Xlm Roberta Large Qe V1

🚀 机器翻译质量评估模型

本模型用于机器翻译系统的无参考质量评估（QE），基于FacebookAI/xlm-roberta-large模型在ymoslem/wmt-da-human-evaluation数据集上微调得到，能有效评估机器翻译的质量。

🚀 快速开始

安装依赖库

pip3 install --upgrade datasets accelerate transformers
pip3 install --upgrade flash_attn triton

加载测试数据集

from datasets import load_dataset

test_dataset = load_dataset("ymoslem/wmt-da-human-evaluation",
                             split="test",
                             trust_remote_code=True
                            )
print(test_dataset)

加载模型和分词器

from transformers import AutoModelForSequenceClassification, AutoTokenizer
import torch

# Load the fine-tuned model and tokenizer
model_name = "ymoslem/ModernBERT-large-qe-v1"
model = AutoModelForSequenceClassification.from_pretrained(
    model_name,
    device_map="auto",
    torch_dtype=torch.bfloat16,
    attn_implementation="flash_attention_2",
)
tokenizer = AutoTokenizer.from_pretrained(model_name)

# Move model to GPU if available
device = "cuda" if torch.cuda.is_available() else "cpu"
model.to(device)
model.eval()

准备数据集

sep_token = tokenizer.sep_token
input_test_texts = [f"{src} {sep_token} {tgt}" for src, tgt in zip(test_dataset["src"], test_dataset["mt"])]

生成预测结果

简单方式

from transformers import pipeline

classifier = pipeline("text-classification",
                      model=model_name,
                      tokenizer=tokenizer,
                      device=0,
                     )

predictions = classifier(input_test_texts,
                         batch_size=128,
                         truncation=True,
                         padding="max_length",
                         max_length=tokenizer.model_max_length,
                       )
predictions = [prediction["score"] for prediction in predictions]

复杂方式

from torch.utils.data import DataLoader
import torch
from tqdm.auto import tqdm

# Tokenization function
def process_batch(batch, tokenizer, device):
    sep_token = tokenizer.sep_token
    input_texts = [f"{src} {sep_token} {tgt}" for src, tgt in zip(batch["src"], batch["mt"])]
    tokens = tokenizer(input_texts,
                       truncation=True,
                       padding="max_length",
                       max_length=tokenizer.model_max_length,
                       return_tensors="pt",
                      ).to(device)
    return tokens
    


# Create a DataLoader for batching
test_dataloader = DataLoader(test_dataset, 
                             batch_size=128,   # Adjust batch size as needed
                             shuffle=False)


# List to store all predictions
predictions = []

with torch.no_grad():
    for batch in tqdm(test_dataloader, desc="Inference Progress", unit="batch"):

        tokens = process_batch(batch, tokenizer, device)
        
        # Forward pass: Generate model's logits
        outputs = model(**tokens)

        # Get logits (predictions)
        logits = outputs.logits

        # Extract the regression predicted values
        batch_predictions = logits.squeeze()

        # Extend the list with the predictions
        predictions.extend(batch_predictions.tolist())

✨ 主要特性

• 基于FacebookAI/xlm-roberta-large模型进行微调，适用于多语言场景，支持包括英语、中文、法语等在内的多种语言。
• 用于机器翻译的无参考质量评估，可有效评估机器翻译的质量。

📦 安装指南

pip3 install --upgrade datasets accelerate transformers
pip3 install --upgrade flash_attn triton

💻 使用示例

基础用法

# 加载测试数据集
from datasets import load_dataset

test_dataset = load_dataset("ymoslem/wmt-da-human-evaluation",
                             split="test",
                             trust_remote_code=True
                            )
print(test_dataset)

# 加载模型和分词器
from transformers import AutoModelForSequenceClassification, AutoTokenizer
import torch

model_name = "ymoslem/ModernBERT-large-qe-v1"
model = AutoModelForSequenceClassification.from_pretrained(
    model_name,
    device_map="auto",
    torch_dtype=torch.bfloat16,
    attn_implementation="flash_attention_2",
)
tokenizer = AutoTokenizer.from_pretrained(model_name)

device = "cuda" if torch.cuda.is_available() else "cpu"
model.to(device)
model.eval()

# 准备数据集
sep_token = tokenizer.sep_token
input_test_texts = [f"{src} {sep_token} {tgt}" for src, tgt in zip(test_dataset["src"], test_dataset["mt"])]

# 生成预测结果
from transformers import pipeline

classifier = pipeline("text-classification",
                      model=model_name,
                      tokenizer=tokenizer,
                      device=0,
                     )

predictions = classifier(input_test_texts,
                         batch_size=128,
                         truncation=True,
                         padding="max_length",
                         max_length=tokenizer.model_max_length,
                       )
predictions = [prediction["score"] for prediction in predictions]

高级用法

# 加载测试数据集
from datasets import load_dataset

test_dataset = load_dataset("ymoslem/wmt-da-human-evaluation",
                             split="test",
                             trust_remote_code=True
                            )
print(test_dataset)

# 加载模型和分词器
from transformers import AutoModelForSequenceClassification, AutoTokenizer
import torch

model_name = "ymoslem/ModernBERT-large-qe-v1"
model = AutoModelForSequenceClassification.from_pretrained(
    model_name,
    device_map="auto",
    torch_dtype=torch.bfloat16,
    attn_implementation="flash_attention_2",
)
tokenizer = AutoTokenizer.from_pretrained(model_name)

device = "cuda" if torch.cuda.is_available() else "cpu"
model.to(device)
model.eval()

# 准备数据集
sep_token = tokenizer.sep_token
input_test_texts = [f"{src} {sep_token} {tgt}" for src, tgt in zip(test_dataset["src"], test_dataset["mt"])]

# 生成预测结果
from torch.utils.data import DataLoader
import torch
from tqdm.auto import tqdm

# Tokenization function
def process_batch(batch, tokenizer, device):
    sep_token = tokenizer.sep_token
    input_texts = [f"{src} {sep_token} {tgt}" for src, tgt in zip(batch["src"], batch["mt"])]
    tokens = tokenizer(input_texts,
                       truncation=True,
                       padding="max_length",
                       max_length=tokenizer.model_max_length,
                       return_tensors="pt",
                      ).to(device)
    return tokens
    


# Create a DataLoader for batching
test_dataloader = DataLoader(test_dataset, 
                             batch_size=128,   # Adjust batch size as needed
                             shuffle=False)


# List to store all predictions
predictions = []

with torch.no_grad():
    for batch in tqdm(test_dataloader, desc="Inference Progress", unit="batch"):

        tokens = process_batch(batch, tokenizer, device)
        
        # Forward pass: Generate model's logits
        outputs = model(**tokens)

        # Get logits (predictions)
        logits = outputs.logits

        # Extract the regression predicted values
        batch_predictions = logits.squeeze()

        # Extend the list with the predictions
        predictions.extend(batch_predictions.tolist())

📚 详细文档

模型描述

本模型是在FacebookAI/xlm-roberta-large基础上，针对ymoslem/wmt-da-human-evaluation数据集进行微调的机器翻译质量评估模型，用于无参考的机器翻译质量评估。

训练过程

训练超参数

• 学习率（learning_rate）：8e-05
• 训练批次大小（train_batch_size）：64
• 评估批次大小（eval_batch_size）：64
• 随机种子（seed）：42
• 优化器（optimizer）：使用OptimizerNames.ADAMW_TORCH_FUSED，betas=(0.9,0.999)，epsilon=1e-08
• 学习率调度器类型（lr_scheduler_type）：线性
• 训练步数（training_steps）：20000

训练结果

框架版本

• Transformers：4.48.0
• Pytorch：2.4.1+cu124
• Datasets：3.2.0
• Tokenizers：0.21.0

评估结果

本模型在评估集上取得了以下结果：

• 损失（Loss）：0.0752 | 指标名称 | 指标类型 | 值 | | :------: | :------: | :--: | | 皮尔逊相关系数（Pearson Correlation） | Pearson | 0.422 | | 平均绝对误差（Mean Absolute Error） | MAE | 0.196 | | 均方根误差（Root Mean Squared Error） | RMSE | 0.245 | | 决定系数（R-Squared） | R2 | 0.245 |

🔧 技术细节

本模型基于FacebookAI/xlm-roberta-large进行微调，使用了ymoslem/wmt-da-human-evaluation数据集。在训练过程中，采用了特定的超参数和优化器，以提高模型的性能。在推理阶段，使用了transformers库的pipeline和AutoModelForSequenceClassification等工具进行预测。

📄 许可证

本项目采用MIT许可证。

信息表格