# Copyright (c) 2022 PaddlePaddle Authors. All Rights Reserved. # Copyright 2020 The HuggingFace 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. from __future__ import annotations import math import random import unittest import numpy as np import paddle from parameterized import parameterized_class from paddlenlp.transformers import GPTTokenizer, OPTConfig, OPTForCausalLM, OPTModel from tests.testing_utils import PaddleNLPModelTest, require_package, slow from tests.transformers.test_generation_utils import GenerationTesterMixin from tests.transformers.test_modeling_common import ( # GenerationD2STestMixin, ModelTesterMixin, floats_tensor, ids_tensor, ) OPT_PRETRAINED_MODEL_ARCHIVE_LIST = [ "facebook/opt-125m", ] class OPTModelTester: def __init__( self, parent, batch_size=14, seq_length=7, is_training=True, use_input_mask=True, vocab_size=99, hidden_size=32, num_hidden_layers=5, num_attention_heads=4, intermediate_size=37, hidden_act="relu", hidden_dropout_prob=0.1, attention_probs_dropout_prob=0.1, max_position_embeddings=512, type_vocab_size=16, type_sequence_label_size=2, initializer_range=0.02, normalize_before=True, word_embed_proj_dim=32, num_labels=3, num_choices=4, scope=None, ): self.parent = parent self.batch_size = batch_size self.seq_length = seq_length self.is_training = is_training self.use_input_mask = use_input_mask self.vocab_size = vocab_size self.hidden_size = hidden_size self.num_hidden_layers = num_hidden_layers self.num_attention_heads = num_attention_heads self.intermediate_size = intermediate_size self.hidden_act = hidden_act self.hidden_dropout_prob = hidden_dropout_prob self.attention_probs_dropout_prob = attention_probs_dropout_prob self.max_position_embeddings = max_position_embeddings self.type_vocab_size = type_vocab_size self.type_sequence_label_size = type_sequence_label_size self.initializer_range = initializer_range self.num_labels = num_labels self.num_choices = num_choices self.scope = None self.normalize_before = normalize_before self.word_embed_proj_dim = word_embed_proj_dim self.bos_token_id = vocab_size - 1 self.eos_token_id = vocab_size - 1 self.pad_token_id = vocab_size - 1 def prepare_config_and_inputs(self): input_ids = ids_tensor([self.batch_size, self.seq_length], self.vocab_size, dtype="int64") input_mask = None if self.use_input_mask: # construct input_mask filling with 0 and -1e4 # left padding: [[-1e4, -1e4, -1e4, 0, 0], [-1e4, -1e4, -1e4, 0, 0], ...] input_mask = [] for _ in range(self.batch_size): pad_length = random.randint(0, self.seq_length) input_mask.append([0] * (self.seq_length - pad_length) + [1] * pad_length) input_mask = paddle.to_tensor(input_mask, dtype="int64") sequence_labels = None token_labels = None choice_labels = None if self.parent.use_labels: sequence_labels = ids_tensor([self.batch_size], self.type_sequence_label_size, dtype="int64") token_labels = ids_tensor([self.batch_size, self.seq_length], self.num_labels, dtype="int64") choice_labels = ids_tensor([self.batch_size], self.num_choices, dtype="int64") config = self.get_config() return ( config, input_ids, input_mask, sequence_labels, token_labels, choice_labels, ) def get_config(self): return OPTConfig( vocab_size=self.vocab_size, hidden_size=self.hidden_size, num_hidden_layers=self.num_hidden_layers, num_attention_heads=self.num_attention_heads, intermediate_size=self.intermediate_size, hidden_act=self.hidden_act, hidden_dropout_prob=self.hidden_dropout_prob, attention_probs_dropout_prob=self.attention_probs_dropout_prob, max_position_embeddings=self.max_position_embeddings, type_vocab_size=self.type_vocab_size, initializer_range=self.initializer_range, bos_token_id=self.bos_token_id, eos_token_id=self.eos_token_id, pad_token_id=self.pad_token_id, normalize_before=self.normalize_before, word_embed_proj_dim=self.word_embed_proj_dim, ) def prepare_config_and_inputs_for_decoder(self): ( config, input_ids, input_mask, sequence_labels, token_labels, choice_labels, ) = self.prepare_config_and_inputs() encoder_hidden_states = floats_tensor([self.batch_size, self.seq_length, self.hidden_size]) encoder_attention_mask = paddle.cast( ids_tensor([self.batch_size, self.seq_length], vocab_size=2), dtype="int64" ) return ( config, input_ids, input_mask, sequence_labels, token_labels, choice_labels, encoder_hidden_states, encoder_attention_mask, ) def create_and_check_opt_model(self, config, input_ids, input_mask, *args): model = OPTModel(config) model.eval() result = model(input_ids, use_cache=True, return_dict=self.parent.return_dict) result = model(input_ids, use_cache=True, return_dict=self.parent.return_dict) result = model(input_ids, use_cache=True, return_dict=self.parent.return_dict) self.parent.assertEqual(result[0].shape, [self.batch_size, self.seq_length, self.hidden_size]) self.parent.assertEqual(len(result[1]), config.num_hidden_layers) def create_and_check_opt_model_past(self, config, input_ids, input_mask, *args): model = OPTModel(config) model.eval() # first forward pass outputs = model(input_ids, use_cache=True, return_dict=self.parent.return_dict) outputs_use_cache_conf = model(input_ids, use_cache=True, return_dict=self.parent.return_dict) model(input_ids, use_cache=False, return_dict=self.parent.return_dict) self.parent.assertTrue(len(outputs) == len(outputs_use_cache_conf)) output, past = outputs[:2] # create hypothetical next token and extent to next_input_ids next_tokens = ids_tensor((self.batch_size, 1), config.vocab_size, dtype="int64") # append to next input_ids next_input_ids = paddle.concat([input_ids, next_tokens], axis=-1) output_from_no_past = model(next_input_ids, return_dict=self.parent.return_dict) if self.parent.return_dict: output_from_no_past = output_from_no_past[0] past_key_values_length = paddle.shape(past[0].k)[2] attention_mask = paddle.ones(shape=[next_tokens.shape[0], 1 + past_key_values_length]) output_from_past = model( next_tokens, use_cache=True, attention_mask=attention_mask, cache=past, return_dict=self.parent.return_dict )[0] # select random slice random_slice_idx = ids_tensor((1,), output_from_past.shape[-1]).item() output_from_no_past_slice = output_from_no_past[:, -1, random_slice_idx].detach() output_from_past_slice = output_from_past[:, 0, random_slice_idx].detach() # test that outputs are equal for slice self.parent.assertTrue(paddle.allclose(output_from_past_slice, output_from_no_past_slice, atol=1e-3)) def create_and_check_opt_model_past_large_inputs(self, config, input_ids, input_mask, *args): model = OPTModel(config) model.eval() # first forward pass outputs = model(input_ids, attention_mask=input_mask, use_cache=True, return_dict=self.parent.return_dict) output, past = outputs[:2] # create hypothetical next token and extent to next_input_ids next_tokens = ids_tensor((self.batch_size, 3), config.vocab_size, dtype="int64") next_mask = paddle.ones_like(next_tokens, dtype=paddle.int64) # append to next input_ids next_input_ids = paddle.concat([input_ids, next_tokens], axis=-1) next_attention_mask = paddle.concat([input_mask, next_mask], axis=-1) output_from_no_past = model( next_input_ids, attention_mask=next_attention_mask, return_dict=self.parent.return_dict ) if self.parent.return_dict: output_from_no_past = output_from_no_past[0] output_from_past = model( next_tokens, attention_mask=next_attention_mask, cache=past, use_cache=True, return_dict=self.parent.return_dict, )[0] self.parent.assertTrue(output_from_past.shape[1] == next_tokens.shape[1]) # select random slice random_slice_idx = ids_tensor((1,), output_from_past.shape[-1], dtype="int64").item() output_from_no_past_slice = output_from_no_past[:, -3:, random_slice_idx].detach() output_from_past_slice = output_from_past[:, :, random_slice_idx].detach() # test that outputs are equal for slice self.parent.assertTrue(paddle.allclose(output_from_past_slice, output_from_no_past_slice, atol=1e-3)) def create_and_check_opt_for_causal_lm(self, config, input_ids, input_mask, *args): model = OPTForCausalLM(config) model.eval() result = model( input_ids, use_cache=True, labels=input_ids if self.parent.use_labels else None, return_dict=self.parent.return_dict, ) if self.parent.use_labels: self.parent.assertIsInstance(result[0].item(), float) self.parent.assertEqual(result[1].shape, [self.batch_size, self.seq_length, self.vocab_size]) else: self.parent.assertEqual(result[0].shape, [self.batch_size, self.seq_length, self.vocab_size]) def create_and_check_forward_and_backwards(self, config, input_ids, input_mask, *args): model = OPTForCausalLM(config) if self.parent.use_labels: loss, logits = model(input_ids, labels=input_ids, return_dict=self.parent.return_dict) self.parent.assertEqual(loss.shape, [1]) self.parent.assertEqual(logits.shape, [self.batch_size, self.seq_length, self.vocab_size]) loss.backward() def create_and_check_opt_weight_initialization(self, config, *args): model = OPTModel(config) model_std = model.config.initializer_range / math.sqrt(2 * model.config.num_hidden_layers) for key in model.state_dict().keys(): if "out_proj" in key and "weight" in key: self.parent.assertLessEqual(abs((paddle.std(model.state_dict()[key]) - model_std).numpy()), 0.02) self.parent.assertLessEqual(abs((paddle.mean(model.state_dict()[key]) - 0.0).numpy()), 0.01) def prepare_config_and_inputs_for_common(self): config_and_inputs = self.prepare_config_and_inputs() ( config, input_ids, input_mask, sequence_labels, token_labels, choice_labels, ) = config_and_inputs inputs_dict = { "input_ids": input_ids, } return config, inputs_dict def create_and_check_model_cache(self, config, input_ids, input_mask, *args): model = OPTModel(config) model.eval() # first forward pass outputs = model(input_ids, attention_mask=input_mask, use_cache=True, return_dict=self.parent.return_dict) past_key_values = outputs.past_key_values if self.parent.return_dict else outputs[1] # create hypothetical multiple next token and extent to next_input_ids next_tokens = ids_tensor((self.batch_size, 3), self.vocab_size, dtype="int64") # all next mask is ones next_mask = paddle.ones_like(next_tokens, dtype="int64") # append to next input_ids and next_input_ids = paddle.concat([input_ids, next_tokens], axis=-1) next_attention_mask = paddle.concat([input_mask, next_mask], axis=-1) outputs = model( next_input_ids, attention_mask=next_attention_mask, output_hidden_states=True, return_dict=self.parent.return_dict, ) output_from_no_past = outputs[1][0] outputs = model( next_tokens, attention_mask=next_attention_mask, cache=past_key_values, output_hidden_states=True, return_dict=self.parent.return_dict, ) output_from_past = outputs[1][0] # select random slice random_slice_idx = ids_tensor((1,), output_from_past.shape[-1]).item() output_from_no_past_slice = output_from_no_past[:, -3:, random_slice_idx].detach() output_from_past_slice = output_from_past[:, :, random_slice_idx].detach() self.parent.assertTrue(output_from_past_slice.shape[1] == next_tokens.shape[1]) # test that outputs are equal for slice self.parent.assertTrue(paddle.allclose(output_from_past_slice, output_from_no_past_slice, atol=1e-3)) @parameterized_class( ("return_dict", "use_labels"), [ [False, False], [False, True], [True, False], [True, True], ], ) class OPTModelTest(ModelTesterMixin, GenerationTesterMixin, PaddleNLPModelTest): base_model_class = OPTModel use_labels = False return_dict = False use_test_inputs_embeds = True all_model_classes = [ OPTModel, ] all_generative_model_classes = {OPTForCausalLM: (OPTModel, "opt")} test_missing_keys = False test_model_parallel = True # special case for DoubleHeads model def _prepare_for_class(self, inputs_dict, model_class): inputs_dict = super()._prepare_for_class(inputs_dict, model_class) return inputs_dict def setUp(self): self.model_tester = OPTModelTester(self) random.seed(128) np.random.seed(128) paddle.seed(128) def test_opt_model(self): config_and_inputs = self.model_tester.prepare_config_and_inputs() self.model_tester.create_and_check_opt_model(*config_and_inputs) def test_opt_model_past(self): config_and_inputs = self.model_tester.prepare_config_and_inputs() self.model_tester.create_and_check_opt_model_past(*config_and_inputs) def test_opt_model_past_large_inputs(self): config_and_inputs = self.model_tester.prepare_config_and_inputs() self.model_tester.create_and_check_opt_model_past_large_inputs(*config_and_inputs) def test_opt_causal_lm_model(self): config_and_inputs = self.model_tester.prepare_config_and_inputs() self.model_tester.create_and_check_opt_for_causal_lm(*config_and_inputs) def test_opt_weight_initialization(self): config_and_inputs = self.model_tester.prepare_config_and_inputs() self.model_tester.create_and_check_opt_weight_initialization(*config_and_inputs) def test_for_model_cache(self): config_and_inputs = self.model_tester.prepare_config_and_inputs() self.model_tester.create_and_check_model_cache(*config_and_inputs) @slow def test_batch_generation(self): model = OPTForCausalLM.from_pretrained("facebook/opt-1.3b") model.eval() tokenizer = GPTTokenizer.from_pretrained("facebook/opt-1.3b") tokenizer.padding_side = "left" # use different length sentences to test batching sentences = [ "my dog is", "Today, I", ] inputs = tokenizer(sentences, return_tensors="pd", padding=True) input_ids = inputs["input_ids"] outputs, _ = model.generate( input_ids=input_ids, decode_strategy="greedy_search", use_cache=True, ) batch_out_sentence = tokenizer.batch_decode(outputs, skip_special_tokens=True) inputs_non_padded = tokenizer(sentences[0], return_tensors="pd")["input_ids"] output_non_padded, _ = model.generate( input_ids=inputs_non_padded, use_cache=True, decode_strategy="greedy_search" ) non_padded_sentence = tokenizer.decode(output_non_padded[0], skip_special_tokens=True) inputs_padded = tokenizer(sentences[1], return_tensors="pd")["input_ids"] output_padded, _ = model.generate(input_ids=inputs_padded, use_cache=True, decode_strategy="greedy_search") padded_sentence = tokenizer.decode(output_padded[0], skip_special_tokens=True) expected_output_sentence = [ " a rescue and she's the best dog ever. she's a little bitch but she's the best", " am going to share with you a few of my favorite recipes.\nI have been cooking for a", ] self.assertListEqual(expected_output_sentence, batch_out_sentence) self.assertListEqual(expected_output_sentence, [non_padded_sentence, padded_sentence]) def _get_input_ids_and_config(self): config, inputs_dict = self.model_tester.prepare_config_and_inputs_for_common() input_ids = inputs_dict[self.input_name] max_batch_size = 2 sequence_length = input_ids.shape[-1] // 2 input_ids = input_ids[:max_batch_size, :sequence_length] attention_mask = paddle.ones_like(input_ids, dtype=paddle.int64) # generate max 3 tokens max_length = 3 if config.eos_token_id or config.pad_token_id: config["pad_token_id"] = config["eos_token_id"] return config, input_ids, attention_mask, max_length @slow def test_model_from_pretrained(self): for model_name in OPT_PRETRAINED_MODEL_ARCHIVE_LIST[:1]: model = OPTModel.from_pretrained(model_name) self.assertIsNotNone(model) class OPTCompatibilityTest(unittest.TestCase): test_model_id = "hf-internal-testing/tiny-random-OPTModel" @require_package("transformers", "torch") def test_model_config_mapping(self): # 1. create common input input_ids = np.random.randint(100, 200, [1, 20]) # 2. forward the torch model import torch import transformers torch_model_class = getattr(transformers, "OPTModel") torch_model = torch_model_class.from_pretrained(OPTCompatibilityTest.test_model_id) torch_model.eval() torch_logit = torch_model(torch.tensor(input_ids), return_dict=False)[0] # 3. forward the paddle model from paddlenlp import transformers paddle_model_class = getattr(transformers, "OPTModel") paddle_model = paddle_model_class.from_pretrained(OPTCompatibilityTest.test_model_id, from_hf_hub=True) paddle_model.eval() paddle_logit = paddle_model(paddle.to_tensor(input_ids), return_dict=False)[0] self.assertTrue( np.allclose( paddle_logit.detach().cpu().reshape([-1])[:9].numpy(), torch_logit.detach().cpu().reshape([-1])[:9].numpy(), atol=1e-4, ) ) class OPTModelIntegrationTest(unittest.TestCase): @slow def test_inference_no_attention(self): model = OPTModel.from_pretrained("facebook/opt-1.3b") model.eval() input_ids = paddle.to_tensor([[0, 345, 232, 328, 740, 140, 1695, 69, 6078, 1588, 2]]) with paddle.no_grad(): output = model(input_ids) expected_shape = [1, 11, 2048] self.assertEqual(output.shape, expected_shape) expected_slice = paddle.to_tensor( [ [ [0.81907797, -1.08688772, 1.26071370], [0.96454084, -0.42267877, 1.70609033], [0.78616256, -0.27438506, 0.74083930], ] ] ) self.assertTrue(paddle.allclose(output[:, 1:4, 1:4], expected_slice, atol=1e-4)) @slow def test_inference_with_attention(self): model = OPTModel.from_pretrained("facebook/opt-1.3b") model.eval() input_ids = paddle.to_tensor([[0, 345, 232, 328, 740, 140, 1695, 69, 6078, 1588, 2]]) attention_mask = paddle.to_tensor([[0, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1]]) with paddle.no_grad(): output = model(input_ids, attention_mask=attention_mask) expected_shape = [1, 11, 2048] self.assertEqual(output.shape, expected_shape) expected_slice = paddle.to_tensor( [ [ [0.15988758, -0.21016182, -0.28532112], [-0.18293847, -0.35511413, 0.56858277], [0.39969346, -0.33906624, -0.43125907], ] ] ) self.assertTrue(paddle.allclose(output[:, 1:4, 1:4], expected_slice, atol=1e-4)) # class OPTGenerationD2STest(GenerationD2STestMixin, unittest.TestCase): # internal_testing_model = "__internal_testing__/tiny-random-opt" # TokenizerClass = GPTTokenizer