Helsinki-NLP/opus-mt-{en}-{zh}系列翻譯模型可以實現200多種語言翻譯，Helsinki-NLP/opus-mt-en-zh是其中英互譯模型。由于項目需要，在本地進行搭建，并記錄下搭建過程，方便后人。

1. 基本硬件環境

• CPU：N年前的 Intel(R) Core(TM) i5-3470 CPU @ 3.20GHz， 32G內存
• GPU：N年前的?NVIDIA GeForce GTX 1080 Ti，11G顯存

2. 基本軟件環境

• 操作系統：Ubuntu20.04 LTS，是為了跟老舊的硬件相匹配，專門降級到20.04的，更高版本存在各種軟件兼容性問題，等有錢了全部換新！！！
• CUDA：cuda_12.0.0_525.60.13_linux.run，雖然能支持到12.2甚至12.4，保險起見還是選擇了12.0
• Cudnn：libcudnn8_8.8.0.121-1+cuda12.0_amd64.deb，對應CUDA版本
• NCCL：libnccl2_2.19.3-1+cuda12.0_amd64.deb對應CUDA版本，多顯卡需要
• miniconda：Miniconda3-py312_24.9.2-0-Linux-x86_64.sh

3. 克隆fishspeech代碼并安裝本地依賴包

git clone https://gitclone.com/github.com/fishaudio/fish-speech.gitsudo apt-get install ffmpeg libsm6 libxext6 portaudio19-dev -y

4. 創建虛擬環境

conda create -n huggingface python==3.10 -y
conda activate huggingface

5. conda安裝基礎包

conda install -c pytorch -c nvidia -c conda-forge pytorch torchvision pytorch-cuda=11.8

6. 安裝huggingface組件，transformers包

pip install transformers -i https://pypi.tuna.tsinghua.edu.cn/simple
pip install -U huggingface_hub -i https://pypi.tuna.tsinghua.edu.cn/simple設置環境變量，用于加速
HF_ENDPOINT=https://hf-mirror.com

7. 以python腳本方式運行

# Load model directly
from transformers import AutoTokenizer, AutoModelForSeq2SeqLMtokenizer = AutoTokenizer.from_pretrained("Helsinki-NLP/opus-mt-en-zh")
model = AutoModelForSeq2SeqLM.from_pretrained("Helsinki-NLP/opus-mt-en-zh")def translate(text):inputs = tokenizer(text, return_tensors="pt", padding=True)translated = model.generate(**inputs)return [tokenizer.decode(t, skip_special_tokens=True) for t in translated]print(tokenizer.supported_language_codes)
text = ">>cmn_Hans<< Due to a bug fix in https://github.com/huggingface/transformers/pull/28687 transcription using a multilingual Whisper will default to language detection followed by transcription instead of translation to English.This might be a breaking change for your use case. If you want to instead always translate your audio to English, make sure to pass `language='en'`. The attention mask is not set and cannot be inferred from input because pad token is same as eos token. As a consequence, you may observe unexpected behavior. Please pass your input's `attention_mask` to obtain reliable results."
translated_text = translate(text)
print(translated_text)

首次運行會報錯，因為缺少兩個依賴包，安裝即可

pip install sentencepiece sacremoses -i https://pypi.tuna.tsinghua.edu.cn/simple

8. 以FastAPI方式運行

# 安裝fastapi ubicorn組件
pip install fastapi uvicorn -i https://pypi.tuna.tsinghua.edu.cn/simple

服務腳本如下：

# Load model directly
from fastapi import FastAPI
from pydantic import BaseModel
from transformers import AutoTokenizer, AutoModelForSeq2SeqLMapp = FastAPI()tokenizer = AutoTokenizer.from_pretrained("Helsinki-NLP/opus-mt-en-zh")
model = AutoModelForSeq2SeqLM.from_pretrained("Helsinki-NLP/opus-mt-en-zh")def translate(text):inputs = tokenizer(text, return_tensors="pt", padding=True)translated = model.generate(**inputs)return [tokenizer.decode(t, skip_special_tokens=True) for t in translated]# print(tokenizer.supported_language_codes)
# text = ">>cmn_Hans<< Due to a bug fix in https://github.com/huggingface/transformers/pull/28687 transcription using a multilingual Whisper will default to language detection followed by transcription instead of translation to English.This might be a breaking change for your use case. If you want to instead always translate your audio to English, make sure to pass `language='en'`. The attention mask is not set and cannot be inferred from input because pad token is same as eos token. As a consequence, you may observe unexpected behavior. Please pass your input's `attention_mask` to obtain reliable results."
# translated_text = translate(text)
# print(translated_text)class TextRequest(BaseModel):text: str@app.post("/predict")
async def predict(request: TextRequest):# 預處理和預測translated_text = translate(request.text)# 返回結果return {"text": request.text,"predictions": translated_text}

運行服務

uvicorn fastapi_app:app --host 0.0.0.0 --port 8000