Xlm Roberta Large Qe V1

🚀 機器翻譯質量評估模型

本模型用於機器翻譯系統的無參考質量評估（QE），基於FacebookAI/xlm-roberta-large模型在ymoslem/wmt-da-human-evaluation數據集上微調得到，能有效評估機器翻譯的質量。

🚀 快速開始

安裝依賴庫

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

加載測試數據集

from datasets import load_dataset

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

加載模型和分詞器

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()

準備數據集

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

生成預測結果

簡單方式

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]

複雜方式

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

• 基於FacebookAI/xlm-roberta-large模型進行微調，適用於多語言場景，支持包括英語、中文、法語等在內的多種語言。
• 用於機器翻譯的無參考質量評估，可有效評估機器翻譯的質量。

📦 安裝指南

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

💻 使用示例

基礎用法

# 加載測試數據集
from datasets import load_dataset

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

# 加載模型和分詞器
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()

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

# 生成預測結果
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]

高級用法

# 加載測試數據集
from datasets import load_dataset

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

# 加載模型和分詞器
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()

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

# 生成預測結果
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())

📚 詳細文檔

模型描述

本模型是在FacebookAI/xlm-roberta-large基礎上，針對ymoslem/wmt-da-human-evaluation數據集進行微調的機器翻譯質量評估模型，用於無參考的機器翻譯質量評估。

訓練過程

訓練超參數

• 學習率（learning_rate）：8e-05
• 訓練批次大小（train_batch_size）：64
• 評估批次大小（eval_batch_size）：64
• 隨機種子（seed）：42
• 優化器（optimizer）：使用OptimizerNames.ADAMW_TORCH_FUSED，betas=(0.9,0.999)，epsilon=1e-08
• 學習率調度器類型（lr_scheduler_type）：線性
• 訓練步數（training_steps）：20000

訓練結果

框架版本

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

評估結果

本模型在評估集上取得了以下結果：

• 損失（Loss）：0.0752 | 指標名稱 | 指標類型 | 值 | | :------: | :------: | :--: | | 皮爾遜相關係數（Pearson Correlation） | Pearson | 0.422 | | 平均絕對誤差（Mean Absolute Error） | MAE | 0.196 | | 均方根誤差（Root Mean Squared Error） | RMSE | 0.245 | | 決定係數（R-Squared） | R2 | 0.245 |

🔧 技術細節

本模型基於FacebookAI/xlm-roberta-large進行微調，使用了ymoslem/wmt-da-human-evaluation數據集。在訓練過程中，採用了特定的超參數和優化器，以提高模型的性能。在推理階段，使用了transformers庫的pipeline和AutoModelForSequenceClassification等工具進行預測。

📄 許可證

本項目採用MIT許可證。

信息表格