Skip to content

Commit

Permalink
[Librispeech] Add 'all' config (#4184)
Browse files Browse the repository at this point in the history 
* [Librispeech] Add 'all' config

* Update datasets/librispeech_asr/librispeech_asr.py

* apply suggestions

* correct paths

* up

* up

* up

* up

* up

Co-authored-by: Patrick von Platen <[email protected]>
  • Loading branch information
@patrickvonplaten
patrickvonplaten and Patrick von Platen authored Apr 22, 2022
1 parent 4134e89 commit 91d7171
Show file tree
Hide file tree
Showing 3 changed files with 132 additions and 30 deletions.
2 changes: 1 addition & 1 deletion datasets/librispeech_asr/README.md
View file
Original file line number Diff line number Diff line change
Expand Up @@ -20,7 +20,7 @@ task_categories:
- automatic-speech-recognition
- audio-classification
task_ids:
- audio-speaker-identification
- speaker-identification
---

# Dataset Card for librispeech_asr
Expand Down
2 changes: 1 addition & 1 deletion datasets/librispeech_asr/dataset_infos.json
View file
Original file line number Diff line number Diff line change
@@ -1 +1 @@
{"clean": {"description": "LibriSpeech is a corpus of approximately 1000 hours of read English speech with sampling rate of 16 kHz,\nprepared by Vassil Panayotov with the assistance of Daniel Povey. The data is derived from read\naudiobooks from the LibriVox project, and has been carefully segmented and aligned.87\n\nNote that in order to limit the required storage for preparing this dataset, the audio\nis stored in the .flac format and is not converted to a float32 array. To convert, the audio\nfile to a float32 array, please make use of the `.map()` function as follows:\n\n\n```python\nimport soundfile as sf\n\ndef map_to_array(batch):\n speech_array, _ = sf.read(batch[\"file\"])\n batch[\"speech\"] = speech_array\n return batch\n\ndataset = dataset.map(map_to_array, remove_columns=[\"file\"])\n```\n", "citation": "@inproceedings{panayotov2015librispeech,\n title={Librispeech: an ASR corpus based on public domain audio books},\n author={Panayotov, Vassil and Chen, Guoguo and Povey, Daniel and Khudanpur, Sanjeev},\n booktitle={Acoustics, Speech and Signal Processing (ICASSP), 2015 IEEE International Conference on},\n pages={5206--5210},\n year={2015},\n organization={IEEE}\n}\n", "homepage": "http://www.openslr.org/12", "license": "", "features": {"file": {"dtype": "string", "id": null, "_type": "Value"}, "audio": {"sampling_rate": 16000, "mono": true, "_storage_dtype": "struct", "id": null, "_type": "Audio"}, "text": {"dtype": "string", "id": null, "_type": "Value"}, "speaker_id": {"dtype": "int64", "id": null, "_type": "Value"}, "chapter_id": {"dtype": "int64", "id": null, "_type": "Value"}, "id": {"dtype": "string", "id": null, "_type": "Value"}}, "post_processed": null, "supervised_keys": {"input": "file", "output": "text"}, "task_templates": [{"task": "automatic-speech-recognition", "audio_column": "audio", "transcription_column": "text"}], "builder_name": "librispeech_asr", "config_name": "clean", "version": {"version_str": "2.1.0", "description": "", "major": 2, "minor": 1, "patch": 0}, "splits": {"train.100": {"name": "train.100", "num_bytes": 6619683041, "num_examples": 28539, "dataset_name": "librispeech_asr"}, "train.360": {"name": "train.360", "num_bytes": 23898214592, "num_examples": 104014, "dataset_name": "librispeech_asr"}, "validation": {"name": "validation", "num_bytes": 359572231, "num_examples": 2703, "dataset_name": "librispeech_asr"}, "test": {"name": "test", "num_bytes": 367705423, "num_examples": 2620, "dataset_name": "librispeech_asr"}}, "download_checksums": {"http://www.openslr.org/resources/12/dev-clean.tar.gz": {"num_bytes": 337926286, "checksum": "76f87d090650617fca0cac8f88b9416e0ebf80350acb97b343a85fa903728ab3"}, "http://www.openslr.org/resources/12/test-clean.tar.gz": {"num_bytes": 346663984, "checksum": "39fde525e59672dc6d1551919b1478f724438a95aa55f874b576be21967e6c23"}, "http://www.openslr.org/resources/12/train-clean-100.tar.gz": {"num_bytes": 6387309499, "checksum": "d4ddd1d5a6ab303066f14971d768ee43278a5f2a0aa43dc716b0e64ecbbbf6e2"}, "http://www.openslr.org/resources/12/train-clean-360.tar.gz": {"num_bytes": 23049477885, "checksum": "146a56496217e96c14334a160df97fffedd6e0a04e66b9c5af0d40be3c792ecf"}}, "download_size": 30121377654, "post_processing_size": null, "dataset_size": 31245175287, "size_in_bytes": 61366552941}, "other": {"description": "LibriSpeech is a corpus of approximately 1000 hours of read English speech with sampling rate of 16 kHz,\nprepared by Vassil Panayotov with the assistance of Daniel Povey. The data is derived from read\naudiobooks from the LibriVox project, and has been carefully segmented and aligned.87\n\nNote that in order to limit the required storage for preparing this dataset, the audio\nis stored in the .flac format and is not converted to a float32 array. To convert, the audio\nfile to a float32 array, please make use of the `.map()` function as follows:\n\n\n```python\nimport soundfile as sf\n\ndef map_to_array(batch):\n speech_array, _ = sf.read(batch[\"file\"])\n batch[\"speech\"] = speech_array\n return batch\n\ndataset = dataset.map(map_to_array, remove_columns=[\"file\"])\n```\n", "citation": "@inproceedings{panayotov2015librispeech,\n title={Librispeech: an ASR corpus based on public domain audio books},\n author={Panayotov, Vassil and Chen, Guoguo and Povey, Daniel and Khudanpur, Sanjeev},\n booktitle={Acoustics, Speech and Signal Processing (ICASSP), 2015 IEEE International Conference on},\n pages={5206--5210},\n year={2015},\n organization={IEEE}\n}\n", "homepage": "http://www.openslr.org/12", "license": "", "features": {"file": {"dtype": "string", "id": null, "_type": "Value"}, "audio": {"sampling_rate": 16000, "mono": true, "_storage_dtype": "struct", "id": null, "_type": "Audio"}, "text": {"dtype": "string", "id": null, "_type": "Value"}, "speaker_id": {"dtype": "int64", "id": null, "_type": "Value"}, "chapter_id": {"dtype": "int64", "id": null, "_type": "Value"}, "id": {"dtype": "string", "id": null, "_type": "Value"}}, "post_processed": null, "supervised_keys": {"input": "file", "output": "text"}, "task_templates": [{"task": "automatic-speech-recognition", "audio_column": "audio", "transcription_column": "text"}], "builder_name": "librispeech_asr", "config_name": "other", "version": {"version_str": "2.1.0", "description": "", "major": 2, "minor": 1, "patch": 0}, "splits": {"train.500": {"name": "train.500", "num_bytes": 31810256902, "num_examples": 148688, "dataset_name": "librispeech_asr"}, "validation": {"name": "validation", "num_bytes": 337283304, "num_examples": 2864, "dataset_name": "librispeech_asr"}, "test": {"name": "test", "num_bytes": 352396474, "num_examples": 2939, "dataset_name": "librispeech_asr"}}, "download_checksums": {"http://www.openslr.org/resources/12/test-other.tar.gz": {"num_bytes": 328757843, "checksum": "d09c181bba5cf717b3dee7d4d592af11a3ee3a09e08ae025c5506f6ebe961c29"}, "http://www.openslr.org/resources/12/dev-other.tar.gz": {"num_bytes": 314305928, "checksum": "12661c48e8c3fe1de2c1caa4c3e135193bfb1811584f11f569dd12645aa84365"}, "http://www.openslr.org/resources/12/train-other-500.tar.gz": {"num_bytes": 30593501606, "checksum": "ddb22f27f96ec163645d53215559df6aa36515f26e01dd70798188350adcb6d2"}}, "download_size": 31236565377, "post_processing_size": null, "dataset_size": 32499936680, "size_in_bytes": 63736502057}}
{"clean": {"description": "LibriSpeech is a corpus of approximately 1000 hours of read English speech with sampling rate of 16 kHz,\nprepared by Vassil Panayotov with the assistance of Daniel Povey. The data is derived from read\naudiobooks from the LibriVox project, and has been carefully segmented and aligned.87\n\nNote that in order to limit the required storage for preparing this dataset, the audio\nis stored in the .flac format and is not converted to a float32 array. To convert, the audio\nfile to a float32 array, please make use of the `.map()` function as follows:\n\n\n```python\nimport soundfile as sf\n\ndef map_to_array(batch):\n speech_array, _ = sf.read(batch[\"file\"])\n batch[\"speech\"] = speech_array\n return batch\n\ndataset = dataset.map(map_to_array, remove_columns=[\"file\"])\n```\n", "citation": "@inproceedings{panayotov2015librispeech,\n title={Librispeech: an ASR corpus based on public domain audio books},\n author={Panayotov, Vassil and Chen, Guoguo and Povey, Daniel and Khudanpur, Sanjeev},\n booktitle={Acoustics, Speech and Signal Processing (ICASSP), 2015 IEEE International Conference on},\n pages={5206--5210},\n year={2015},\n organization={IEEE}\n}\n", "homepage": "http://www.openslr.org/12", "license": "", "features": {"file": {"dtype": "string", "id": null, "_type": "Value"}, "audio": {"sampling_rate": 16000, "mono": true, "decode": true, "id": null, "_type": "Audio"}, "text": {"dtype": "string", "id": null, "_type": "Value"}, "speaker_id": {"dtype": "int64", "id": null, "_type": "Value"}, "chapter_id": {"dtype": "int64", "id": null, "_type": "Value"}, "id": {"dtype": "string", "id": null, "_type": "Value"}}, "post_processed": null, "supervised_keys": {"input": "file", "output": "text"}, "task_templates": [{"task": "automatic-speech-recognition", "audio_column": "audio", "transcription_column": "text"}], "builder_name": "librispeech_asr", "config_name": "clean", "version": {"version_str": "2.1.0", "description": "", "major": 2, "minor": 1, "patch": 0}, "splits": {"train.100": {"name": "train.100", "num_bytes": 6619683041, "num_examples": 28539, "dataset_name": "librispeech_asr"}, "train.360": {"name": "train.360", "num_bytes": 23898214592, "num_examples": 104014, "dataset_name": "librispeech_asr"}, "validation": {"name": "validation", "num_bytes": 359572231, "num_examples": 2703, "dataset_name": "librispeech_asr"}, "test": {"name": "test", "num_bytes": 367705423, "num_examples": 2620, "dataset_name": "librispeech_asr"}}, "download_checksums": {"http://www.openslr.org/resources/12/dev-clean.tar.gz": {"num_bytes": 337926286, "checksum": "76f87d090650617fca0cac8f88b9416e0ebf80350acb97b343a85fa903728ab3"}, "http://www.openslr.org/resources/12/test-clean.tar.gz": {"num_bytes": 346663984, "checksum": "39fde525e59672dc6d1551919b1478f724438a95aa55f874b576be21967e6c23"}, "http://www.openslr.org/resources/12/train-clean-100.tar.gz": {"num_bytes": 6387309499, "checksum": "d4ddd1d5a6ab303066f14971d768ee43278a5f2a0aa43dc716b0e64ecbbbf6e2"}, "http://www.openslr.org/resources/12/train-clean-360.tar.gz": {"num_bytes": 23049477885, "checksum": "146a56496217e96c14334a160df97fffedd6e0a04e66b9c5af0d40be3c792ecf"}}, "download_size": 30121377654, "post_processing_size": null, "dataset_size": 31245175287, "size_in_bytes": 61366552941}, "other": {"description": "LibriSpeech is a corpus of approximately 1000 hours of read English speech with sampling rate of 16 kHz,\nprepared by Vassil Panayotov with the assistance of Daniel Povey. The data is derived from read\naudiobooks from the LibriVox project, and has been carefully segmented and aligned.87\n\nNote that in order to limit the required storage for preparing this dataset, the audio\nis stored in the .flac format and is not converted to a float32 array. To convert, the audio\nfile to a float32 array, please make use of the `.map()` function as follows:\n\n\n```python\nimport soundfile as sf\n\ndef map_to_array(batch):\n speech_array, _ = sf.read(batch[\"file\"])\n batch[\"speech\"] = speech_array\n return batch\n\ndataset = dataset.map(map_to_array, remove_columns=[\"file\"])\n```\n", "citation": "@inproceedings{panayotov2015librispeech,\n title={Librispeech: an ASR corpus based on public domain audio books},\n author={Panayotov, Vassil and Chen, Guoguo and Povey, Daniel and Khudanpur, Sanjeev},\n booktitle={Acoustics, Speech and Signal Processing (ICASSP), 2015 IEEE International Conference on},\n pages={5206--5210},\n year={2015},\n organization={IEEE}\n}\n", "homepage": "http://www.openslr.org/12", "license": "", "features": {"file": {"dtype": "string", "id": null, "_type": "Value"}, "audio": {"sampling_rate": 16000, "mono": true, "decode": true, "id": null, "_type": "Audio"}, "text": {"dtype": "string", "id": null, "_type": "Value"}, "speaker_id": {"dtype": "int64", "id": null, "_type": "Value"}, "chapter_id": {"dtype": "int64", "id": null, "_type": "Value"}, "id": {"dtype": "string", "id": null, "_type": "Value"}}, "post_processed": null, "supervised_keys": {"input": "file", "output": "text"}, "task_templates": [{"task": "automatic-speech-recognition", "audio_column": "audio", "transcription_column": "text"}], "builder_name": "librispeech_asr", "config_name": "other", "version": {"version_str": "2.1.0", "description": "", "major": 2, "minor": 1, "patch": 0}, "splits": {"train.500": {"name": "train.500", "num_bytes": 31810256902, "num_examples": 148688, "dataset_name": "librispeech_asr"}, "validation": {"name": "validation", "num_bytes": 337283304, "num_examples": 2864, "dataset_name": "librispeech_asr"}, "test": {"name": "test", "num_bytes": 352396474, "num_examples": 2939, "dataset_name": "librispeech_asr"}}, "download_checksums": {"http://www.openslr.org/resources/12/test-other.tar.gz": {"num_bytes": 328757843, "checksum": "d09c181bba5cf717b3dee7d4d592af11a3ee3a09e08ae025c5506f6ebe961c29"}, "http://www.openslr.org/resources/12/dev-other.tar.gz": {"num_bytes": 314305928, "checksum": "12661c48e8c3fe1de2c1caa4c3e135193bfb1811584f11f569dd12645aa84365"}, "http://www.openslr.org/resources/12/train-other-500.tar.gz": {"num_bytes": 30593501606, "checksum": "ddb22f27f96ec163645d53215559df6aa36515f26e01dd70798188350adcb6d2"}}, "download_size": 31236565377, "post_processing_size": null, "dataset_size": 32499936680, "size_in_bytes": 63736502057}, "all": {"description": "LibriSpeech is a corpus of approximately 1000 hours of read English speech with sampling rate of 16 kHz,\nprepared by Vassil Panayotov with the assistance of Daniel Povey. The data is derived from read\naudiobooks from the LibriVox project, and has been carefully segmented and aligned.87\n", "citation": "@inproceedings{panayotov2015librispeech,\n title={Librispeech: an ASR corpus based on public domain audio books},\n author={Panayotov, Vassil and Chen, Guoguo and Povey, Daniel and Khudanpur, Sanjeev},\n booktitle={Acoustics, Speech and Signal Processing (ICASSP), 2015 IEEE International Conference on},\n pages={5206--5210},\n year={2015},\n organization={IEEE}\n}\n", "homepage": "http://www.openslr.org/12", "license": "", "features": {"file": {"dtype": "string", "id": null, "_type": "Value"}, "audio": {"sampling_rate": 16000, "mono": true, "decode": true, "id": null, "_type": "Audio"}, "text": {"dtype": "string", "id": null, "_type": "Value"}, "speaker_id": {"dtype": "int64", "id": null, "_type": "Value"}, "chapter_id": {"dtype": "int64", "id": null, "_type": "Value"}, "id": {"dtype": "string", "id": null, "_type": "Value"}}, "post_processed": null, "supervised_keys": {"input": "file", "output": "text"}, "task_templates": [{"task": "automatic-speech-recognition", "audio_column": "audio", "transcription_column": "text"}], "builder_name": "librispeech_asr", "config_name": "all", "version": {"version_str": "2.1.0", "description": "", "major": 2, "minor": 1, "patch": 0}, "splits": {"train.clean.100": {"name": "train.clean.100", "num_bytes": 6627791685, "num_examples": 28539, "dataset_name": "librispeech_asr"}, "train.clean.360": {"name": "train.clean.360", "num_bytes": 23927767570, "num_examples": 104014, "dataset_name": "librispeech_asr"}, "train.other.500": {"name": "train.other.500", "num_bytes": 31852502880, "num_examples": 148688, "dataset_name": "librispeech_asr"}, "validation.clean": {"name": "validation.clean", "num_bytes": 359505691, "num_examples": 2703, "dataset_name": "librispeech_asr"}, "validation.other": {"name": "validation.other", "num_bytes": 337213112, "num_examples": 2864, "dataset_name": "librispeech_asr"}, "test.clean": {"name": "test.clean", "num_bytes": 368449831, "num_examples": 2620, "dataset_name": "librispeech_asr"}, "test.other": {"name": "test.other", "num_bytes": 353231518, "num_examples": 2939, "dataset_name": "librispeech_asr"}}, "download_checksums": {"http://www.openslr.org/resources/12/dev-clean.tar.gz": {"num_bytes": 337926286, "checksum": "76f87d090650617fca0cac8f88b9416e0ebf80350acb97b343a85fa903728ab3"}, "http://www.openslr.org/resources/12/dev-other.tar.gz": {"num_bytes": 314305928, "checksum": "12661c48e8c3fe1de2c1caa4c3e135193bfb1811584f11f569dd12645aa84365"}, "http://www.openslr.org/resources/12/test-clean.tar.gz": {"num_bytes": 346663984, "checksum": "39fde525e59672dc6d1551919b1478f724438a95aa55f874b576be21967e6c23"}, "http://www.openslr.org/resources/12/test-other.tar.gz": {"num_bytes": 328757843, "checksum": "d09c181bba5cf717b3dee7d4d592af11a3ee3a09e08ae025c5506f6ebe961c29"}, "http://www.openslr.org/resources/12/train-clean-100.tar.gz": {"num_bytes": 6387309499, "checksum": "d4ddd1d5a6ab303066f14971d768ee43278a5f2a0aa43dc716b0e64ecbbbf6e2"}, "http://www.openslr.org/resources/12/train-clean-360.tar.gz": {"num_bytes": 23049477885, "checksum": "146a56496217e96c14334a160df97fffedd6e0a04e66b9c5af0d40be3c792ecf"}, "http://www.openslr.org/resources/12/train-other-500.tar.gz": {"num_bytes": 30593501606, "checksum": "ddb22f27f96ec163645d53215559df6aa36515f26e01dd70798188350adcb6d2"}}, "download_size": 61357943031, "post_processing_size": null, "dataset_size": 63826462287, "size_in_bytes": 125184405318}}
Loading

1 comment on commit 91d7171

@github-actions
Copy link

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Show benchmarks

PyArrow==5.0.0

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) 0.010099 / 0.011353 (-0.001254) 0.004258 / 0.011008 (-0.006750) 0.031552 / 0.038508 (-0.006956) 0.036371 / 0.023109 (0.013262) 0.296150 / 0.275898 (0.020252) 0.331774 / 0.323480 (0.008294) 0.008521 / 0.007986 (0.000535) 0.003929 / 0.004328 (-0.000399) 0.009361 / 0.004250 (0.005110) 0.043137 / 0.037052 (0.006084) 0.286031 / 0.258489 (0.027542) 0.328420 / 0.293841 (0.034579) 0.032469 / 0.128546 (-0.096077) 0.010072 / 0.075646 (-0.065574) 0.254420 / 0.419271 (-0.164851) 0.054134 / 0.043533 (0.010601) 0.283849 / 0.255139 (0.028710) 0.308047 / 0.283200 (0.024847) 0.115568 / 0.141683 (-0.026115) 1.722672 / 1.452155 (0.270518) 1.813688 / 1.492716 (0.320972)

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) 0.306566 / 0.018006 (0.288560) 0.555368 / 0.000490 (0.554878) 0.007495 / 0.000200 (0.007295) 0.000403 / 0.000054 (0.000348)

Benchmark: benchmark_indices_mapping.json

metric select shard shuffle sort train_test_split
new / old (diff) 0.024397 / 0.037411 (-0.013014) 0.093660 / 0.014526 (0.079134) 0.103536 / 0.176557 (-0.073020) 0.146170 / 0.737135 (-0.590965) 0.103919 / 0.296338 (-0.192419)

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) 0.362255 / 0.215209 (0.147046) 3.636312 / 2.077655 (1.558657) 1.550471 / 1.504120 (0.046351) 1.360088 / 1.541195 (-0.181106) 1.440120 / 1.468490 (-0.028370) 0.382640 / 4.584777 (-4.202137) 4.534776 / 3.745712 (0.789064) 2.302796 / 5.269862 (-2.967065) 0.908674 / 4.565676 (-3.657003) 0.052784 / 0.424275 (-0.371491) 0.012362 / 0.007607 (0.004755) 0.512205 / 0.226044 (0.286161) 5.116242 / 2.268929 (2.847313) 2.182639 / 55.444624 (-53.261985) 1.833008 / 6.876477 (-5.043468) 2.020157 / 2.142072 (-0.121916) 0.553735 / 4.805227 (-4.251492) 0.123267 / 6.500664 (-6.377397) 0.061548 / 0.075469 (-0.013921)

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) 1.436580 / 1.841788 (-0.405207) 13.125056 / 8.074308 (5.050748) 23.780364 / 10.191392 (13.588972) 0.754520 / 0.680424 (0.074096) 0.452297 / 0.534201 (-0.081904) 0.441918 / 0.579283 (-0.137365) 0.461624 / 0.434364 (0.027260) 0.287378 / 0.540337 (-0.252959) 0.295814 / 1.386936 (-1.091122)
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) 0.007663 / 0.011353 (-0.003690) 0.003810 / 0.011008 (-0.007198) 0.026302 / 0.038508 (-0.012206) 0.030748 / 0.023109 (0.007639) 0.293834 / 0.275898 (0.017936) 0.301242 / 0.323480 (-0.022238) 0.005756 / 0.007986 (-0.002230) 0.004652 / 0.004328 (0.000324) 0.006691 / 0.004250 (0.002441) 0.036321 / 0.037052 (-0.000731) 0.268498 / 0.258489 (0.010009) 0.301568 / 0.293841 (0.007727) 0.028733 / 0.128546 (-0.099814) 0.009065 / 0.075646 (-0.066582) 0.224062 / 0.419271 (-0.195210) 0.045980 / 0.043533 (0.002447) 0.281123 / 0.255139 (0.025984) 0.288517 / 0.283200 (0.005317) 0.087568 / 0.141683 (-0.054115) 1.538256 / 1.452155 (0.086102) 1.637772 / 1.492716 (0.145055)

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) 0.361210 / 0.018006 (0.343204) 0.558216 / 0.000490 (0.557726) 0.008490 / 0.000200 (0.008290) 0.000104 / 0.000054 (0.000049)

Benchmark: benchmark_indices_mapping.json

metric select shard shuffle sort train_test_split
new / old (diff) 0.023976 / 0.037411 (-0.013435) 0.093013 / 0.014526 (0.078487) 0.103831 / 0.176557 (-0.072725) 0.144938 / 0.737135 (-0.592197) 0.104815 / 0.296338 (-0.191523)

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) 0.372910 / 0.215209 (0.157701) 3.743260 / 2.077655 (1.665605) 1.604610 / 1.504120 (0.100490) 1.418733 / 1.541195 (-0.122461) 1.500879 / 1.468490 (0.032389) 0.392282 / 4.584777 (-4.192495) 4.609975 / 3.745712 (0.864263) 2.169978 / 5.269862 (-3.099883) 0.930090 / 4.565676 (-3.635586) 0.053246 / 0.424275 (-0.371029) 0.011962 / 0.007607 (0.004355) 0.529265 / 0.226044 (0.303220) 5.292541 / 2.268929 (3.023612) 2.277513 / 55.444624 (-53.167112) 1.918541 / 6.876477 (-4.957936) 2.045144 / 2.142072 (-0.096929) 0.558690 / 4.805227 (-4.246538) 0.122905 / 6.500664 (-6.377759) 0.061179 / 0.075469 (-0.014290)

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) 1.428077 / 1.841788 (-0.413711) 13.486728 / 8.074308 (5.412420) 26.940811 / 10.191392 (16.749419) 0.874520 / 0.680424 (0.194096) 0.469655 / 0.534201 (-0.064546) 0.436746 / 0.579283 (-0.142537) 0.508204 / 0.434364 (0.073840) 0.293367 / 0.540337 (-0.246970) 0.305694 / 1.386936 (-1.081242)

CML watermark

Please sign in to comment.