Fix push_to_hub by not calling create_branch if PR branch

[albertvillanova]

Fix push_to_hub by not calling create_branch if PR branch (e.g. refs/pr/1).

Note that currently create_branch raises a 400 Bad Request error if the user passes a PR branch (e.g. refs/pr/1).

EDIT:
Fix push_to_hub by not calling create_branch if branch exists.

Note that currently create_branch raises a 403 Forbidden error even if all these conditions are met:

• exist_ok is passed
• the branch already exists
• the user does not have WRITE permission

Fix #7067.

Related issue:

• Different behavior when passing exist_ok huggingface_hub#2419

[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.

[lhoestq]

cc @Wauplin maybe it's a huggingface_hub bug ?

EDIT: ah actually the issue is opened at huggingface/huggingface_hub#2419

[albertvillanova]

Alternatively:

Suggested change

	if revision is not None and not api.revision_exists(repo_id, revision, repo_type="dataset", token=token):
	if revision is not None and not revision.startswith("refs/pr/"):

[albertvillanova]

I think this modification would be more efficient:

• The modification above will fix the 400 Bad Request error without the additional network call revision_exists.
• In the case of a 403 Forbidden error, anyway we will raise an error some code lines below when trying to push to the branch if the user does not have write permission.

[albertvillanova]

On the other hand, it is true that the call to revision_exists is a more general solution.

[albertvillanova]

I think we need to make this fix anyway, unless we pin the lower version of huggingface-hub (once they release the patch).

• Calling create_branch with a PR ref raises an error

[albertvillanova]

Comment by @Wauplin: huggingface/huggingface_hub#2426 (comment)

I think this should be something to fix in datasets directly. Having a 400 Bad request when trying to create the branch refs/pr/1 seems normal to me since it's not a branch.

[lhoestq]

does this mean we should use create_pull_request() in that case ?

[Wauplin]

does this mean we should use create_pull_request() in that case ?

If user wants to push some data to a new PR, they can already pass create_pr=True which will automatically do the job for you (without using revision). If user is passing revision="refs/pr/1" explicitly, you should assume the PR already exists.

[lhoestq]

ah yes we do pass create_pr in preupload_lfs_files() ! sounds good then

[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)	0.005806 / 0.011353 (-0.005547)	0.004082 / 0.011008 (-0.006927)	0.064277 / 0.038508 (0.025769)	0.032289 / 0.023109 (0.009180)	0.242066 / 0.275898 (-0.033832)	0.272574 / 0.323480 (-0.050906)	0.003281 / 0.007986 (-0.004705)	0.002957 / 0.004328 (-0.001371)	0.049930 / 0.004250 (0.045679)	0.047306 / 0.037052 (0.010253)	0.252216 / 0.258489 (-0.006273)	0.286678 / 0.293841 (-0.007163)	0.030182 / 0.128546 (-0.098364)	0.012967 / 0.075646 (-0.062680)	0.204949 / 0.419271 (-0.214323)	0.036999 / 0.043533 (-0.006534)	0.243577 / 0.255139 (-0.011562)	0.265044 / 0.283200 (-0.018156)	0.021149 / 0.141683 (-0.120534)	1.112293 / 1.452155 (-0.339862)	1.186483 / 1.492716 (-0.306233)

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)	0.093239 / 0.018006 (0.075233)	0.286372 / 0.000490 (0.285883)	0.000224 / 0.000200 (0.000024)	0.000062 / 0.000054 (0.000007)

Benchmark: benchmark_indices_mapping.json

metric	select	shard	shuffle	sort	train_test_split
new / old (diff)	0.019042 / 0.037411 (-0.018369)	0.063690 / 0.014526 (0.049164)	0.075034 / 0.176557 (-0.101523)	0.123053 / 0.737135 (-0.614083)	0.076843 / 0.296338 (-0.219495)

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)	0.276554 / 0.215209 (0.061345)	2.749338 / 2.077655 (0.671683)	1.442764 / 1.504120 (-0.061356)	1.327860 / 1.541195 (-0.213335)	1.369885 / 1.468490 (-0.098606)	0.722645 / 4.584777 (-3.862132)	2.430707 / 3.745712 (-1.315005)	3.105293 / 5.269862 (-2.164568)	1.961617 / 4.565676 (-2.604060)	0.077728 / 0.424275 (-0.346547)	0.005189 / 0.007607 (-0.002418)	0.335511 / 0.226044 (0.109467)	3.315618 / 2.268929 (1.046690)	1.858254 / 55.444624 (-53.586371)	1.552173 / 6.876477 (-5.324304)	1.627086 / 2.142072 (-0.514987)	0.790871 / 4.805227 (-4.014356)	0.136958 / 6.500664 (-6.363706)	0.043207 / 0.075469 (-0.032262)

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)	0.969314 / 1.841788 (-0.872473)	12.145318 / 8.074308 (4.071010)	9.834839 / 10.191392 (-0.356553)	0.141896 / 0.680424 (-0.538528)	0.014304 / 0.534201 (-0.519897)	0.306091 / 0.579283 (-0.273192)	0.260994 / 0.434364 (-0.173369)	0.348096 / 0.540337 (-0.192242)	0.441458 / 1.386936 (-0.945478)

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)	0.005989 / 0.011353 (-0.005363)	0.003907 / 0.011008 (-0.007102)	0.050819 / 0.038508 (0.012310)	0.033178 / 0.023109 (0.010069)	0.279059 / 0.275898 (0.003161)	0.300161 / 0.323480 (-0.023319)	0.004383 / 0.007986 (-0.003603)	0.002834 / 0.004328 (-0.001495)	0.048779 / 0.004250 (0.044528)	0.040502 / 0.037052 (0.003450)	0.291786 / 0.258489 (0.033297)	0.323827 / 0.293841 (0.029986)	0.032175 / 0.128546 (-0.096371)	0.012157 / 0.075646 (-0.063489)	0.060796 / 0.419271 (-0.358476)	0.033924 / 0.043533 (-0.009609)	0.278047 / 0.255139 (0.022908)	0.297878 / 0.283200 (0.014678)	0.019137 / 0.141683 (-0.122546)	1.138996 / 1.452155 (-0.313158)	1.172731 / 1.492716 (-0.319985)

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)	0.110148 / 0.018006 (0.092142)	0.307232 / 0.000490 (0.306742)	0.000209 / 0.000200 (0.000009)	0.000044 / 0.000054 (-0.000010)

Benchmark: benchmark_indices_mapping.json

metric	select	shard	shuffle	sort	train_test_split
new / old (diff)	0.023082 / 0.037411 (-0.014330)	0.076590 / 0.014526 (0.062065)	0.088444 / 0.176557 (-0.088113)	0.129293 / 0.737135 (-0.607842)	0.090470 / 0.296338 (-0.205868)

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)	0.305016 / 0.215209 (0.089807)	2.931671 / 2.077655 (0.854016)	1.586055 / 1.504120 (0.081935)	1.463517 / 1.541195 (-0.077678)	1.479654 / 1.468490 (0.011164)	0.726194 / 4.584777 (-3.858583)	0.970512 / 3.745712 (-2.775200)	2.850496 / 5.269862 (-2.419365)	1.920112 / 4.565676 (-2.645564)	0.079921 / 0.424275 (-0.344354)	0.005367 / 0.007607 (-0.002240)	0.347022 / 0.226044 (0.120978)	3.472425 / 2.268929 (1.203497)	1.965400 / 55.444624 (-53.479225)	1.669116 / 6.876477 (-5.207361)	1.859504 / 2.142072 (-0.282568)	0.802703 / 4.805227 (-4.002525)	0.134776 / 6.500664 (-6.365888)	0.041800 / 0.075469 (-0.033669)

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)	1.039665 / 1.841788 (-0.802122)	12.024071 / 8.074308 (3.949763)	10.338743 / 10.191392 (0.147351)	0.139495 / 0.680424 (-0.540929)	0.015249 / 0.534201 (-0.518952)	0.298580 / 0.579283 (-0.280703)	0.124625 / 0.434364 (-0.309739)	0.341868 / 0.540337 (-0.198470)	0.431396 / 1.386936 (-0.955540)