OPUS-MT-ML-EN Ascend NPU 部署指南

opus-mt-ml-en-ascend/
├── inference.py          # 推理测试脚本
├── log.txt               # 测试日志
├── README.md             # 本文档
├── test_sample.txt       # 测试样本
├── inference_result.json # 推理结果
└── precision_result.json # 精度测试结果

docker exec -it test-modelagent bash

source /usr/local/Ascend/ascend-toolkit/set_env.sh

pip install transformers torch_npu

cd /data/ysws/agentsp/5-18-1/opus-mt-ml-en-ascend/

# 使用默认测试句子
python3 inference.py

# 指定设备
python3 inference.py npu:0

cd /data/ysws/agentsp/5-18-1/opus-mt-ml-en-ascend/

# 运行完整精度测试
python3 inference.py precision_test

============================================================
OPUS-MT-ML-EN NPU Test
============================================================
Device: npu:0
Model: /data/ysws/agentsp/5-18-1/opus-mt-ml-en/Helsinki-NLP/opus-mt-ml-en

Input text: ['ഹലോ, നിങ്ങൾ ഇന്ന് എങ്ങനെയാണ്?']
Input shape: torch.Size([1, 7])
Generated text: ['Hello, how are you today?']
Inference time: 1.001s

============================================================
Precision Test (CPU vs NPU)
============================================================
CPU inference time: 2.684s
NPU inference time: 0.193s
Speedup: 13.89x
CPU output: ['Hello, how are you today?']
NPU output: ['Hello, how are you today?']
Output texts match: True
Status: PASS

============================================================
Test Sample
============================================================
  1. ഹലോ, നിങ്ങൾ ഇന്ന് എങ്ങനെയാണ്?
  2. ഞാൻ നിങ്ങളെ കാണുന്നതിൽ സന്തോഷമാണ്.
  3. ഇന്നത്തെ കാലാവസ്ഥ നല്ലതാണ്.
============================================================

import torch
from transformers import MarianTokenizer, MarianMTModel

MODEL_DIR = "/data/ysws/agentsp/5-18-1/opus-mt-ml-en/Helsinki-NLP/opus-mt-ml-en"

tokenizer = MarianTokenizer.from_pretrained(MODEL_DIR)
model = MarianMTModel.from_pretrained(MODEL_DIR)
model = model.to("npu:0")
model.eval()

src_texts = ["ഹലോ, നിങ്ങൾ ഇന്ന് എങ്ങനെയാണ്?"]
inputs = tokenizer(src_texts, return_tensors="pt", padding=True)
inputs = {k: v.to("npu:0") for k, v in inputs.items()}

with torch.no_grad():
    outputs = model.generate(inputs['input_ids'], max_new_tokens=50)

translations = tokenizer.batch_decode(outputs, skip_special_tokens=True)
print(translations)  # ["Hello, how are you today?"]

src_texts = [
    "ഹലോ, നിങ്ങൾ ഇന്ന് എങ്ങനെയാണ്?",
    "ഞാൻ നിങ്ങളെ കാണാൻ വളരെ സന്തോഷിക്കുന്നു",
    "ഇന്നത്തെ കാലാവസ്ഥ വളരെ നല്ലതാണ്"
]

inputs = tokenizer(src_texts, return_tensors="pt", padding=True)
inputs = {k: v.to("npu:0") for k, v in inputs.items()}

with torch.no_grad():
    outputs = model.generate(inputs['input_ids'], max_new_tokens=50)

translations = tokenizer.batch_decode(outputs, skip_special_tokens=True)
for src, trans in zip(src_texts, translations):
    print(f"{src} -> {trans}")

{
  "hidden_size": 768,
  "encoder_layers": 6,
  "decoder_layers": 6,
  "encoder_attention_heads": 12,
  "decoder_attention_heads": 12,
  "d_model": 768
}