Fix UnboundLocalError if preprocessing returns an empty list

[cwallenwein]

If this tokenization function is used with IterableDatasets and no sample is as big as the context length, input_batch will be an empty list.

def tokenize(batch, tokenizer, context_length):
    outputs = tokenizer(
        batch["text"],
        truncation=True,
        max_length=context_length,
        return_overflowing_tokens=True,
        return_length=True
    )
    input_batch = []
    for length, input_ids in zip(outputs["length"], outputs["input_ids"]):
        if length == context_length:
            input_batch.append(input_ids)
    return {"input_ids": input_batch}

dataset.map(tokenize, batched=True, batch_size=batch_size, fn_kwargs={"context_length": context_length, "tokenizer": tokenizer}, remove_columns=dataset.column_names)

This will throw the following error: UnboundLocalError: local variable 'batch_idx' referenced before assignment, because the for loop was not executed a single time

for batch_idx, example in enumerate(_batch_to_examples(transformed_batch)):
     yield new_key, example
current_idx += batch_idx + 1

Some of the possible solutions

for batch_idx, example in enumerate(_batch_to_examples(transformed_batch)):
     yield new_key, example
try:
     current_idx += batch_idx + 1
except:
     current_idx += 1

or

batch_idx = 0
for batch_idx, example in enumerate(_batch_to_examples(transformed_batch)):
     yield new_key, example
current_idx += batch_idx + 1

[mariosasko]

Thanks!

The batched .filter also has this issue, so feel free to fix it too as part of this PR.

[HuggingFaceDocBuilderDev]

The documentation is not available anymore as the PR was closed or merged.

[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)	0.009286 / 0.011353 (-0.002067)	0.005478 / 0.011008 (-0.005530)	0.109768 / 0.038508 (0.071260)	0.088460 / 0.023109 (0.065351)	0.387664 / 0.275898 (0.111766)	0.457379 / 0.323480 (0.133899)	0.006517 / 0.007986 (-0.001469)	0.004037 / 0.004328 (-0.000292)	0.083911 / 0.004250 (0.079661)	0.071658 / 0.037052 (0.034605)	0.385065 / 0.258489 (0.126576)	0.460928 / 0.293841 (0.167087)	0.048062 / 0.128546 (-0.080484)	0.016343 / 0.075646 (-0.059303)	0.373675 / 0.419271 (-0.045597)	0.067640 / 0.043533 (0.024108)	0.391730 / 0.255139 (0.136591)	0.432908 / 0.283200 (0.149708)	0.035748 / 0.141683 (-0.105935)	1.767625 / 1.452155 (0.315471)	1.965606 / 1.492716 (0.472889)

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)	0.277405 / 0.018006 (0.259399)	0.538448 / 0.000490 (0.537958)	0.013795 / 0.000200 (0.013595)	0.000518 / 0.000054 (0.000464)

Benchmark: benchmark_indices_mapping.json

metric	select	shard	shuffle	sort	train_test_split
new / old (diff)	0.043962 / 0.037411 (0.006550)	0.115305 / 0.014526 (0.100780)	0.117572 / 0.176557 (-0.058985)	0.182168 / 0.737135 (-0.554968)	0.114833 / 0.296338 (-0.181505)

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)	0.604209 / 0.215209 (0.389000)	6.186113 / 2.077655 (4.108458)	2.771067 / 1.504120 (1.266947)	2.425420 / 1.541195 (0.884226)	2.475200 / 1.468490 (1.006710)	0.887096 / 4.584777 (-3.697681)	5.214349 / 3.745712 (1.468637)	4.989606 / 5.269862 (-0.280256)	3.092135 / 4.565676 (-1.473541)	0.104464 / 0.424275 (-0.319811)	0.008994 / 0.007607 (0.001387)	0.732819 / 0.226044 (0.506775)	7.396007 / 2.268929 (5.127078)	3.371167 / 55.444624 (-52.073457)	2.645475 / 6.876477 (-4.231001)	2.704215 / 2.142072 (0.562143)	1.034724 / 4.805227 (-3.770504)	0.219063 / 6.500664 (-6.281601)	0.073863 / 0.075469 (-0.001606)

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)	1.625020 / 1.841788 (-0.216768)	23.369980 / 8.074308 (15.295671)	22.480951 / 10.191392 (12.289559)	0.228219 / 0.680424 (-0.452204)	0.026981 / 0.534201 (-0.507220)	0.487670 / 0.579283 (-0.091613)	0.582310 / 0.434364 (0.147946)	0.539182 / 0.540337 (-0.001156)	0.791962 / 1.386936 (-0.594974)

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)	0.008657 / 0.011353 (-0.002696)	0.004971 / 0.011008 (-0.006037)	0.089499 / 0.038508 (0.050991)	0.075963 / 0.023109 (0.052854)	0.497719 / 0.275898 (0.221821)	0.507912 / 0.323480 (0.184432)	0.006067 / 0.007986 (-0.001919)	0.004118 / 0.004328 (-0.000210)	0.079397 / 0.004250 (0.075146)	0.059181 / 0.037052 (0.022129)	0.501108 / 0.258489 (0.242619)	0.565792 / 0.293841 (0.271951)	0.048818 / 0.128546 (-0.079729)	0.014813 / 0.075646 (-0.060833)	0.093863 / 0.419271 (-0.325409)	0.060824 / 0.043533 (0.017292)	0.489289 / 0.255139 (0.234150)	0.533624 / 0.283200 (0.250425)	0.034997 / 0.141683 (-0.106685)	1.770574 / 1.452155 (0.318419)	1.837213 / 1.492716 (0.344496)

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)	0.237319 / 0.018006 (0.219313)	0.594976 / 0.000490 (0.594486)	0.008888 / 0.000200 (0.008688)	0.000124 / 0.000054 (0.000070)

Benchmark: benchmark_indices_mapping.json

metric	select	shard	shuffle	sort	train_test_split
new / old (diff)	0.036955 / 0.037411 (-0.000456)	0.097825 / 0.014526 (0.083299)	0.111139 / 0.176557 (-0.065418)	0.174776 / 0.737135 (-0.562359)	0.117755 / 0.296338 (-0.178584)

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)	0.606498 / 0.215209 (0.391289)	6.089874 / 2.077655 (4.012219)	2.811135 / 1.504120 (1.307015)	2.428486 / 1.541195 (0.887292)	2.399512 / 1.468490 (0.931022)	0.823492 / 4.584777 (-3.761285)	4.897107 / 3.745712 (1.151395)	4.407589 / 5.269862 (-0.862272)	2.868442 / 4.565676 (-1.697235)	0.098774 / 0.424275 (-0.325502)	0.007998 / 0.007607 (0.000391)	0.699489 / 0.226044 (0.473445)	7.139214 / 2.268929 (4.870285)	3.511158 / 55.444624 (-51.933466)	2.775459 / 6.876477 (-4.101018)	2.951549 / 2.142072 (0.809477)	1.006921 / 4.805227 (-3.798306)	0.200105 / 6.500664 (-6.300559)	0.071064 / 0.075469 (-0.004405)

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)	1.680599 / 1.841788 (-0.161189)	23.399777 / 8.074308 (15.325469)	21.776357 / 10.191392 (11.584965)	0.264697 / 0.680424 (-0.415726)	0.034272 / 0.534201 (-0.499929)	0.506984 / 0.579283 (-0.072299)	0.609556 / 0.434364 (0.175192)	0.599014 / 0.540337 (0.058677)	0.824068 / 1.386936 (-0.562868)