File size: 4,329 Bytes
8a48a42 3ce1634 23e8695 ac57c7b 23e8695 3ce1634 2da6076 3ce1634 61b749b 2da6076 61b749b 2da6076 |
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 |
---
language:
- en
- zh
tags:
- translation
license: apache-2.0
datasets:
- DDDSSS/en-zh-dataset
metrics:
- bleu
- sacrebleu
---
该模型主要的训练数据是opus100和CodeAlpaca_20K中的英文作为翻译内容，采用chatglm作为翻译器翻译成中文，并将脏数据筛选后得到DDDSSS/en-zh-dataset数据集，
缺点是这个模型的sentence len 较短，需要自己进行分句，要不然可能会出现，少翻或者不翻译的情况出现
!注意，如果是pretrain方法下载模型的话，可能部分参数会随机初始化，建议直接下载模型，并从本地读取。
from transformers import AutoModelForSeq2SeqLM, AutoTokenizer, pipeline
parser.add_argument('--device', default="cpu", type=str, help='"cuda:1"、"cuda:2"……')
mode_name = opt.model
device = opt.device
model = AutoModelForSeq2SeqLM.from_pretrained(mode_name)
tokenizer = AutoTokenizer.from_pretrained(mode_name)
translation = pipeline("translation_en_to_zh", model=model, tokenizer=tokenizer,
torch_dtype="float", device_map=True,device=device)
x=["If nothing is detected and there is a config.json file, it’s assumed the library is transformers.","By looking into the presence of files such as *.nemo or *saved_model.pb*, the Hub can determine if a model is from NeMo or Keras."]
re = translation(x, max_length=450)
print('翻译为：' ,re)
微调：
import numpy as np
from transformers import AutoModelForSeq2SeqLM, Seq2SeqTrainingArguments, Seq2SeqTrainer
import torch
# books = load_from_disk("")
books = load_dataset("json", data_files=".json")
books = books["train"].train_test_split(test_size=0.2)
checkpoint = "./opus-mt-en-zh"
# checkpoint = "./model/checkpoint-19304"
tokenizer = AutoTokenizer.from_pretrained(checkpoint)
source_lang = "en"
target_lang = "zh"
def preprocess_function(examples):
inputs = [example[source_lang] for example in examples["translation"]]
targets = [example[target_lang] for example in examples["translation"]]
model_inputs = tokenizer(inputs, text_target=targets, max_length=512, truncation=True)
return model_inputs
tokenized_books = books.map(preprocess_function, batched=True)
data_collator = DataCollatorForSeq2Seq(tokenizer=tokenizer, model=checkpoint)
metric = evaluate.load("sacrebleu")
def postprocess_text(preds, labels):
preds = [pred.strip() for pred in preds]
labels = [[label.strip()] for label in labels]
return preds, labels
def compute_metrics(eval_preds):
preds, labels = eval_preds
if isinstance(preds, tuple):
preds = preds[0]
decoded_preds = tokenizer.batch_decode(preds, skip_special_tokens=True)
labels = np.where(labels != -100, labels, tokenizer.pad_token_id)
decoded_labels = tokenizer.batch_decode(labels, skip_special_tokens=True)
decoded_preds, decoded_labels = postprocess_text(decoded_preds, decoded_labels)
result = metric.compute(predictions=decoded_preds, references=decoded_labels)
result = {"bleu": result["score"]}
prediction_lens = [np.count_nonzero(pred != tokenizer.pad_token_id) for pred in preds]
result["gen_len"] = np.mean(prediction_lens)
result = {k: round(v, 4) for k, v in result.items()}
return result
model = AutoModelForSeq2SeqLM.from_pretrained(checkpoint)
batchsize=4
training_args = Seq2SeqTrainingArguments(
output_dir="./my_awesome_opus_books_model",
evaluation_strategy="epoch",
learning_rate=2e-4,
per_device_train_batch_size=batchsize,
per_device_eval_batch_size=batchsize,
weight_decay=0.01,
# save_total_limit=3,
num_train_epochs=4,
predict_with_generate=True,
fp16=True,
push_to_hub=False,
save_strategy="epoch",
jit_mode_eval=True
)
trainer = Seq2SeqTrainer(
model=model,
args=training_args,
train_dataset=tokenized_books["train"],
eval_dataset=tokenized_books["test"],
tokenizer=tokenizer,
data_collator=data_collator,
compute_metrics=compute_metrics,
)
trainer.train() |