<MulBackward0>), False: tensor(1.0000, device='cuda:3', grad_fn=<DivBackward0>), 'Execute Error': tensor(-1.1921e-07, device='cuda:3', grad_fn=<DivBackward0>)}
ANSWER0=VQA(image=LEFT,question='How many birds are in the image?')
ANSWER1=EVAL(expr='{ANSWER0} == 1')
FINAL_ANSWER=RESULT(var=ANSWER1)
torch.Size([7, 3, 448, 448])
dynamic ViT batch size: 13, images per sample: 13.0, dynamic token length: 3396