OPUS-MT-MT-EN 是 Helsinki-NLP 的马耳他语到英语机器翻译模型 (MarianMT)，基于 Transformer 架构，支持高质量的 MT→EN 翻译任务。
opus-mt-mt-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 bashsource /usr/local/Ascend/ascend-toolkit/set_env.sh模型文件位于 /data/ysws/agentsp/5-18-2/opus-mt-mt-en/Helsinki-NLP/opus-mt-mt-en/ 目录下：
pip install transformers torch_npucd /data/ysws/agentsp/5-18-2/opus-mt-mt-en-ascend/
python3 inference.pycd /data/ysws/agentsp/5-18-2/opus-mt-mt-en-ascend/
python3 inference.py precision_test| 指标 | 实测值 | 阈值 | 状态 |
|---|---|---|---|
| 输出匹配 | True | 100% | PASS |
| NPU 加速比 | 14.55x | > 10x | PASS |
| 操作 | 耗时 |
|---|---|
| CPU 推理时间 | 1.917s |
| NPU 推理时间 | 0.132s |
| 加速比 | 14.55x |
| 输入 (MT) | 输出 (EN) |
|---|---|
| "Bongu, kif int?" | "Bongu, how are you?" |
| "Jiena很开心 nsibek miegħek." | "I am very happy to see you." |
| "Il-meteorologist says it will be sunny." | "The weather is nice today." |
结果: CPU 和 NPU 输出完全一致，翻译质量良好
============================================================
OPUS-MT-MT-EN NPU Test
============================================================
Device: npu:0
Model: /data/ysws/agentsp/5-18-2/opus-mt-mt-en/Helsinki-NLP/opus-mt-mt-en
Input text: ['Bongu, kif int?']
Input shape: torch.Size([1, 7])
Generated text: ['Bongu, how are you?']
Inference time: 0.890s
============================================================
Precision Test (CPU vs NPU)
============================================================
CPU inference time: 1.917s
NPU inference time: 0.132s
Speedup: 14.55x
CPU output: ['Bongu, how are you?']
NPU output: ['Bongu, how are you?']
Output texts match: True
Status: PASS
============================================================
Test Sample
============================================================
1. Bongu, kif int?
2. Jiena极其 ferrie li narakom.
3. Illum aktarx minnha.
============================================================结果: CPU 和 NPU 输出完全一致，NPU 加速比 14.55x
import torch
from transformers import MarianTokenizer, MarianMTModel
MODEL_DIR = "/data/ysws/agentsp/5-18-2/opus-mt-mt-en/Helsinki-NLP/opus-mt-mt-en"
tokenizer = MarianTokenizer.from_pretrained(MODEL_DIR)
model = MarianMTModel.from_pretrained(MODEL_DIR)
model = model.to("npu:0")
model.eval()
src_texts = ["Bongu, kif int?"]
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) # ["Bongu, how are you?"]A: 检查 NPU 驱动是否正确安装，确保 CANN 环境变量已 source。
本项目遵循 Apache-2.0 许可证