Spaces:
Runtime error
Runtime error
File size: 7,178 Bytes
455a40f |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 |
# coding=utf-8
# Copyright 2021 The HuggingFace Inc. team. All rights reserved.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
""" Testing suite for the PyTorch TrOCR model. """
import unittest
from transformers import TrOCRConfig
from transformers.testing_utils import is_torch_available, require_torch, torch_device
from ...generation.test_utils import GenerationTesterMixin
from ...test_configuration_common import ConfigTester
from ...test_modeling_common import ModelTesterMixin, ids_tensor
from ...test_pipeline_mixin import PipelineTesterMixin
if is_torch_available():
import torch
from transformers.models.trocr.modeling_trocr import TrOCRDecoder, TrOCRForCausalLM
@require_torch
class TrOCRStandaloneDecoderModelTester:
def __init__(
self,
parent,
vocab_size=99,
batch_size=13,
d_model=16,
decoder_seq_length=7,
is_training=True,
is_decoder=True,
use_attention_mask=True,
use_cache=False,
use_labels=True,
decoder_start_token_id=2,
decoder_ffn_dim=32,
decoder_layers=4,
decoder_attention_heads=4,
max_position_embeddings=30,
pad_token_id=0,
bos_token_id=1,
eos_token_id=2,
scope=None,
):
self.parent = parent
self.batch_size = batch_size
self.decoder_seq_length = decoder_seq_length
# For common tests
self.seq_length = self.decoder_seq_length
self.is_training = is_training
self.use_attention_mask = use_attention_mask
self.use_labels = use_labels
self.vocab_size = vocab_size
self.d_model = d_model
self.hidden_size = d_model
self.num_hidden_layers = decoder_layers
self.decoder_layers = decoder_layers
self.decoder_ffn_dim = decoder_ffn_dim
self.decoder_attention_heads = decoder_attention_heads
self.num_attention_heads = decoder_attention_heads
self.eos_token_id = eos_token_id
self.bos_token_id = bos_token_id
self.pad_token_id = pad_token_id
self.decoder_start_token_id = decoder_start_token_id
self.use_cache = use_cache
self.max_position_embeddings = max_position_embeddings
self.scope = None
self.decoder_key_length = decoder_seq_length
self.base_model_out_len = 2
self.decoder_attention_idx = 1
def prepare_config_and_inputs(self):
input_ids = ids_tensor([self.batch_size, self.decoder_seq_length], self.vocab_size)
attention_mask = None
if self.use_attention_mask:
attention_mask = ids_tensor([self.batch_size, self.decoder_seq_length], vocab_size=2)
lm_labels = None
if self.use_labels:
lm_labels = ids_tensor([self.batch_size, self.decoder_seq_length], self.vocab_size)
config = TrOCRConfig(
vocab_size=self.vocab_size,
d_model=self.d_model,
decoder_layers=self.decoder_layers,
decoder_ffn_dim=self.decoder_ffn_dim,
decoder_attention_heads=self.decoder_attention_heads,
eos_token_id=self.eos_token_id,
bos_token_id=self.bos_token_id,
use_cache=self.use_cache,
pad_token_id=self.pad_token_id,
decoder_start_token_id=self.decoder_start_token_id,
max_position_embeddings=self.max_position_embeddings,
)
return (config, input_ids, attention_mask, lm_labels)
def create_and_check_decoder_model_past(
self,
config,
input_ids,
attention_mask,
lm_labels,
):
config.use_cache = True
model = TrOCRDecoder(config=config).to(torch_device).eval()
input_ids = input_ids[:2]
input_ids[input_ids == 0] += 1
# first forward pass
outputs = model(input_ids, use_cache=True)
outputs_use_cache_conf = model(input_ids)
outputs_no_past = model(input_ids, use_cache=False)
self.parent.assertTrue(len(outputs) == len(outputs_use_cache_conf))
self.parent.assertTrue(len(outputs) == len(outputs_no_past) + 1)
past_key_values = outputs["past_key_values"]
# create hypothetical next token and extent to next_input_ids
next_tokens = ids_tensor((2, 1), config.vocab_size - 1) + 1
# append to next input_ids and
next_input_ids = torch.cat([input_ids, next_tokens], dim=-1)
output_from_no_past = model(next_input_ids)["last_hidden_state"]
output_from_past = model(next_tokens, past_key_values=past_key_values)["last_hidden_state"]
# select random slice
random_slice_idx = ids_tensor((1,), output_from_past.shape[-1]).item()
output_from_no_past_slice = output_from_no_past[:, next_input_ids.shape[-1] - 1, random_slice_idx].detach()
output_from_past_slice = output_from_past[:, 0, random_slice_idx].detach()
# test that outputs are equal for slice
assert torch.allclose(output_from_past_slice, output_from_no_past_slice, atol=1e-3)
def prepare_config_and_inputs_for_common(self):
config_and_inputs = self.prepare_config_and_inputs()
config, input_ids, attention_mask, lm_labels = config_and_inputs
inputs_dict = {"input_ids": input_ids, "attention_mask": attention_mask}
return config, inputs_dict
@require_torch
class TrOCRStandaloneDecoderModelTest(ModelTesterMixin, GenerationTesterMixin, PipelineTesterMixin, unittest.TestCase):
all_model_classes = (TrOCRDecoder, TrOCRForCausalLM) if is_torch_available() else ()
all_generative_model_classes = (TrOCRForCausalLM,) if is_torch_available() else ()
pipeline_model_mapping = {"text-generation": TrOCRForCausalLM} if is_torch_available() else {}
fx_compatible = True
test_pruning = False
def setUp(self):
self.model_tester = TrOCRStandaloneDecoderModelTester(self, is_training=False)
self.config_tester = ConfigTester(self, config_class=TrOCRConfig)
# not implemented currently
def test_inputs_embeds(self):
pass
# trocr has no base model
def test_save_load_fast_init_from_base(self):
pass
# trocr has no base model
def test_save_load_fast_init_to_base(self):
pass
def test_config(self):
self.config_tester.run_common_tests()
def test_decoder_model_past(self):
config_and_inputs = self.model_tester.prepare_config_and_inputs()
self.model_tester.create_and_check_decoder_model_past(*config_and_inputs)
# decoder cannot keep gradients
def test_retain_grad_hidden_states_attentions(self):
return
|