wehub-resource-sync
491 lines
20 KiB
Python
491 lines
20 KiB
Python
# Copyright (c) 2023 PaddlePaddle Authors. 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 unittest
|
|
|
|
import numpy as np
|
|
import paddle
|
|
|
|
from paddlenlp.transformers import ChatGLMConfig, ChatGLMForCausalLM, ChatGLMModel
|
|
|
|
# from paddlenlp.utils.env import (
|
|
# PADDLE_INFERENCE_MODEL_SUFFIX,
|
|
# PADDLE_INFERENCE_WEIGHTS_SUFFIX,
|
|
# )
|
|
from tests.transformers.test_configuration_common import ConfigTester
|
|
from tests.transformers.test_generation_utils import GenerationTesterMixin
|
|
from tests.transformers.test_modeling_common import (
|
|
ModelTesterMixin,
|
|
ids_tensor,
|
|
random_attention_mask,
|
|
)
|
|
|
|
|
|
class ChatGLMTester:
|
|
def __init__(
|
|
self,
|
|
parent,
|
|
vocab_size=130528,
|
|
hidden_size=64,
|
|
num_hidden_layers=2,
|
|
num_attention_heads=8,
|
|
layernorm_epsilon=1e-5,
|
|
use_cache=False,
|
|
bos_token_id=130004,
|
|
eos_token_id=130005,
|
|
pad_token_id=3,
|
|
mask_token_id=130000,
|
|
gmask_token_id=130001,
|
|
max_sequence_length=10,
|
|
inner_hidden_size=256,
|
|
position_encoding_2d=True,
|
|
quantization_bit=0,
|
|
pre_seq_len=None,
|
|
prefix_projection=False,
|
|
output_predict=True,
|
|
recompute=False,
|
|
attention_scale=True,
|
|
activation="gelu",
|
|
batch_size: int = 2,
|
|
seq_length: int = 10,
|
|
num_image_tokens=0,
|
|
use_labels: bool = False,
|
|
return_dict=False,
|
|
):
|
|
self.parent: ChatGLMTest = parent
|
|
self.num_hidden_layers = num_hidden_layers
|
|
self.vocab_size = vocab_size
|
|
self.hidden_size = hidden_size
|
|
self.num_attention_heads = num_attention_heads
|
|
self.max_sequence_length = max_sequence_length
|
|
self.layernorm_epsilon = layernorm_epsilon
|
|
self.inner_hidden_size = inner_hidden_size
|
|
self.use_cache = use_cache
|
|
self.bos_token_id = bos_token_id
|
|
self.eos_token_id = eos_token_id
|
|
self.pad_token_id = pad_token_id
|
|
self.mask_token_id = mask_token_id
|
|
self.gmask_token_id = gmask_token_id
|
|
self.position_encoding_2d = position_encoding_2d
|
|
self.quantization_bit = quantization_bit
|
|
self.pre_seq_len = pre_seq_len
|
|
self.prefix_projection = prefix_projection
|
|
self.output_predict = output_predict
|
|
self.recompute = recompute
|
|
self.attention_scale = attention_scale
|
|
self.activation = activation
|
|
self.num_image_tokens = num_image_tokens
|
|
|
|
self.batch_size = batch_size
|
|
self.seq_length = seq_length
|
|
self.use_labels = use_labels
|
|
self.return_dict = return_dict
|
|
|
|
def prepare_config_and_inputs(self):
|
|
input_ids = ids_tensor([self.batch_size, self.seq_length], self.vocab_size, dtype="int64")
|
|
input_ids[input_ids == self.gmask_token_id] = self.mask_token_id
|
|
input_ids[input_ids == self.bos_token_id] = self.mask_token_id
|
|
|
|
context_length = np.random.randint(1, self.seq_length - 2)
|
|
input_ids[:, context_length - 2] = self.gmask_token_id
|
|
input_ids[:, context_length - 1] = self.bos_token_id
|
|
|
|
attention_mask = paddle.ones_like(input_ids, dtype=paddle.int64)
|
|
attention_mask = attention_mask.unsqueeze([1, 2])
|
|
attention_mask = attention_mask * attention_mask.transpose([0, 1, 3, 2])
|
|
|
|
MASK, gMASK = self.mask_token_id, self.gmask_token_id
|
|
use_gmasks = []
|
|
mask_positions = []
|
|
context_lengths = []
|
|
for seq in input_ids:
|
|
mask_token = gMASK if gMASK in seq else MASK
|
|
use_gmask = mask_token == gMASK
|
|
use_gmasks.append(use_gmask)
|
|
mask_positions.append(paddle.where(seq == mask_token)[0][0])
|
|
context_lengths.append(context_length)
|
|
|
|
position_ids = paddle.arange(self.seq_length, dtype="int64").unsqueeze(0).tile([self.batch_size, 1])
|
|
for i, context_length in enumerate(context_lengths):
|
|
position_ids[i, context_length:] = mask_positions[i]
|
|
|
|
block_position_ids = [
|
|
paddle.concat(
|
|
(
|
|
paddle.zeros([context_length], dtype="int64"),
|
|
paddle.arange(self.seq_length - context_length, dtype="int64") + 1,
|
|
)
|
|
)
|
|
for context_length in context_lengths
|
|
]
|
|
block_position_ids = paddle.stack(block_position_ids, axis=0)
|
|
position_ids = paddle.stack((position_ids, block_position_ids), axis=1)
|
|
|
|
labels = None
|
|
if self.use_labels:
|
|
labels = paddle.ones([self.batch_size, self.seq_length]) * -100
|
|
labels[:, context_length:] = input_ids[:, context_length:]
|
|
|
|
config = self.get_config()
|
|
return config, input_ids, labels, attention_mask, position_ids
|
|
|
|
def get_config(self):
|
|
return ChatGLMConfig(
|
|
num_hidden_layers=self.num_hidden_layers,
|
|
vocab_size=self.vocab_size,
|
|
hidden_size=self.hidden_size,
|
|
num_attention_heads=self.num_attention_heads,
|
|
max_sequence_length=self.max_sequence_length,
|
|
layernorm_epsilon=self.layernorm_epsilon,
|
|
inner_hidden_size=self.inner_hidden_size,
|
|
use_cache=self.use_cache,
|
|
bos_token_id=self.bos_token_id,
|
|
eos_token_id=self.eos_token_id,
|
|
pad_token_id=self.pad_token_id,
|
|
mask_token_id=self.mask_token_id,
|
|
gmask_token_id=self.gmask_token_id,
|
|
position_encoding_2d=self.position_encoding_2d,
|
|
quantization_bit=self.quantization_bit,
|
|
pre_seq_len=self.pre_seq_len,
|
|
prefix_projection=self.prefix_projection,
|
|
output_predict=self.output_predict,
|
|
recompute=self.recompute,
|
|
attention_scale=self.attention_scale,
|
|
activation=self.activation,
|
|
num_image_tokens=self.num_image_tokens,
|
|
)
|
|
|
|
def create_and_check_model(self, config, input_ids, labels, attention_mask, position_ids):
|
|
model = ChatGLMModel(config)
|
|
model.eval()
|
|
|
|
result = model(input_ids, attention_mask=attention_mask, position_ids=position_ids)
|
|
self.parent.assertEqual(result[0].shape, [self.seq_length, self.batch_size, self.hidden_size])
|
|
|
|
# def create_and_check_model_past_large_inputs(self, config, input_ids, labels):
|
|
# model = ChatGLMModel(config)
|
|
# model.eval()
|
|
|
|
# outputs = model(input_ids, attention_mask=attention_mask, return_dict=self.return_dict)
|
|
# past_key_values = outputs.past_key_values[0] if self.return_dict else outputs[1][0]
|
|
|
|
# next_tokens = ids_tensor([self.batch_size, 3], self.vocab_size)
|
|
# next_input_ids = paddle.concat([input_ids, next_tokens], axis=-1)
|
|
# next_attention_mask = model.get_masks(next_input_ids)
|
|
|
|
# outputs = model(next_input_ids, attention_mask=next_attention_mask, return_dict=self.return_dict)
|
|
# output_from_no_past = outputs.past_key_values[0] if self.return_dict else outputs[1][0]
|
|
|
|
# outputs = model(
|
|
# next_tokens,
|
|
# attention_mask=next_attention_mask,
|
|
# past_key_values=past_key_values,
|
|
# return_dict=self.return_dict,
|
|
# )
|
|
# output_from_past = outputs.past_key_values[0] if self.return_dict else 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))
|
|
|
|
def prepare_config_and_inputs_for_common(self):
|
|
config_and_inputs = self.prepare_config_and_inputs()
|
|
config, input_ids, labels, attention_mask, position_ids = config_and_inputs
|
|
inputs_dict = {"input_ids": input_ids, "attention_mask": attention_mask, "position_ids": position_ids}
|
|
return config, inputs_dict
|
|
|
|
def create_and_check_lm_head_model(self, config, input_ids, labels, attention_mask, position_ids):
|
|
model = ChatGLMForCausalLM(config)
|
|
model.eval()
|
|
|
|
result = model(
|
|
input_ids,
|
|
attention_mask=attention_mask,
|
|
labels=labels if self.parent.use_labels else None,
|
|
return_dict=self.parent.return_dict,
|
|
)
|
|
if self.parent.use_labels:
|
|
loss = result.loss if self.parent.return_dict else result[0]
|
|
self.parent.assertEqual(loss.shape, [1])
|
|
logits = result.logits if self.parent.return_dict else result[1]
|
|
past_key_values = result.past_key_values[0] if self.parent.return_dict else result[2][0]
|
|
else:
|
|
loss = result.loss if self.parent.return_dict else None
|
|
self.parent.assertTrue(loss is None)
|
|
logits = result.logits if self.parent.return_dict else result[0]
|
|
past_key_values = result.past_key_values[0] if self.parent.return_dict else result[1][0]
|
|
self.parent.assertEqual(logits.shape, [self.batch_size, self.seq_length, self.vocab_size])
|
|
if config.use_cache:
|
|
self.parent.assertTrue(isinstance(past_key_values, tuple))
|
|
self.parent.assertEqual(past_key_values[0].shape, [self.batch_size, self.seq_length, self.hidden_size])
|
|
else:
|
|
self.parent.assertTrue(past_key_values is None)
|
|
|
|
def create_and_check_model_attention_mask(self, config: ChatGLMConfig, input_ids, labels):
|
|
model = ChatGLMModel(config)
|
|
model.eval()
|
|
attn_mask_2d = random_attention_mask([self.batch_size, self.seq_length])
|
|
result_2d = model(input_ids, attention_mask=attn_mask_2d)[0].transpose([1, 0, 2])
|
|
batch, seq_length = input_ids.shape
|
|
causal_mask = paddle.tril(paddle.ones((batch, seq_length, seq_length), dtype=attn_mask_2d.dtype))
|
|
attn_mask_3d = causal_mask & attn_mask_2d.unsqueeze(-1)
|
|
result_3d = model(input_ids, attention_mask=attn_mask_3d)[0].transpose([1, 0, 2])
|
|
|
|
# use 4d mask for chatglm must prepocess prefix mask and padding mask
|
|
attn_mask_4d = attn_mask_3d.unsqueeze(1)
|
|
context_lengths, pad_lengths = [], []
|
|
for seq in input_ids:
|
|
context_lengths.append(paddle.where(seq == self.bos_token_id)[0][0])
|
|
pad_lengths.append(paddle.where(seq != self.pad_token_id)[0][0])
|
|
|
|
for i, context_length in enumerate(context_lengths):
|
|
attn_mask_4d[i, :, :, :context_length] = 1
|
|
print(attn_mask_4d)
|
|
|
|
for i, pad_length in enumerate(pad_lengths):
|
|
attn_mask_4d[i, :pad_length, :pad_length] = 0
|
|
print(attn_mask_4d)
|
|
|
|
result_4d = model(input_ids, attention_mask=attn_mask_4d)[0].transpose([1, 0, 2])
|
|
result_no_attention_mask = model(input_ids, attention_mask=None)[0].transpose([1, 0, 2])
|
|
# Assert non-padding tokens have the same logits with different attention_mask shape
|
|
self.parent.assertTrue((result_2d[attn_mask_2d] == result_3d[attn_mask_2d]).all())
|
|
self.parent.assertTrue((result_2d[attn_mask_2d] == result_4d[attn_mask_2d]).all())
|
|
self.parent.assertTrue((result_2d[attn_mask_2d] == result_no_attention_mask[attn_mask_2d]).all())
|
|
|
|
|
|
class ChatGLMTest(ModelTesterMixin, GenerationTesterMixin, unittest.TestCase):
|
|
base_model_class = ChatGLMModel
|
|
return_dict = False
|
|
use_labels = False
|
|
|
|
all_model_classes = (ChatGLMModel, ChatGLMForCausalLM)
|
|
all_generative_model_classes = {ChatGLMForCausalLM: (ChatGLMModel, "chatglm")}
|
|
|
|
def setUp(self):
|
|
super().setUp()
|
|
|
|
self.model_tester = ChatGLMTester(self)
|
|
self.config_tester = ConfigTester(self, config_class=ChatGLMConfig, vocab_size=256, hidden_size=24)
|
|
|
|
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]
|
|
attention_mask = inputs_dict["attention_mask"]
|
|
position_ids = inputs_dict["position_ids"]
|
|
|
|
max_batch_size = 2
|
|
sequence_length = input_ids.shape[-1]
|
|
input_ids = input_ids[:max_batch_size, :sequence_length]
|
|
attention_mask = attention_mask[:max_batch_size, :, :sequence_length, :sequence_length]
|
|
position_ids = position_ids[:max_batch_size, :, :sequence_length]
|
|
|
|
# generate max 3 tokens
|
|
max_length = 3
|
|
|
|
if config.eos_token_id or config.pad_token_id:
|
|
# hack to allow generate for models such as GPT2 as is done in `generate()`
|
|
config["pad_token_id"] = config["eos_token_id"]
|
|
|
|
return config, input_ids, attention_mask, max_length
|
|
|
|
def test_model(self):
|
|
config_and_inputs = self.model_tester.prepare_config_and_inputs()
|
|
self.model_tester.create_and_check_model(*config_and_inputs)
|
|
|
|
def test_model_name_list(self):
|
|
pass
|
|
|
|
def test_group_beam_search_generate(self):
|
|
pass
|
|
|
|
def test_beam_search_generate(self):
|
|
pass
|
|
|
|
def test_generate_without_input_ids(self):
|
|
pass
|
|
|
|
def test_resize_tokens_embeddings(self):
|
|
pass
|
|
|
|
def test_chatglm_lm_head_model(self):
|
|
config_and_inputs = self.model_tester.prepare_config_and_inputs()
|
|
self.model_tester.create_and_check_lm_head_model(*config_and_inputs)
|
|
|
|
# def test_model_attention_mask(self):
|
|
# config_and_inputs = self.model_tester.prepare_config_and_inputs()
|
|
# self.model_tester.create_and_check_model_attention_mask(*config_and_inputs)
|
|
|
|
|
|
# class ChatGLMGenerationD2STest(GenerationD2STestMixin, unittest.TestCase):
|
|
# internal_testing_model = "__internal_testing__/tiny-random-chatglm"
|
|
# TokenizerClass = ChatGLMTokenizer
|
|
# CausalLMClass = ChatGLMForCausalLM
|
|
|
|
# def test_to_static_use_top_k(self):
|
|
# tokenizer = self.TokenizerClass.from_pretrained(self.internal_testing_model)
|
|
# model = self.CausalLMClass.from_pretrained(self.internal_testing_model)
|
|
|
|
# model_kwargs = tokenizer(
|
|
# self.article,
|
|
# max_length=self.max_length,
|
|
# truncation=True,
|
|
# truncation_side="left",
|
|
# return_tensors="pd",
|
|
# padding=True,
|
|
# add_special_tokens=True,
|
|
# )
|
|
|
|
# model.eval()
|
|
|
|
# # Llama model do not contains ``
|
|
# model.is_encoder_decoder = False
|
|
|
|
# max_length = self.max_length
|
|
|
|
# model_kwargs["use_cache"] = True
|
|
# model_kwargs["max_length"] = max_length + model_kwargs["input_ids"].shape[-1]
|
|
|
|
# decoded_ids = model.greedy_search(
|
|
# logits_processors=None,
|
|
# bos_token_id=model.config.bos_token_id,
|
|
# pad_token_id=model.config.pad_token_id,
|
|
# eos_token_id=model.config.eos_token_id,
|
|
# **model_kwargs,
|
|
# )[0]
|
|
|
|
# dygraph_decoded_ids = decoded_ids.tolist()
|
|
|
|
# with static_mode_guard():
|
|
# with tempfile.TemporaryDirectory() as tempdir:
|
|
# path = os.path.join(tempdir, "model")
|
|
# model.to_static(
|
|
# path,
|
|
# config=dict(
|
|
# use_top_p=False,
|
|
# ),
|
|
# )
|
|
|
|
# model_path = os.path.join(tempdir, f"model{PADDLE_INFERENCE_MODEL_SUFFIX}")
|
|
# params_path = os.path.join(tempdir, f"model{PADDLE_INFERENCE_WEIGHTS_SUFFIX}")
|
|
# config = paddle.inference.Config(model_path, params_path)
|
|
|
|
# config.disable_gpu()
|
|
# config.disable_glog_info()
|
|
# predictor = paddle.inference.create_predictor(config)
|
|
|
|
# model_kwargs["top_k"] = 1
|
|
# model_kwargs["attention_mask"] = model_kwargs["attention_mask"].astype("int64")
|
|
# model_kwargs["max_length"] = self.max_length
|
|
# # create input
|
|
# for key in model_kwargs.keys():
|
|
# if paddle.is_tensor(model_kwargs[key]):
|
|
# model_kwargs[key] = model_kwargs[key].numpy()
|
|
# elif isinstance(model_kwargs[key], float):
|
|
# model_kwargs[key] = np.array(model_kwargs[key], dtype="float32")
|
|
# else:
|
|
# model_kwargs[key] = np.array(model_kwargs[key], dtype="int64")
|
|
|
|
# input_handles = {}
|
|
# for name in predictor.get_input_names():
|
|
# input_handles[name] = predictor.get_input_handle(name)
|
|
|
|
# input_handles[name].copy_from_cpu(model_kwargs[name])
|
|
|
|
# predictor.run()
|
|
# output_names = predictor.get_output_names()
|
|
# output_handle = predictor.get_output_handle(output_names[0])
|
|
# results = output_handle.copy_to_cpu()
|
|
|
|
# static_decoded_ids = results.tolist()
|
|
|
|
# self.assertEqual(dygraph_decoded_ids, static_decoded_ids)
|
|
|
|
# def test_to_static_use_top_p(self):
|
|
# tokenizer = self.TokenizerClass.from_pretrained(self.internal_testing_model)
|
|
# model = self.CausalLMClass.from_pretrained(self.internal_testing_model)
|
|
|
|
# model_kwargs = tokenizer(
|
|
# self.article,
|
|
# max_length=self.max_length,
|
|
# truncation=True,
|
|
# truncation_side="left",
|
|
# return_tensors="pd",
|
|
# padding=True,
|
|
# add_special_tokens=True,
|
|
# )
|
|
|
|
# model.eval()
|
|
|
|
# # Llama model do not contains ``
|
|
# model.is_encoder_decoder = False
|
|
|
|
# max_length = self.max_length
|
|
|
|
# model_kwargs["use_cache"] = True
|
|
# model_kwargs["max_length"] = max_length + model_kwargs["input_ids"].shape[-1]
|
|
|
|
# with static_mode_guard():
|
|
# with tempfile.TemporaryDirectory() as tempdir:
|
|
|
|
# path = os.path.join(tempdir, "model")
|
|
# model.to_static(
|
|
# path,
|
|
# config=dict(
|
|
# use_top_p=False,
|
|
# ),
|
|
# )
|
|
|
|
# model_path = os.path.join(tempdir, f"model{PADDLE_INFERENCE_MODEL_SUFFIX}")
|
|
# params_path = os.path.join(tempdir, f"model{PADDLE_INFERENCE_WEIGHTS_SUFFIX}")
|
|
# config = paddle.inference.Config(model_path, params_path)
|
|
|
|
# config.disable_gpu()
|
|
# config.disable_glog_info()
|
|
# predictor = paddle.inference.create_predictor(config)
|
|
|
|
# model_kwargs["attention_mask"] = model_kwargs["attention_mask"].astype("int64")
|
|
# model_kwargs["top_k"] = 1
|
|
# model_kwargs["max_length"] = self.max_length
|
|
# # create input
|
|
# for key in model_kwargs.keys():
|
|
# if paddle.is_tensor(model_kwargs[key]):
|
|
# model_kwargs[key] = model_kwargs[key].numpy()
|
|
# else:
|
|
# model_kwargs[key] = np.array(model_kwargs[key])
|
|
|
|
# input_handles = {}
|
|
# for name in predictor.get_input_names():
|
|
# input_handles[name] = predictor.get_input_handle(name)
|
|
# input_handles[name].copy_from_cpu(model_kwargs[name])
|
|
|
|
# predictor.run()
|
|
# output_names = predictor.get_output_names()
|
|
# output_handle = predictor.get_output_handle(output_names[0])
|
|
# results = output_handle.copy_to_cpu()
|
|
|
|
# self.assertIsNotNone(results)
|
|
|
|
|
|
if __name__ == "__main__":
|
|
unittest.main()
|