Fix dump of bfloat16 torch tensor

[lhoestq]

close #7000

[HuggingFaceDocBuilderDev]

The docs for this PR live here. All of your documentation changes will be reflected on that endpoint. The docs are available until 30 days after the last update.

[github-actions]

Show benchmarks

PyArrow==8.0.0

Show updated benchmarks!

Benchmark: benchmark_array_xd.json

metric	read_batch_formatted_as_numpy after write_array2d	read_batch_formatted_as_numpy after write_flattened_sequence	read_batch_formatted_as_numpy after write_nested_sequence	read_batch_unformated after write_array2d	read_batch_unformated after write_flattened_sequence	read_batch_unformated after write_nested_sequence	read_col_formatted_as_numpy after write_array2d	read_col_formatted_as_numpy after write_flattened_sequence	read_col_formatted_as_numpy after write_nested_sequence	read_col_unformated after write_array2d	read_col_unformated after write_flattened_sequence	read_col_unformated after write_nested_sequence	read_formatted_as_numpy after write_array2d	read_formatted_as_numpy after write_flattened_sequence	read_formatted_as_numpy after write_nested_sequence	read_unformated after write_array2d	read_unformated after write_flattened_sequence	read_unformated after write_nested_sequence	write_array2d	write_flattened_sequence	write_nested_sequence
new / old (diff)	0.005321 / 0.011353 (-0.006032)	0.003495 / 0.011008 (-0.007514)	0.065577 / 0.038508 (0.027069)	0.030876 / 0.023109 (0.007767)	0.255216 / 0.275898 (-0.020682)	0.265111 / 0.323480 (-0.058368)	0.003149 / 0.007986 (-0.004837)	0.004062 / 0.004328 (-0.000267)	0.051142 / 0.004250 (0.046891)	0.042460 / 0.037052 (0.005408)	0.270692 / 0.258489 (0.012203)	0.284957 / 0.293841 (-0.008884)	0.030143 / 0.128546 (-0.098403)	0.012148 / 0.075646 (-0.063498)	0.203706 / 0.419271 (-0.215565)	0.035948 / 0.043533 (-0.007584)	0.251391 / 0.255139 (-0.003748)	0.270908 / 0.283200 (-0.012292)	0.018496 / 0.141683 (-0.123187)	1.118567 / 1.452155 (-0.333587)	1.157695 / 1.492716 (-0.335021)

Benchmark: benchmark_getitem_100B.json

metric	get_batch_of_1024_random_rows	get_batch_of_1024_rows	get_first_row	get_last_row
new / old (diff)	0.135649 / 0.018006 (0.117643)	0.281489 / 0.000490 (0.281000)	0.000244 / 0.000200 (0.000044)	0.000042 / 0.000054 (-0.000012)

Benchmark: benchmark_indices_mapping.json

metric	select	shard	shuffle	sort	train_test_split
new / old (diff)	0.018700 / 0.037411 (-0.018711)	0.062305 / 0.014526 (0.047779)	0.074968 / 0.176557 (-0.101589)	0.121490 / 0.737135 (-0.615645)	0.075585 / 0.296338 (-0.220754)

Benchmark: benchmark_iterating.json

metric	read 5000	read 50000	read_batch 50000 10	read_batch 50000 100	read_batch 50000 1000	read_formatted numpy 5000	read_formatted pandas 5000	read_formatted tensorflow 5000	read_formatted torch 5000	read_formatted_batch numpy 5000 10	read_formatted_batch numpy 5000 1000	shuffled read 5000	shuffled read 50000	shuffled read_batch 50000 10	shuffled read_batch 50000 100	shuffled read_batch 50000 1000	shuffled read_formatted numpy 5000	shuffled read_formatted_batch numpy 5000 10	shuffled read_formatted_batch numpy 5000 1000
new / old (diff)	0.276929 / 0.215209 (0.061720)	2.733543 / 2.077655 (0.655888)	1.414585 / 1.504120 (-0.089535)	1.301975 / 1.541195 (-0.239220)	1.336698 / 1.468490 (-0.131792)	0.720650 / 4.584777 (-3.864127)	2.374796 / 3.745712 (-1.370917)	2.866534 / 5.269862 (-2.403327)	1.819607 / 4.565676 (-2.746069)	0.077914 / 0.424275 (-0.346361)	0.005146 / 0.007607 (-0.002461)	0.331722 / 0.226044 (0.105678)	3.290875 / 2.268929 (1.021946)	1.799806 / 55.444624 (-53.644818)	1.476816 / 6.876477 (-5.399660)	1.511441 / 2.142072 (-0.630631)	0.798043 / 4.805227 (-4.007185)	0.134577 / 6.500664 (-6.366087)	0.042055 / 0.075469 (-0.033415)

Benchmark: benchmark_map_filter.json

metric	filter	map fast-tokenizer batched	map identity	map identity batched	map no-op batched	map no-op batched numpy	map no-op batched pandas	map no-op batched pytorch	map no-op batched tensorflow
new / old (diff)	0.967908 / 1.841788 (-0.873880)	11.215688 / 8.074308 (3.141380)	9.486403 / 10.191392 (-0.704989)	0.141864 / 0.680424 (-0.538560)	0.013462 / 0.534201 (-0.520739)	0.302601 / 0.579283 (-0.276682)	0.266870 / 0.434364 (-0.167494)	0.336963 / 0.540337 (-0.203375)	0.425374 / 1.386936 (-0.961562)

PyArrow==latest

Show updated benchmarks!

Benchmark: benchmark_array_xd.json

metric	read_batch_formatted_as_numpy after write_array2d	read_batch_formatted_as_numpy after write_flattened_sequence	read_batch_formatted_as_numpy after write_nested_sequence	read_batch_unformated after write_array2d	read_batch_unformated after write_flattened_sequence	read_batch_unformated after write_nested_sequence	read_col_formatted_as_numpy after write_array2d	read_col_formatted_as_numpy after write_flattened_sequence	read_col_formatted_as_numpy after write_nested_sequence	read_col_unformated after write_array2d	read_col_unformated after write_flattened_sequence	read_col_unformated after write_nested_sequence	read_formatted_as_numpy after write_array2d	read_formatted_as_numpy after write_flattened_sequence	read_formatted_as_numpy after write_nested_sequence	read_unformated after write_array2d	read_unformated after write_flattened_sequence	read_unformated after write_nested_sequence	write_array2d	write_flattened_sequence	write_nested_sequence
new / old (diff)	0.005549 / 0.011353 (-0.005803)	0.003464 / 0.011008 (-0.007544)	0.051421 / 0.038508 (0.012913)	0.032320 / 0.023109 (0.009211)	0.269591 / 0.275898 (-0.006307)	0.292015 / 0.323480 (-0.031465)	0.004351 / 0.007986 (-0.003634)	0.002772 / 0.004328 (-0.001556)	0.048836 / 0.004250 (0.044586)	0.039501 / 0.037052 (0.002449)	0.282419 / 0.258489 (0.023930)	0.312289 / 0.293841 (0.018448)	0.031788 / 0.128546 (-0.096759)	0.012074 / 0.075646 (-0.063572)	0.060457 / 0.419271 (-0.358814)	0.033106 / 0.043533 (-0.010427)	0.270323 / 0.255139 (0.015184)	0.287855 / 0.283200 (0.004655)	0.017865 / 0.141683 (-0.123818)	1.130406 / 1.452155 (-0.321749)	1.178679 / 1.492716 (-0.314038)

Benchmark: benchmark_getitem_100B.json

metric	get_batch_of_1024_random_rows	get_batch_of_1024_rows	get_first_row	get_last_row
new / old (diff)	0.093606 / 0.018006 (0.075600)	0.297328 / 0.000490 (0.296838)	0.000211 / 0.000200 (0.000011)	0.000043 / 0.000054 (-0.000011)

Benchmark: benchmark_indices_mapping.json

metric	select	shard	shuffle	sort	train_test_split
new / old (diff)	0.022498 / 0.037411 (-0.014913)	0.076927 / 0.014526 (0.062401)	0.088013 / 0.176557 (-0.088544)	0.127279 / 0.737135 (-0.609857)	0.089424 / 0.296338 (-0.206914)

Benchmark: benchmark_iterating.json

metric	read 5000	read 50000	read_batch 50000 10	read_batch 50000 100	read_batch 50000 1000	read_formatted numpy 5000	read_formatted pandas 5000	read_formatted tensorflow 5000	read_formatted torch 5000	read_formatted_batch numpy 5000 10	read_formatted_batch numpy 5000 1000	shuffled read 5000	shuffled read 50000	shuffled read_batch 50000 10	shuffled read_batch 50000 100	shuffled read_batch 50000 1000	shuffled read_formatted numpy 5000	shuffled read_formatted_batch numpy 5000 10	shuffled read_formatted_batch numpy 5000 1000
new / old (diff)	0.296441 / 0.215209 (0.081232)	2.913051 / 2.077655 (0.835396)	1.581816 / 1.504120 (0.077696)	1.451575 / 1.541195 (-0.089620)	1.458968 / 1.468490 (-0.009522)	0.727191 / 4.584777 (-3.857586)	0.954607 / 3.745712 (-2.791106)	2.824357 / 5.269862 (-2.445505)	1.886779 / 4.565676 (-2.678898)	0.079397 / 0.424275 (-0.344878)	0.005566 / 0.007607 (-0.002041)	0.351655 / 0.226044 (0.125611)	3.395790 / 2.268929 (1.126862)	1.886238 / 55.444624 (-53.558387)	1.615413 / 6.876477 (-5.261064)	1.723922 / 2.142072 (-0.418150)	0.807858 / 4.805227 (-3.997369)	0.132998 / 6.500664 (-6.367667)	0.040396 / 0.075469 (-0.035073)

Benchmark: benchmark_map_filter.json

metric	filter	map fast-tokenizer batched	map identity	map identity batched	map no-op batched	map no-op batched numpy	map no-op batched pandas	map no-op batched pytorch	map no-op batched tensorflow
new / old (diff)	1.008527 / 1.841788 (-0.833261)	11.736104 / 8.074308 (3.661796)	10.283367 / 10.191392 (0.091975)	0.141386 / 0.680424 (-0.539038)	0.015722 / 0.534201 (-0.518479)	0.301785 / 0.579283 (-0.277498)	0.123073 / 0.434364 (-0.311291)	0.340478 / 0.540337 (-0.199859)	0.462936 / 1.386936 (-0.924000)

[stoical07]

For future compatibility? 😄

Suggested change

	def create_torchTensor(np_array, dtype=None):
	tensor = torch.from_numpy(np_array)
	if dtype:
	tensor = tensor.type(torch.bfloat16)
	return tensor
	def create_torchTensor(np_array, dtype=None):
	tensor = torch.from_numpy(np_array)
	if dtype:
	tensor = tensor.type(dtype)
	return tensor

[lhoestq]

ah good catch, I went too fast x)

[lhoestq]

fixed at #7003