我想测量 pretrained_model 的困惑度。这个 pretrained_model 是 self_made。代码如下。
import pickle
import os
import logging
logger = logging.getLogger(__name__)
logging.basicConfig(
format="%(asctime)s - %(levelname)s - %(name)s - %(message)s",
datefmt="%m/%d/%Y %H:%M:%S",
level=logging.INFO,
)
import threading
os.environ["CUDA_LAUNCH_BLOCKING"] = "1"
import torch,sys
from torch.utils.data import DataLoader
from transformers import CamembertForMaskedLM, CamembertTokenizer
seed = 666
torch.manual_seed(seed)
torch.cuda.manual_seed(seed)
tokenizer = CamembertTokenizer.from_pretrained('/home/~/tokenizers_')
import numpy as np
from tqdm import tqdm
import sys, pickle5
def dynamic_masking(inputs, mlm_probability=0.15):
labels = inputs.clone()
probability_matrix = torch.full(labels.shape, mlm_probability)
special_tokens_mask = list()
for row in labels:
special_tokens_mask.append([1 if token_id in tokenizer.convert_tokens_to_ids(['<s>', '</s>', '<spk>', '</spk>']) else 0 for token_id in row])
# special_tokens_mask = [
# tokenizer.get_special_tokens_mask(val, already_has_special_tokens=True) for row in labels.tolist()
# ]
probability_matrix.masked_fill_(torch.tensor(special_tokens_mask, dtype=torch.bool), value=0.0)
if tokenizer._pad_token is not None:
padding_mask = labels.eq(tokenizer.pad_token_id)
probability_matrix.masked_fill_(padding_mask, value=0.0)
masked_indices = torch.bernoulli(probability_matrix).bool()
labels[~masked_indices] = -100 # We only compute loss on masked tokens. (# -100 means ignore target index in pytorch's CrossEntropy.)
# 80% of the time, we replace masked input tokens with tokenizer.mask_token ([MASK])
indices_replaced = torch.bernoulli(torch.full(labels.shape, 0.8)).bool() & masked_indices
inputs[indices_replaced] = tokenizer.convert_tokens_to_ids(tokenizer.mask_token)
# 10% of the time, we replace masked input tokens with random word
indices_random = torch.bernoulli(torch.full(labels.shape, 0.5)).bool() & masked_indices & ~indices_replaced
random_words = torch.randint(len(tokenizer), labels.shape, dtype=torch.long)
inputs[indices_random] = random_words[indices_random]
# The rest of the time (10% of the time) we keep the masked input tokens unchanged
return inputs, labels
def syncronized_dynamic_masking(inputs):
half_size = inputs.size(1) // 2
inputs, labels = dynamic_masking(inputs)
tmp_list = list()
semi_inputs = torch.empty(0, half_size+1, dtype=torch.long)
for row in inputs:
row_l = row.tolist()
tmp_list.append(row_l[:(half_size)] + [6])
# print(tokenizer.convert_ids_to_tokens(row_l[:(half)] + [6]))
tmp_list.append([5] + row_l[half_size:])
# print(tokenizer.convert_ids_to_tokens([5] + row_l[half:]))
semi_inputs = torch.cat((semi_inputs, torch.tensor(tmp_list)), dim=0)
tmp_list.clear()
semi_labels = torch.empty(0, half_size+1, dtype=torch.long)
for row in labels:
row_l = row.tolist()
tmp_list.append(row_l[:(half_size)] + [-100])
# print(tokenizer.convert_ids_to_tokens(row_l[:(half)] + [6]))
tmp_list.append([-100] + row_l[half_size:])
# print(tokenizer.convert_ids_to_tokens([5] + row_l[half:]))
semi_labels = torch.cat((semi_labels, torch.tensor(tmp_list)), dim=0)
return (inputs, labels), (semi_inputs, semi_labels)
def evaluate(model, batch):
# device = torch.device('cuda', index=gpu_nums[0])#for debug
device = torch.device('cpu')
inputs, labels = batch
# print(inputs.shape)#torch.Size([128, 128])
# print('-' * 50)
# print(labels.shape)#torch.Size([128, 128])
inputs = inputs.to(device)
labels = labels.to(device)
with torch.no_grad():
# print(inputs.shape)#torch.Size([128, 128])
# print(type(inputs))#class 'torch.Tensor'
# print(labels.shape)#torch.Size([128, 128])
# print(type(labels))#class 'torch.Tensor'
# sys.exit()
outputs = model(inputs, masked_lm_labels=labels) # tuple
lm_loss = outputs[0] # torch.Size([]), the mean of the loss per mask token
eval_loss = lm_loss.mean().item() # mean() will not be needed.
return eval_loss
def evaluate_wrapper(model, batch, gpu_nums:list, result:dict):
print('evaluate score: ', evaluate(model, batch, gpu_nums))
result[threading.current_thread().name] = evaluate(model, batch, gpu_nums)
if __name__ == "__main__":
model = '/home/~/checkpoint-1000000'
model = CamembertForMaskedLM.from_pretrained(model)
# model = torch.nn.DataParallel(model, device_ids=[1,2])
# model['large'].to(torch.device('cuda', index=1))#for debug
model.to(torch.device('cpu'))#for debug
batch_size = 128
corpus = "/home/~/test-128.pickle"
dataset = pickle5.load(open(corpus, mode='rb'))
dl = DataLoader(dataset, batch_size=batch_size, shuffle=False)
logger.info("***** Running evaluation *****")
logger.info(" Num examples = %d", len(dataset))
logger.info(" large: Batch size = %d", batch_size)
# logger.info(" small: Batch size = %d", batch_size*2)
large_eval_loss = 0.0
nb_eval_steps = 0
model.eval()
for batch in tqdm(dl, desc="Evaluating"):
# inputs, labels = dynamic_masking(batch)
batch = dynamic_masking(batch)
# b_for_large, b_for_small = syncronized_dynamic_masking(batch)
result = dict()
large_eval_loss += evaluate(model, batch)
nb_eval_steps += 1
large_eval_loss = large_eval_loss / nb_eval_steps
perplexity = {
'large': torch.exp(torch.tensor(large_eval_loss)).item()
}
logger.info("***** Eval results *****")
for key in sorted(perplexity):
logger.info("perplexity:")
logger.info(" %s = %s", key, str(perplexity[key]))
print('perplexity:', str(perplexity[key]))模型内容如下。
CamembertForMaskedLM(
(roberta): RobertaModel(
(embeddings): RobertaEmbeddings(
(word_embeddings): Embedding(32005, 768, padding_idx=1)
(position_embeddings): Embedding(130, 768, padding_idx=1)
(token_type_embeddings): Embedding(1, 768)
(LayerNorm): LayerNorm((768,), eps=1e-05, elementwise_affine=True)
(dropout): Dropout(p=0.1, inplace=False)
)
(encoder): BertEncoder(
(layer): ModuleList(
(0): BertLayer(
(attention): BertAttention(
(self): BertSelfAttention(
(query): Linear(in_features=768, out_features=768, bias=True)
(key): Linear(in_features=768, out_features=768, bias=True)
(value): Linear(in_features=768, out_features=768, bias=True)
(dropout): Dropout(p=0.1, inplace=False)
)
(output): BertSelfOutput(
(dense): Linear(in_features=768, out_features=768, bias=True)
(LayerNorm): LayerNorm((768,), eps=1e-05, elementwise_affine=True)
(dropout): Dropout(p=0.1, inplace=False)
)
)
(intermediate): BertIntermediate(
(dense): Linear(in_features=768, out_features=3072, bias=True)
)
(output): BertOutput(
(dense): Linear(in_features=3072, out_features=768, bias=True)
(LayerNorm): LayerNorm((768,), eps=1e-05, elementwise_affine=True)
(dropout): Dropout(p=0.1, inplace=False)
)
)
(1): BertLayer(
(attention): BertAttention(
(self): BertSelfAttention(
(query): Linear(in_features=768, out_features=768, bias=True)
(key): Linear(in_features=768, out_features=768, bias=True)
(value): Linear(in_features=768, out_features=768, bias=True)
(dropout): Dropout(p=0.1, inplace=False)
)
(output): BertSelfOutput(
(dense): Linear(in_features=768, out_features=768, bias=True)
(LayerNorm): LayerNorm((768,), eps=1e-05, elementwise_affine=True)
(dropout): Dropout(p=0.1, inplace=False)
)
)
(intermediate): BertIntermediate(
(dense): Linear(in_features=768, out_features=3072, bias=True)
)
(output): BertOutput(
(dense): Linear(in_features=3072, out_features=768, bias=True)
(LayerNorm): LayerNorm((768,), eps=1e-05, elementwise_affine=True)
(dropout): Dropout(p=0.1, inplace=False)
)
)
(2): BertLayer(
(attention): BertAttention(
(self): BertSelfAttention(
(query): Linear(in_features=768, out_features=768, bias=True)
(key): Linear(in_features=768, out_features=768, bias=True)
(value): Linear(in_features=768, out_features=768, bias=True)
(dropout): Dropout(p=0.1, inplace=False)
)
(output): BertSelfOutput(
(dense): Linear(in_features=768, out_features=768, bias=True)
(LayerNorm): LayerNorm((768,), eps=1e-05, elementwise_affine=True)
(dropout): Dropout(p=0.1, inplace=False)
)
)
(intermediate): BertIntermediate(
(dense): Linear(in_features=768, out_features=3072, bias=True)
)
(output): BertOutput(
(dense): Linear(in_features=3072, out_features=768, bias=True)
(LayerNorm): LayerNorm((768,), eps=1e-05, elementwise_affine=True)
(dropout): Dropout(p=0.1, inplace=False)
)
)
(3): BertLayer(
(attention): BertAttention(
(self): BertSelfAttention(
(query): Linear(in_features=768, out_features=768, bias=True)
(key): Linear(in_features=768, out_features=768, bias=True)
(value): Linear(in_features=768, out_features=768, bias=True)
(dropout): Dropout(p=0.1, inplace=False)
)
(output): BertSelfOutput(
(dense): Linear(in_features=768, out_features=768, bias=True)
(LayerNorm): LayerNorm((768,), eps=1e-05, elementwise_affine=True)
(dropout): Dropout(p=0.1, inplace=False)
)
)
(intermediate): BertIntermediate(
(dense): Linear(in_features=768, out_features=3072, bias=True)
)
(output): BertOutput(
(dense): Linear(in_features=3072, out_features=768, bias=True)
(LayerNorm): LayerNorm((768,), eps=1e-05, elementwise_affine=True)
(dropout): Dropout(p=0.1, inplace=False)
)
)
(4): BertLayer(
(attention): BertAttention(
(self): BertSelfAttention(
(query): Linear(in_features=768, out_features=768, bias=True)
(key): Linear(in_features=768, out_features=768, bias=True)
(value): Linear(in_features=768, out_features=768, bias=True)
(dropout): Dropout(p=0.1, inplace=False)
)
(output): BertSelfOutput(
(dense): Linear(in_features=768, out_features=768, bias=True)
(LayerNorm): LayerNorm((768,), eps=1e-05, elementwise_affine=True)
(dropout): Dropout(p=0.1, inplace=False)
)
)
(intermediate): BertIntermediate(
(dense): Linear(in_features=768, out_features=3072, bias=True)
)
(output): BertOutput(
(dense): Linear(in_features=3072, out_features=768, bias=True)
(LayerNorm): LayerNorm((768,), eps=1e-05, elementwise_affine=True)
(dropout): Dropout(p=0.1, inplace=False)
)
)
(5): BertLayer(
(attention): BertAttention(
(self): BertSelfAttention(
(query): Linear(in_features=768, out_features=768, bias=True)
(key): Linear(in_features=768, out_features=768, bias=True)
(value): Linear(in_features=768, out_features=768, bias=True)
(dropout): Dropout(p=0.1, inplace=False)
)
(output): BertSelfOutput(
(dense): Linear(in_features=768, out_features=768, bias=True)
(LayerNorm): LayerNorm((768,), eps=1e-05, elementwise_affine=True)
(dropout): Dropout(p=0.1, inplace=False)
)
)
(intermediate): BertIntermediate(
(dense): Linear(in_features=768, out_features=3072, bias=True)
)
(output): BertOutput(
(dense): Linear(in_features=3072, out_features=768, bias=True)
(LayerNorm): LayerNorm((768,), eps=1e-05, elementwise_affine=True)
(dropout): Dropout(p=0.1, inplace=False)
)
)
(6): BertLayer(
(attention): BertAttention(
(self): BertSelfAttention(
(query): Linear(in_features=768, out_features=768, bias=True)
(key): Linear(in_features=768, out_features=768, bias=True)
(value): Linear(in_features=768, out_features=768, bias=True)
(dropout): Dropout(p=0.1, inplace=False)
)
(output): BertSelfOutput(
(dense): Linear(in_features=768, out_features=768, bias=True)
(LayerNorm): LayerNorm((768,), eps=1e-05, elementwise_affine=True)
(dropout): Dropout(p=0.1, inplace=False)
)
)
(intermediate): BertIntermediate(
(dense): Linear(in_features=768, out_features=3072, bias=True)
)
(output): BertOutput(
(dense): Linear(in_features=3072, out_features=768, bias=True)
(LayerNorm): LayerNorm((768,), eps=1e-05, elementwise_affine=True)
(dropout): Dropout(p=0.1, inplace=False)
)
)
(7): BertLayer(
(attention): BertAttention(
(self): BertSelfAttention(
(query): Linear(in_features=768, out_features=768, bias=True)
(key): Linear(in_features=768, out_features=768, bias=True)
(value): Linear(in_features=768, out_features=768, bias=True)
(dropout): Dropout(p=0.1, inplace=False)
)
(output): BertSelfOutput(
(dense): Linear(in_features=768, out_features=768, bias=True)
(LayerNorm): LayerNorm((768,), eps=1e-05, elementwise_affine=True)
(dropout): Dropout(p=0.1, inplace=False)
)
)
(intermediate): BertIntermediate(
(dense): Linear(in_features=768, out_features=3072, bias=True)
)
(output): BertOutput(
(dense): Linear(in_features=3072, out_features=768, bias=True)
(LayerNorm): LayerNorm((768,), eps=1e-05, elementwise_affine=True)
(dropout): Dropout(p=0.1, inplace=False)
)
)
(8): BertLayer(
(attention): BertAttention(
(self): BertSelfAttention(
(query): Linear(in_features=768, out_features=768, bias=True)
(key): Linear(in_features=768, out_features=768, bias=True)
(value): Linear(in_features=768, out_features=768, bias=True)
(dropout): Dropout(p=0.1, inplace=False)
)
(output): BertSelfOutput(
(dense): Linear(in_features=768, out_features=768, bias=True)
(LayerNorm): LayerNorm((768,), eps=1e-05, elementwise_affine=True)
(dropout): Dropout(p=0.1, inplace=False)
)
)
(intermediate): BertIntermediate(
(dense): Linear(in_features=768, out_features=3072, bias=True)
)
(output): BertOutput(
(dense): Linear(in_features=3072, out_features=768, bias=True)
(LayerNorm): LayerNorm((768,), eps=1e-05, elementwise_affine=True)
(dropout): Dropout(p=0.1, inplace=False)
)
)
(9): BertLayer(
(attention): BertAttention(
(self): BertSelfAttention(
(query): Linear(in_features=768, out_features=768, bias=True)
(key): Linear(in_features=768, out_features=768, bias=True)
(value): Linear(in_features=768, out_features=768, bias=True)
(dropout): Dropout(p=0.1, inplace=False)
)
(output): BertSelfOutput(
(dense): Linear(in_features=768, out_features=768, bias=True)
(LayerNorm): LayerNorm((768,), eps=1e-05, elementwise_affine=True)
(dropout): Dropout(p=0.1, inplace=False)
)
)
(intermediate): BertIntermediate(
(dense): Linear(in_features=768, out_features=3072, bias=True)
)
(output): BertOutput(
(dense): Linear(in_features=3072, out_features=768, bias=True)
(LayerNorm): LayerNorm((768,), eps=1e-05, elementwise_affine=True)
(dropout): Dropout(p=0.1, inplace=False)
)
)
(10): BertLayer(
(attention): BertAttention(
(self): BertSelfAttention(
(query): Linear(in_features=768, out_features=768, bias=True)
(key): Linear(in_features=768, out_features=768, bias=True)
(value): Linear(in_features=768, out_features=768, bias=True)
(dropout): Dropout(p=0.1, inplace=False)
)
(output): BertSelfOutput(
(dense): Linear(in_features=768, out_features=768, bias=True)
(LayerNorm): LayerNorm((768,), eps=1e-05, elementwise_affine=True)
(dropout): Dropout(p=0.1, inplace=False)
)
)
(intermediate): BertIntermediate(
(dense): Linear(in_features=768, out_features=3072, bias=True)
)
(output): BertOutput(
(dense): Linear(in_features=3072, out_features=768, bias=True)
(LayerNorm): LayerNorm((768,), eps=1e-05, elementwise_affine=True)
(dropout): Dropout(p=0.1, inplace=False)
)
)
(11): BertLayer(
(attention): BertAttention(
(self): BertSelfAttention(
(query): Linear(in_features=768, out_features=768, bias=True)
(key): Linear(in_features=768, out_features=768, bias=True)
(value): Linear(in_features=768, out_features=768, bias=True)
(dropout): Dropout(p=0.1, inplace=False)
)
(output): BertSelfOutput(
(dense): Linear(in_features=768, out_features=768, bias=True)
(LayerNorm): LayerNorm((768,), eps=1e-05, elementwise_affine=True)
(dropout): Dropout(p=0.1, inplace=False)
)
)
(intermediate): BertIntermediate(
(dense): Linear(in_features=768, out_features=3072, bias=True)
)
(output): BertOutput(
(dense): Linear(in_features=3072, out_features=768, bias=True)
(LayerNorm): LayerNorm((768,), eps=1e-05, elementwise_affine=True)
(dropout): Dropout(p=0.1, inplace=False)
)
)
)
)
(pooler): BertPooler(
(dense): Linear(in_features=768, out_features=768, bias=True)
(activation): Tanh()
)
)
(lm_head): RobertaLMHead(
(dense): Linear(in_features=768, out_features=768, bias=True)
(layer_norm): LayerNorm((768,), eps=1e-05, elementwise_affine=True)
(decoder): Linear(in_features=768, out_features=32005, bias=True)
)
)错误IndexError: index out of range in self在outputs = model(inputs, masked_lm_labels=labels) # tuple
输入和标签形状都是 torch.Size([128, 128])，这有什么问题？
有人可以告诉我该怎么做吗？如果我遗漏了什么，请指出。