Fix TypeError: Object of type int64 is not JSON serializable (#24340)
* Fix TypeError: Object of type int64 is not JSON serializable * Convert numpy.float64 and numpy.int64 to float and int for json serialization * Black reformatted examples/pytorch/token-classification/run_ner_no_trainer.py * * make style
This commit is contained in:
parent
ac19871ce2
commit
239ace152b
|
@ -28,6 +28,7 @@ from pathlib import Path
|
|||
|
||||
import datasets
|
||||
import evaluate
|
||||
import numpy as np
|
||||
import torch
|
||||
from accelerate import Accelerator
|
||||
from accelerate.logging import get_logger
|
||||
|
@ -777,6 +778,12 @@ def main():
|
|||
if args.with_tracking:
|
||||
all_results.update({"train_loss": total_loss.item() / len(train_dataloader)})
|
||||
with open(os.path.join(args.output_dir, "all_results.json"), "w") as f:
|
||||
# Convert all float64 & int64 type numbers to float & int for json serialization
|
||||
for key, value in all_results.items():
|
||||
if isinstance(value, np.float64):
|
||||
all_results[key] = float(value)
|
||||
elif isinstance(value, np.int64):
|
||||
all_results[key] = int(value)
|
||||
json.dump(all_results, f)
|
||||
|
||||
|
||||
|
|
|
@ -60,7 +60,7 @@ class EndOfFunctionCriteria(StoppingCriteria):
|
|||
decoded_generations = self.tokenizer.batch_decode(input_ids[:, self.start_length :])
|
||||
done = []
|
||||
for decoded_generation in decoded_generations:
|
||||
done.append(any([stop_string in decoded_generation for stop_string in self.eof_strings]))
|
||||
done.append(any(stop_string in decoded_generation for stop_string in self.eof_strings))
|
||||
return all(done)
|
||||
|
||||
|
||||
|
|
|
@ -17,7 +17,7 @@ class FSNERTokenizerUtils(object):
|
|||
`transformers.tokenization_utils_base.BatchEncoding` dict with additional keys and values for start_token_id, end_token_id and sizes of example lists for each entity type
|
||||
"""
|
||||
|
||||
if isinstance(x, list) and all([isinstance(_x, list) for _x in x]):
|
||||
if isinstance(x, list) and all(isinstance(_x, list) for _x in x):
|
||||
d = None
|
||||
for l in x:
|
||||
t = self.tokenizer(
|
||||
|
@ -37,7 +37,7 @@ class FSNERTokenizerUtils(object):
|
|||
d["start_token_id"] = torch.tensor(self.tokenizer.convert_tokens_to_ids("[E]"))
|
||||
d["end_token_id"] = torch.tensor(self.tokenizer.convert_tokens_to_ids("[/E]"))
|
||||
|
||||
elif isinstance(x, list) and all([isinstance(_x, str) for _x in x]):
|
||||
elif isinstance(x, list) and all(isinstance(_x, str) for _x in x):
|
||||
d = self.tokenizer(
|
||||
x,
|
||||
padding="max_length",
|
||||
|
|
|
@ -50,7 +50,7 @@ def _get_single_answer(example):
|
|||
answer["remove_it"] = False
|
||||
|
||||
cols = ["start_token", "end_token", "start_byte", "end_byte", "text"]
|
||||
if not all([isinstance(answer[k], list) for k in cols]):
|
||||
if not all(isinstance(answer[k], list) for k in cols):
|
||||
raise ValueError("Issue in ID", example["id"])
|
||||
|
||||
return answer
|
||||
|
|
|
@ -610,7 +610,7 @@ def main():
|
|||
predicted_sequence = [label_list[0]] * len(true_tags)
|
||||
|
||||
for _, span, label in sorted(predictions, key=lambda o: o[0], reverse=True):
|
||||
if all([o == label_list[0] for o in predicted_sequence[span[0] : span[1]]]):
|
||||
if all(o == label_list[0] for o in predicted_sequence[span[0] : span[1]]):
|
||||
predicted_sequence[span[0]] = label
|
||||
if span[1] - span[0] > 1:
|
||||
predicted_sequence[span[0] + 1 : span[1]] = [label] * (span[1] - span[0] - 1)
|
||||
|
|
|
@ -554,8 +554,8 @@ class Matcher(object):
|
|||
assert thresholds[0] > 0
|
||||
thresholds.insert(0, -float("inf"))
|
||||
thresholds.append(float("inf"))
|
||||
assert all([low <= high for (low, high) in zip(thresholds[:-1], thresholds[1:])])
|
||||
assert all([label_i in [-1, 0, 1] for label_i in labels])
|
||||
assert all(low <= high for (low, high) in zip(thresholds[:-1], thresholds[1:]))
|
||||
assert all(label_i in [-1, 0, 1] for label_i in labels)
|
||||
assert len(labels) == len(thresholds) - 1
|
||||
self.thresholds = thresholds
|
||||
self.labels = labels
|
||||
|
|
|
@ -554,8 +554,8 @@ class Matcher(object):
|
|||
assert thresholds[0] > 0
|
||||
thresholds.insert(0, -float("inf"))
|
||||
thresholds.append(float("inf"))
|
||||
assert all([low <= high for (low, high) in zip(thresholds[:-1], thresholds[1:])])
|
||||
assert all([label_i in [-1, 0, 1] for label_i in labels])
|
||||
assert all(low <= high for (low, high) in zip(thresholds[:-1], thresholds[1:]))
|
||||
assert all(label_i in [-1, 0, 1] for label_i in labels)
|
||||
assert len(labels) == len(thresholds) - 1
|
||||
self.thresholds = thresholds
|
||||
self.labels = labels
|
||||
|
|
|
@ -110,7 +110,7 @@ class MinLengthLogitsProcessor(LogitsProcessor):
|
|||
|
||||
if isinstance(eos_token_id, int):
|
||||
eos_token_id = [eos_token_id]
|
||||
if not all([isinstance(i, int) for i in eos_token_id]) or any([i < 0 for i in eos_token_id]):
|
||||
if not all(isinstance(i, int) for i in eos_token_id) or any(i < 0 for i in eos_token_id):
|
||||
logger.warning(f"`eos_token_id` has to be a list of positive integers, but is {eos_token_id}")
|
||||
|
||||
self.min_length = min_length
|
||||
|
@ -147,7 +147,7 @@ class MinNewTokensLengthLogitsProcessor(LogitsProcessor):
|
|||
|
||||
if isinstance(eos_token_id, int):
|
||||
eos_token_id = [eos_token_id]
|
||||
if not all([isinstance(i, int) for i in eos_token_id]) or any([i < 0 for i in eos_token_id]):
|
||||
if not all(isinstance(i, int) for i in eos_token_id) or any(i < 0 for i in eos_token_id):
|
||||
logger.warning(f"`eos_token_id` has to be a list of positive integers, but is {eos_token_id}")
|
||||
|
||||
self.prompt_length_to_skip = prompt_length_to_skip
|
||||
|
@ -731,7 +731,7 @@ class NoBadWordsLogitsProcessor(SequenceBiasLogitsProcessor):
|
|||
if isinstance(eos_token_id, int):
|
||||
eos_token_id = [eos_token_id]
|
||||
bad_words_ids = list(
|
||||
filter(lambda bad_token_seq: all([bad_token_seq != [i] for i in eos_token_id]), bad_words_ids)
|
||||
filter(lambda bad_token_seq: all(bad_token_seq != [i] for i in eos_token_id), bad_words_ids)
|
||||
)
|
||||
|
||||
# Forbidding a sequence is equivalent to setting its bias to -inf
|
||||
|
|
|
@ -318,7 +318,7 @@ class TFNoBadWordsLogitsProcessor(TFLogitsProcessor):
|
|||
self.bad_word_seqs_ids = tf.ragged.constant(bad_words_ids).to_tensor(default_value=-1)
|
||||
# 2. a tensor with the unpadded length of each forbidden sequence, for quick length comparisons
|
||||
bad_word_seqs_len = [len(bad_words) for bad_words in bad_words_ids]
|
||||
if any([word_len == 0 for word_len in bad_word_seqs_len]):
|
||||
if any(word_len == 0 for word_len in bad_word_seqs_len):
|
||||
raise ValueError(f"Banned words token sequences {bad_words_ids} cannot have an empty list")
|
||||
self.bad_word_seqs_len = tf.convert_to_tensor(bad_word_seqs_len, dtype=tf.int32)
|
||||
# 3. a tensor containing the last token for each sequence, for easy access to the tokens that may be banned
|
||||
|
|
|
@ -1638,7 +1638,7 @@ class TFGenerationMixin:
|
|||
# TODO (Joao): fix cache format or find programatic way to detect cache index
|
||||
# GPT2 and other models has a slightly different cache structure, with a different batch axis
|
||||
model_name = str(self.decoder) if "EncoderDecoder" in str(self) else str(self)
|
||||
cache_batch_axis = 1 if any([model_prefix in model_name for model_prefix in ("TFGPT2", "TFCTRL")]) else 0
|
||||
cache_batch_axis = 1 if any(model_prefix in model_name for model_prefix in ("TFGPT2", "TFCTRL")) else 0
|
||||
# some models, like XLNet, need more than the last token in the presence of past_key_values
|
||||
needs_full_input = "use_mems" in set(inspect.signature(self.prepare_inputs_for_generation).parameters.keys())
|
||||
|
||||
|
@ -1922,7 +1922,7 @@ class TFGenerationMixin:
|
|||
# TODO (Joao): fix cache format or find programatic way to detect cache index
|
||||
# GPT2 and other models has a slightly different cache structure, with a different batch axis
|
||||
model_name = str(self.decoder) if "EncoderDecoder" in str(self) else str(self)
|
||||
cache_batch_axis = 1 if any([model_prefix in model_name for model_prefix in ("TFGPT2", "TFCTRL")]) else 0
|
||||
cache_batch_axis = 1 if any(model_prefix in model_name for model_prefix in ("TFGPT2", "TFCTRL")) else 0
|
||||
# some models, like XLNet, need more than the last token in the presence of past_key_values
|
||||
needs_full_input = "use_mems" in set(inspect.signature(self.prepare_inputs_for_generation).parameters.keys())
|
||||
|
||||
|
@ -2265,7 +2265,7 @@ class TFGenerationMixin:
|
|||
# TODO (Joao): fix cache format or find programatic way to detect cache index
|
||||
# GPT2 and other models has a slightly different cache structure, with a different batch axis
|
||||
model_name = str(self.decoder) if "EncoderDecoder" in str(self) else str(self)
|
||||
cache_batch_axis = 1 if any([model_prefix in model_name for model_prefix in ("TFGPT2", "TFCTRL")]) else 0
|
||||
cache_batch_axis = 1 if any(model_prefix in model_name for model_prefix in ("TFGPT2", "TFCTRL")) else 0
|
||||
# some models, like XLNet, need more than the last token in the presence of past_key_values
|
||||
needs_full_input = "use_mems" in set(inspect.signature(self.prepare_inputs_for_generation).parameters.keys())
|
||||
|
||||
|
@ -2779,7 +2779,7 @@ class TFGenerationMixin:
|
|||
# TODO (Joao): fix cache format or find programatic way to detect cache index
|
||||
# GPT2 and other models has a slightly different cache structure, with a different batch axis
|
||||
model_name = str(self.decoder) if "EncoderDecoder" in str(self) else str(self)
|
||||
cache_batch_axis = 1 if any([model_prefix in model_name for model_prefix in ("TFGPT2", "TFCTRL")]) else 0
|
||||
cache_batch_axis = 1 if any(model_prefix in model_name for model_prefix in ("TFGPT2", "TFCTRL")) else 0
|
||||
|
||||
# 2. init `attentions`, `hidden_states`, and `scores` tuples
|
||||
scores = [] if (return_dict_in_generate and output_scores) else None
|
||||
|
|
|
@ -144,7 +144,7 @@ class KerasMetricCallback(Callback):
|
|||
@staticmethod
|
||||
def _concatenate_batches(batches, padding_index=-100):
|
||||
# If all batches are unidimensional or same length, do a simple concatenation
|
||||
if batches[0].ndim == 1 or all([batch.shape[1] == batches[0].shape[1] for batch in batches]):
|
||||
if batches[0].ndim == 1 or all(batch.shape[1] == batches[0].shape[1] for batch in batches):
|
||||
return np.concatenate(batches, axis=0)
|
||||
|
||||
# Welp, they're not the same length. Let's do some padding
|
||||
|
|
|
@ -78,7 +78,7 @@ def convert_pytorch_checkpoint_to_tf(model: BertModel, ckpt_dir: str, model_name
|
|||
for var_name in state_dict:
|
||||
tf_name = to_tf_var_name(var_name)
|
||||
torch_tensor = state_dict[var_name].numpy()
|
||||
if any([x in var_name for x in tensors_to_transpose]):
|
||||
if any(x in var_name for x in tensors_to_transpose):
|
||||
torch_tensor = torch_tensor.T
|
||||
tf_var = create_tf_var(tensor=torch_tensor, name=tf_name, session=session)
|
||||
tf.keras.backend.set_value(tf_var, torch_tensor)
|
||||
|
|
|
@ -104,7 +104,7 @@ def convert_bigbird_pegasus(tf_weights: dict, config_update: dict) -> BigBirdPeg
|
|||
new_k = rename_state_dict_key(k, patterns)
|
||||
if new_k not in state_dict:
|
||||
raise ValueError(f"could not find new key {new_k} in state dict. (converted from {k})")
|
||||
if any([True if i in k else False for i in ["dense", "query", "key", "value"]]):
|
||||
if any(True if i in k else False for i in ["dense", "query", "key", "value"]):
|
||||
v = v.T
|
||||
mapping[new_k] = torch.from_numpy(v)
|
||||
assert v.shape == state_dict[new_k].shape, f"{new_k}, {k}, {v.shape}, {state_dict[new_k].shape}"
|
||||
|
@ -117,7 +117,7 @@ def convert_bigbird_pegasus(tf_weights: dict, config_update: dict) -> BigBirdPeg
|
|||
new_k = rename_state_dict_key(k, patterns)
|
||||
if new_k not in state_dict and k != "pegasus/embeddings/position_embeddings":
|
||||
raise ValueError(f"could not find new key {new_k} in state dict. (converted from {k})")
|
||||
if any([True if i in k else False for i in ["dense", "query", "key", "value"]]):
|
||||
if any(True if i in k else False for i in ["dense", "query", "key", "value"]):
|
||||
v = v.T
|
||||
mapping[new_k] = torch.from_numpy(v)
|
||||
if k != "pegasus/embeddings/position_embeddings":
|
||||
|
@ -147,7 +147,7 @@ def get_tf_weights_as_numpy(path) -> Dict:
|
|||
tf_weights = {}
|
||||
ignore_name = ["global_step"]
|
||||
for name, shape in tqdm(init_vars, desc="converting tf checkpoint to dict"):
|
||||
skip_key = any([pat in name for pat in ignore_name])
|
||||
skip_key = any(pat in name for pat in ignore_name)
|
||||
if skip_key:
|
||||
continue
|
||||
array = tf.train.load_variable(path, name)
|
||||
|
|
|
@ -2485,9 +2485,9 @@ class DetaMatcher(object):
|
|||
thresholds.insert(0, -float("inf"))
|
||||
thresholds.append(float("inf"))
|
||||
# Currently torchscript does not support all + generator
|
||||
if not all([low <= high for (low, high) in zip(thresholds[:-1], thresholds[1:])]):
|
||||
if not all(low <= high for (low, high) in zip(thresholds[:-1], thresholds[1:])):
|
||||
raise ValueError("Thresholds should be sorted.")
|
||||
if not all([l in [-1, 0, 1] for l in labels]):
|
||||
if not all(l in [-1, 0, 1] for l in labels):
|
||||
raise ValueError("All labels should be either -1, 0 or 1")
|
||||
if len(labels) != len(thresholds) - 1:
|
||||
raise ValueError("Number of labels should be equal to number of thresholds - 1")
|
||||
|
|
|
@ -379,11 +379,9 @@ class CustomDPRReaderTokenizerMixin:
|
|||
if length > max_answer_length:
|
||||
raise ValueError(f"Span is too long: {length} > {max_answer_length}")
|
||||
if any(
|
||||
[
|
||||
start_index <= prev_start_index <= prev_end_index <= end_index
|
||||
or prev_start_index <= start_index <= end_index <= prev_end_index
|
||||
for (prev_start_index, prev_end_index) in chosen_span_intervals
|
||||
]
|
||||
start_index <= prev_start_index <= prev_end_index <= end_index
|
||||
or prev_start_index <= start_index <= end_index <= prev_end_index
|
||||
for (prev_start_index, prev_end_index) in chosen_span_intervals
|
||||
):
|
||||
continue
|
||||
chosen_span_intervals.append((start_index, end_index))
|
||||
|
|
|
@ -377,11 +377,9 @@ class CustomDPRReaderTokenizerMixin:
|
|||
length = end_index - start_index + 1
|
||||
assert length <= max_answer_length, f"Span is too long: {length} > {max_answer_length}"
|
||||
if any(
|
||||
[
|
||||
start_index <= prev_start_index <= prev_end_index <= end_index
|
||||
or prev_start_index <= start_index <= end_index <= prev_end_index
|
||||
for (prev_start_index, prev_end_index) in chosen_span_intervals
|
||||
]
|
||||
start_index <= prev_start_index <= prev_end_index <= end_index
|
||||
or prev_start_index <= start_index <= end_index <= prev_end_index
|
||||
for (prev_start_index, prev_end_index) in chosen_span_intervals
|
||||
):
|
||||
continue
|
||||
chosen_span_intervals.append((start_index, end_index))
|
||||
|
|
|
@ -90,7 +90,7 @@ def get_tf_weights_as_numpy(path="./ckpt/aeslc/model.ckpt-32000") -> Dict:
|
|||
tf_weights = {}
|
||||
ignore_name = ["Adafactor", "global_step"]
|
||||
for name, shape in tqdm(init_vars, desc="converting tf checkpoint to dict"):
|
||||
skip_key = any([pat in name for pat in ignore_name])
|
||||
skip_key = any(pat in name for pat in ignore_name)
|
||||
if skip_key:
|
||||
continue
|
||||
array = tf.train.load_variable(path, name)
|
||||
|
|
|
@ -115,7 +115,7 @@ class SamProcessor(ProcessorMixin):
|
|||
for point, original_size in zip(input_points, original_sizes)
|
||||
]
|
||||
# check that all arrays have the same shape
|
||||
if not all([point.shape == input_points[0].shape for point in input_points]):
|
||||
if not all(point.shape == input_points[0].shape for point in input_points):
|
||||
if input_labels is not None:
|
||||
input_points, input_labels = self._pad_points_and_labels(input_points, input_labels)
|
||||
|
||||
|
|
|
@ -647,7 +647,7 @@ class GenerationIntegrationTestsMixin:
|
|||
generated_tokens = model.generate(**tokens, eos_token_id=eos_token_id, **generation_kwargs)
|
||||
unpadded_correct_condition = expectation == len(generated_tokens[0])
|
||||
padded_correct_condition = expectation < len(generated_tokens[0]) and all(
|
||||
[token == model.config.pad_token_id for token in generated_tokens[0][expectation:]]
|
||||
token == model.config.pad_token_id for token in generated_tokens[0][expectation:]
|
||||
)
|
||||
self.assertTrue(unpadded_correct_condition or padded_correct_condition)
|
||||
|
||||
|
@ -655,7 +655,7 @@ class GenerationIntegrationTestsMixin:
|
|||
generated_tokens = model.generate(**tokens, eos_token_id=eos_token_id, **generation_kwargs)
|
||||
unpadded_correct_condition = expectation == len(generated_tokens[0])
|
||||
padded_correct_condition = expectation < len(generated_tokens[0]) and all(
|
||||
[token == model.config.pad_token_id for token in generated_tokens[0][expectation:]]
|
||||
token == model.config.pad_token_id for token in generated_tokens[0][expectation:]
|
||||
)
|
||||
self.assertTrue(unpadded_correct_condition or padded_correct_condition)
|
||||
|
||||
|
|
|
@ -521,7 +521,7 @@ class CodeGenModelLanguageGenerationTest(unittest.TestCase):
|
|||
|
||||
self.assertEqual(output_str, EXPECTED_OUTPUT_STR)
|
||||
self.assertTrue(
|
||||
all([output_seq_strs[idx] != output_seq_tt_strs[idx] for idx in range(len(output_seq_tt_strs))])
|
||||
all(output_seq_strs[idx] != output_seq_tt_strs[idx] for idx in range(len(output_seq_tt_strs)))
|
||||
) # token_type_ids should change output
|
||||
|
||||
@is_flaky(max_attempts=3, description="measure of timing is somehow flaky.")
|
||||
|
|
|
@ -516,7 +516,7 @@ class Data2VecAudioModelTest(ModelTesterMixin, PipelineTesterMixin, unittest.Tes
|
|||
"objective.weight",
|
||||
]
|
||||
if param.requires_grad:
|
||||
if any([x in name for x in uniform_init_parms]):
|
||||
if any(x in name for x in uniform_init_parms):
|
||||
self.assertTrue(
|
||||
-1.0 <= ((param.data.mean() * 1e9).round() / 1e9).item() <= 1.0,
|
||||
msg=f"Parameter {name} of model {model_class} seems not properly initialized",
|
||||
|
|
|
@ -373,12 +373,12 @@ class EncodecModelTest(ModelTesterMixin, PipelineTesterMixin, unittest.TestCase)
|
|||
uniform_init_parms = ["conv"]
|
||||
ignore_init = ["lstm"]
|
||||
if param.requires_grad:
|
||||
if any([x in name for x in uniform_init_parms]):
|
||||
if any(x in name for x in uniform_init_parms):
|
||||
self.assertTrue(
|
||||
-1.0 <= ((param.data.mean() * 1e9).round() / 1e9).item() <= 1.0,
|
||||
msg=f"Parameter {name} of model {model_class} seems not properly initialized",
|
||||
)
|
||||
elif not any([x in name for x in ignore_init]):
|
||||
elif not any(x in name for x in ignore_init):
|
||||
self.assertIn(
|
||||
((param.data.mean() * 1e9).round() / 1e9).item(),
|
||||
[0.0, 1.0],
|
||||
|
|
|
@ -768,7 +768,7 @@ class GPT2ModelLanguageGenerationTest(unittest.TestCase):
|
|||
)
|
||||
self.assertEqual(output_str, EXPECTED_OUTPUT_STR)
|
||||
self.assertTrue(
|
||||
all([output_seq_strs[idx] != output_seq_tt_strs[idx] for idx in range(len(output_seq_tt_strs))])
|
||||
all(output_seq_strs[idx] != output_seq_tt_strs[idx] for idx in range(len(output_seq_tt_strs)))
|
||||
) # token_type_ids should change output
|
||||
|
||||
@slow
|
||||
|
|
|
@ -571,7 +571,7 @@ class GPTJModelLanguageGenerationTest(unittest.TestCase):
|
|||
|
||||
self.assertEqual(output_str, EXPECTED_OUTPUT_STR)
|
||||
self.assertTrue(
|
||||
all([output_seq_strs[idx] != output_seq_tt_strs[idx] for idx in range(len(output_seq_tt_strs))])
|
||||
all(output_seq_strs[idx] != output_seq_tt_strs[idx] for idx in range(len(output_seq_tt_strs)))
|
||||
) # token_type_ids should change output
|
||||
|
||||
@slow
|
||||
|
|
|
@ -423,7 +423,7 @@ class HubertModelTest(ModelTesterMixin, PipelineTesterMixin, unittest.TestCase):
|
|||
"quantizer.weight_proj.weight",
|
||||
]
|
||||
if param.requires_grad:
|
||||
if any([x in name for x in uniform_init_parms]):
|
||||
if any(x in name for x in uniform_init_parms):
|
||||
self.assertTrue(
|
||||
-1.0 <= ((param.data.mean() * 1e9).round() / 1e9).item() <= 1.0,
|
||||
msg=f"Parameter {name} of model {model_class} seems not properly initialized",
|
||||
|
@ -684,7 +684,7 @@ class HubertRobustModelTest(ModelTesterMixin, unittest.TestCase):
|
|||
"quantizer.weight_proj.weight",
|
||||
]
|
||||
if param.requires_grad:
|
||||
if any([x in name for x in uniform_init_parms]):
|
||||
if any(x in name for x in uniform_init_parms):
|
||||
self.assertTrue(
|
||||
-1.0 <= ((param.data.mean() * 1e9).round() / 1e9).item() <= 1.0,
|
||||
msg=f"Parameter {name} of model {model_class} seems not properly initialized",
|
||||
|
|
|
@ -386,7 +386,7 @@ class MCTCTModelTest(ModelTesterMixin, PipelineTesterMixin, unittest.TestCase):
|
|||
"objective.weight",
|
||||
]
|
||||
if param.requires_grad:
|
||||
if any([x in name for x in uniform_init_parms]):
|
||||
if any(x in name for x in uniform_init_parms):
|
||||
self.assertTrue(
|
||||
-1.0 <= ((param.data.mean() * 1e9).round() / 1e9).item() <= 1.0,
|
||||
msg=f"Parameter {name} of model {model_class} seems not properly initialized",
|
||||
|
@ -533,7 +533,7 @@ class MCTCTRobustModelTest(ModelTesterMixin, unittest.TestCase):
|
|||
"objective.weight",
|
||||
]
|
||||
if param.requires_grad:
|
||||
if any([x in name for x in uniform_init_parms]):
|
||||
if any(x in name for x in uniform_init_parms):
|
||||
self.assertTrue(
|
||||
-1.0 <= ((param.data.mean() * 1e9).round() / 1e9).item() <= 1.0,
|
||||
msg=f"Parameter {name} of model {model_class} seems not properly initialized",
|
||||
|
|
|
@ -334,7 +334,7 @@ class RwkvModelTest(ModelTesterMixin, GenerationTesterMixin, PipelineTesterMixin
|
|||
if param.requires_grad:
|
||||
# check if it's a ones like
|
||||
self.assertTrue(torch.allclose(param.data, torch.ones_like(param.data), atol=1e-5, rtol=1e-5))
|
||||
elif any([x in name for x in ["time_mix_key", "time_mix_receptance"]]):
|
||||
elif any(x in name for x in ["time_mix_key", "time_mix_receptance"]):
|
||||
if param.requires_grad:
|
||||
self.assertInterval(
|
||||
param.data,
|
||||
|
|
|
@ -417,7 +417,7 @@ class SEWModelTest(ModelTesterMixin, PipelineTesterMixin, unittest.TestCase):
|
|||
"quantizer.weight_proj.weight",
|
||||
]
|
||||
if param.requires_grad:
|
||||
if any([x in name for x in uniform_init_parms]):
|
||||
if any(x in name for x in uniform_init_parms):
|
||||
self.assertTrue(
|
||||
-1.0 <= ((param.data.mean() * 1e9).round() / 1e9).item() <= 1.0,
|
||||
msg=f"Parameter {name} of model {model_class} seems not properly initialized",
|
||||
|
|
|
@ -431,7 +431,7 @@ class SEWDModelTest(ModelTesterMixin, PipelineTesterMixin, unittest.TestCase):
|
|||
"quantizer.weight_proj.weight",
|
||||
]
|
||||
if param.requires_grad:
|
||||
if any([x in name for x in uniform_init_parms]):
|
||||
if any(x in name for x in uniform_init_parms):
|
||||
self.assertTrue(
|
||||
-1.0 <= ((param.data.mean() * 1e9).round() / 1e9).item() <= 1.0,
|
||||
msg=f"Parameter {name} of model {model_class} seems not properly initialized",
|
||||
|
|
|
@ -583,7 +583,7 @@ class SpeechT5ForSpeechToTextTest(ModelTesterMixin, unittest.TestCase):
|
|||
"feature_projection.projection.bias",
|
||||
]
|
||||
if param.requires_grad:
|
||||
if any([x in name for x in uniform_init_parms]):
|
||||
if any(x in name for x in uniform_init_parms):
|
||||
self.assertTrue(
|
||||
-1.0 <= ((param.data.mean() * 1e9).round() / 1e9).item() <= 1.0,
|
||||
msg=f"Parameter {name} of model {model_class} seems not properly initialized",
|
||||
|
@ -927,7 +927,7 @@ class SpeechT5ForTextToSpeechTest(ModelTesterMixin, unittest.TestCase):
|
|||
"conv.weight",
|
||||
]
|
||||
if param.requires_grad:
|
||||
if any([x in name for x in uniform_init_parms]):
|
||||
if any(x in name for x in uniform_init_parms):
|
||||
self.assertTrue(
|
||||
-1.0 <= ((param.data.mean() * 1e9).round() / 1e9).item() <= 1.0,
|
||||
msg=f"Parameter {name} of model {model_class} seems not properly initialized",
|
||||
|
@ -1337,7 +1337,7 @@ class SpeechT5ForSpeechToSpeechTest(ModelTesterMixin, unittest.TestCase):
|
|||
"feature_projection.projection.bias",
|
||||
]
|
||||
if param.requires_grad:
|
||||
if any([x in name for x in uniform_init_parms]):
|
||||
if any(x in name for x in uniform_init_parms):
|
||||
self.assertTrue(
|
||||
-1.0 <= ((param.data.mean() * 1e9).round() / 1e9).item() <= 1.0,
|
||||
msg=f"Parameter {name} of model {model_class} seems not properly initialized",
|
||||
|
|
|
@ -432,7 +432,7 @@ class UniSpeechRobustModelTest(ModelTesterMixin, PipelineTesterMixin, unittest.T
|
|||
"feature_projection.projection.bias",
|
||||
]
|
||||
if param.requires_grad:
|
||||
if any([x in name for x in uniform_init_parms]):
|
||||
if any(x in name for x in uniform_init_parms):
|
||||
self.assertTrue(
|
||||
-1.0 <= ((param.data.mean() * 1e9).round() / 1e9).item() <= 1.0,
|
||||
msg=f"Parameter {name} of model {model_class} seems not properly initialized",
|
||||
|
|
|
@ -484,7 +484,7 @@ class UniSpeechSatModelTest(ModelTesterMixin, PipelineTesterMixin, unittest.Test
|
|||
"objective.weight",
|
||||
]
|
||||
if param.requires_grad:
|
||||
if any([x in name for x in uniform_init_parms]):
|
||||
if any(x in name for x in uniform_init_parms):
|
||||
self.assertTrue(
|
||||
-1.0 <= ((param.data.mean() * 1e9).round() / 1e9).item() <= 1.0,
|
||||
msg=f"Parameter {name} of model {model_class} seems not properly initialized",
|
||||
|
@ -695,7 +695,7 @@ class UniSpeechSatRobustModelTest(ModelTesterMixin, unittest.TestCase):
|
|||
"objective.weight",
|
||||
]
|
||||
if param.requires_grad:
|
||||
if any([x in name for x in uniform_init_parms]):
|
||||
if any(x in name for x in uniform_init_parms):
|
||||
self.assertTrue(
|
||||
-1.0 <= ((param.data.mean() * 1e9).round() / 1e9).item() <= 1.0,
|
||||
msg=f"Parameter {name} of model {model_class} seems not properly initialized",
|
||||
|
|
|
@ -464,7 +464,7 @@ class FlaxWav2Vec2UtilsTest(unittest.TestCase):
|
|||
negative_indices = _sample_negative_indices(features.shape, num_negatives, attention_mask=attention_mask)
|
||||
|
||||
# make sure that no padding tokens are sampled
|
||||
self.assertTrue(all([idx not in negative_indices for idx in forbidden_indices]))
|
||||
self.assertTrue(all(idx not in negative_indices for idx in forbidden_indices))
|
||||
|
||||
features = features.reshape(-1, hidden_size) # BTC => (BxT)C
|
||||
# take negative vectors from sampled indices
|
||||
|
|
|
@ -637,7 +637,7 @@ class Wav2Vec2ModelTest(ModelTesterMixin, PipelineTesterMixin, unittest.TestCase
|
|||
"objective.weight",
|
||||
]
|
||||
if param.requires_grad:
|
||||
if any([x in name for x in uniform_init_parms]):
|
||||
if any(x in name for x in uniform_init_parms):
|
||||
self.assertTrue(
|
||||
-1.0 <= ((param.data.mean() * 1e9).round() / 1e9).item() <= 1.0,
|
||||
msg=f"Parameter {name} of model {model_class} seems not properly initialized",
|
||||
|
@ -971,7 +971,7 @@ class Wav2Vec2RobustModelTest(ModelTesterMixin, unittest.TestCase):
|
|||
"objective.weight",
|
||||
]
|
||||
if param.requires_grad:
|
||||
if any([x in name for x in uniform_init_parms]):
|
||||
if any(x in name for x in uniform_init_parms):
|
||||
self.assertTrue(
|
||||
-1.0 <= ((param.data.mean() * 1e9).round() / 1e9).item() <= 1.0,
|
||||
msg=f"Parameter {name} of model {model_class} seems not properly initialized",
|
||||
|
|
|
@ -569,7 +569,7 @@ class Wav2Vec2ConformerModelTest(ModelTesterMixin, PipelineTesterMixin, unittest
|
|||
"objective.weight",
|
||||
]
|
||||
if param.requires_grad:
|
||||
if any([x in name for x in uniform_init_parms]):
|
||||
if any(x in name for x in uniform_init_parms):
|
||||
self.assertTrue(
|
||||
-1.0 <= ((param.data.mean() * 1e9).round() / 1e9).item() <= 1.0,
|
||||
msg=f"Parameter {name} of model {model_class} seems not properly initialized",
|
||||
|
|
|
@ -438,7 +438,7 @@ class WavLMModelTest(ModelTesterMixin, PipelineTesterMixin, unittest.TestCase):
|
|||
"objective.weight",
|
||||
]
|
||||
if param.requires_grad:
|
||||
if any([x in name for x in uniform_init_parms]):
|
||||
if any(x in name for x in uniform_init_parms):
|
||||
self.assertTrue(
|
||||
-1.0 <= ((param.data.mean() * 1e9).round() / 1e9).item() <= 1.0,
|
||||
msg=f"Parameter {name} of model {model_class} seems not properly initialized",
|
||||
|
|
|
@ -1535,7 +1535,7 @@ class WhisperModelIntegrationTests(unittest.TestCase):
|
|||
text = processor.decode(output[0])
|
||||
|
||||
self.assertTrue(prompt in text)
|
||||
self.assertTrue(all([token in text for token in expected_tokens]))
|
||||
self.assertTrue(all(token in text for token in expected_tokens))
|
||||
|
||||
@slow
|
||||
def test_generate_with_prompt_ids_and_no_non_prompt_forced_decoder_ids(self):
|
||||
|
|
|
@ -145,7 +145,7 @@ class OnnxExportTestCase(unittest.TestCase):
|
|||
|
||||
# Assert all variables are present
|
||||
self.assertEqual(len(shapes), len(variable_names))
|
||||
self.assertTrue(all([var_name in shapes for var_name in variable_names]))
|
||||
self.assertTrue(all(var_name in shapes for var_name in variable_names))
|
||||
self.assertSequenceEqual(variable_names[:3], input_vars)
|
||||
self.assertSequenceEqual(variable_names[3:], output_vars)
|
||||
|
||||
|
|
|
@ -1566,7 +1566,7 @@ class TFModelTesterMixin:
|
|||
return_labels=True if "labels" in inspect.signature(model_class.call).parameters.keys() else False,
|
||||
)
|
||||
if not any(
|
||||
[tensor.dtype.is_integer for tensor in prepared_for_class.values() if isinstance(tensor, tf.Tensor)]
|
||||
tensor.dtype.is_integer for tensor in prepared_for_class.values() if isinstance(tensor, tf.Tensor)
|
||||
):
|
||||
return # No integer inputs means no need for this test
|
||||
|
||||
|
|
|
@ -79,7 +79,7 @@ SMALL_TRAINING_CORPUS = [
|
|||
|
||||
def filter_non_english(_, pretrained_name: str):
|
||||
"""Filter all the model for non-english language"""
|
||||
return not any([lang in pretrained_name for lang in NON_ENGLISH_TAGS])
|
||||
return not any(lang in pretrained_name for lang in NON_ENGLISH_TAGS)
|
||||
|
||||
|
||||
def filter_roberta_detectors(_, pretrained_name: str):
|
||||
|
|
|
@ -56,8 +56,8 @@ class Seq2seqTrainerTester(TestCasePlus):
|
|||
]
|
||||
batch["decoder_attention_mask"] = outputs.attention_mask
|
||||
|
||||
assert all([len(x) == 512 for x in inputs.input_ids])
|
||||
assert all([len(x) == 128 for x in outputs.input_ids])
|
||||
assert all(len(x) == 512 for x in inputs.input_ids)
|
||||
assert all(len(x) == 128 for x in outputs.input_ids)
|
||||
|
||||
return batch
|
||||
|
||||
|
|
|
@ -362,7 +362,7 @@ def convert_to_localized_md(model_list, localized_model_list, format_str):
|
|||
model_keys = [re.search(r"\*\*\[([^\]]*)", line).groups()[0] for line in model_list.strip().split("\n")]
|
||||
|
||||
# We exclude keys in localized README not in the main one.
|
||||
readmes_match = not any([k not in model_keys for k in localized_model_index])
|
||||
readmes_match = not any(k not in model_keys for k in localized_model_index)
|
||||
localized_model_index = {k: v for k, v in localized_model_index.items() if k in model_keys}
|
||||
|
||||
for model in model_list.strip().split("\n"):
|
||||
|
|
|
@ -735,7 +735,7 @@ def build_model(model_arch, tiny_config, output_dir):
|
|||
|
||||
tiny_config = copy.deepcopy(tiny_config)
|
||||
|
||||
if any([model_arch.__name__.endswith(x) for x in ["ForCausalLM", "LMHeadModel"]]):
|
||||
if any(model_arch.__name__.endswith(x) for x in ["ForCausalLM", "LMHeadModel"]):
|
||||
tiny_config.is_encoder_decoder = False
|
||||
tiny_config.is_decoder = True
|
||||
|
||||
|
|
|
@ -428,7 +428,7 @@ def get_module_dependencies(module_fname, cache=None):
|
|||
# So we get the imports from that init then try to find where our objects come from.
|
||||
new_imported_modules = extract_imports(module, cache=cache)
|
||||
for new_module, new_imports in new_imported_modules:
|
||||
if any([i in new_imports for i in imports]):
|
||||
if any(i in new_imports for i in imports):
|
||||
if new_module not in dependencies:
|
||||
new_modules.append((new_module, [i for i in new_imports if i in imports]))
|
||||
imports = [i for i in imports if i not in new_imports]
|
||||
|
|
Loading…
Reference in New Issue