Better multi-gpu example

[lhoestq]

Use Qwen1.5-0.5B-Chat as an easy example for multi-GPU

the previous example was using a model for translation and the way it was setup was not really the right way to use the model.

[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.

[albertvillanova]

Thanks, better now.

[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)	0.005598 / 0.011353 (-0.005755)	0.003640 / 0.011008 (-0.007369)	0.064557 / 0.038508 (0.026049)	0.029645 / 0.023109 (0.006536)	0.243695 / 0.275898 (-0.032203)	0.261252 / 0.323480 (-0.062228)	0.004067 / 0.007986 (-0.003919)	0.002883 / 0.004328 (-0.001446)	0.049192 / 0.004250 (0.044942)	0.045299 / 0.037052 (0.008246)	0.273207 / 0.258489 (0.014718)	0.288668 / 0.293841 (-0.005173)	0.028114 / 0.128546 (-0.100432)	0.010597 / 0.075646 (-0.065049)	0.215345 / 0.419271 (-0.203927)	0.036119 / 0.043533 (-0.007414)	0.243718 / 0.255139 (-0.011421)	0.266657 / 0.283200 (-0.016543)	0.018176 / 0.141683 (-0.123507)	1.127926 / 1.452155 (-0.324229)	1.168066 / 1.492716 (-0.324650)

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)	0.096001 / 0.018006 (0.077994)	0.304317 / 0.000490 (0.303828)	0.000209 / 0.000200 (0.000009)	0.000051 / 0.000054 (-0.000004)

Benchmark: benchmark_indices_mapping.json

metric	select	shard	shuffle	sort	train_test_split
new / old (diff)	0.018241 / 0.037411 (-0.019170)	0.061505 / 0.014526 (0.046979)	0.072456 / 0.176557 (-0.104101)	0.118315 / 0.737135 (-0.618821)	0.075154 / 0.296338 (-0.221184)

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)	0.278748 / 0.215209 (0.063538)	2.729923 / 2.077655 (0.652268)	1.416835 / 1.504120 (-0.087285)	1.294016 / 1.541195 (-0.247179)	1.323249 / 1.468490 (-0.145241)	0.575389 / 4.584777 (-4.009388)	2.404923 / 3.745712 (-1.340789)	2.769233 / 5.269862 (-2.500629)	1.742340 / 4.565676 (-2.823336)	0.062664 / 0.424275 (-0.361611)	0.004951 / 0.007607 (-0.002656)	0.335024 / 0.226044 (0.108979)	3.291446 / 2.268929 (1.022518)	1.797095 / 55.444624 (-53.647530)	1.532963 / 6.876477 (-5.343513)	1.529315 / 2.142072 (-0.612758)	0.654922 / 4.805227 (-4.150305)	0.118772 / 6.500664 (-6.381892)	0.042034 / 0.075469 (-0.033435)

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)	0.983646 / 1.841788 (-0.858141)	11.518625 / 8.074308 (3.444317)	9.538781 / 10.191392 (-0.652611)	0.140300 / 0.680424 (-0.540124)	0.013966 / 0.534201 (-0.520235)	0.287071 / 0.579283 (-0.292212)	0.270201 / 0.434364 (-0.164163)	0.323294 / 0.540337 (-0.217044)	0.418130 / 1.386936 (-0.968806)

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)	0.005508 / 0.011353 (-0.005844)	0.003714 / 0.011008 (-0.007294)	0.050031 / 0.038508 (0.011523)	0.031866 / 0.023109 (0.008756)	0.272248 / 0.275898 (-0.003650)	0.295105 / 0.323480 (-0.028375)	0.005179 / 0.007986 (-0.002807)	0.002820 / 0.004328 (-0.001508)	0.048896 / 0.004250 (0.044646)	0.045975 / 0.037052 (0.008922)	0.287662 / 0.258489 (0.029173)	0.321139 / 0.293841 (0.027298)	0.049242 / 0.128546 (-0.079304)	0.010732 / 0.075646 (-0.064914)	0.057943 / 0.419271 (-0.361328)	0.033527 / 0.043533 (-0.010006)	0.271746 / 0.255139 (0.016607)	0.291404 / 0.283200 (0.008204)	0.019351 / 0.141683 (-0.122332)	1.157221 / 1.452155 (-0.294934)	1.215757 / 1.492716 (-0.276959)

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)	0.096950 / 0.018006 (0.078944)	0.312002 / 0.000490 (0.311512)	0.000223 / 0.000200 (0.000023)	0.000055 / 0.000054 (0.000001)

Benchmark: benchmark_indices_mapping.json

metric	select	shard	shuffle	sort	train_test_split
new / old (diff)	0.022288 / 0.037411 (-0.015123)	0.075282 / 0.014526 (0.060756)	0.087445 / 0.176557 (-0.089112)	0.125617 / 0.737135 (-0.611519)	0.088878 / 0.296338 (-0.207460)

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)	0.291961 / 0.215209 (0.076752)	2.881445 / 2.077655 (0.803790)	1.586128 / 1.504120 (0.082008)	1.458636 / 1.541195 (-0.082558)	1.487001 / 1.468490 (0.018511)	0.575466 / 4.584777 (-4.009311)	2.454941 / 3.745712 (-1.290771)	2.878077 / 5.269862 (-2.391785)	1.787215 / 4.565676 (-2.778462)	0.064010 / 0.424275 (-0.360265)	0.005092 / 0.007607 (-0.002516)	0.360500 / 0.226044 (0.134455)	3.465574 / 2.268929 (1.196646)	1.957516 / 55.444624 (-53.487108)	1.666282 / 6.876477 (-5.210195)	1.690070 / 2.142072 (-0.452002)	0.661323 / 4.805227 (-4.143905)	0.117824 / 6.500664 (-6.382840)	0.042286 / 0.075469 (-0.033183)

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)	1.026517 / 1.841788 (-0.815270)	12.083347 / 8.074308 (4.009039)	10.269319 / 10.191392 (0.077927)	0.139253 / 0.680424 (-0.541171)	0.016258 / 0.534201 (-0.517943)	0.290583 / 0.579283 (-0.288700)	0.284338 / 0.434364 (-0.150026)	0.335865 / 0.540337 (-0.204473)	0.416600 / 1.386936 (-0.970336)

[mfidabel]

Thanks, I was needing this example today <3