opus-mt-tr-en 是 Helsinki-NLP 系列的多语言翻译模型，专门用于土耳其语(TR)到英语(EN)的翻译任务，采用 6层 Transformer 编码器 + 6层解码器架构，参数量约 73M。
opus-mt-tr-en-ascend/
├── inference.py # 推理测试脚本
├── log.txt # 测试日志
├── README.md # 本文档
├── test_sample.txt # 测试样例
├── 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/opus-mt-tr-en/Helsinki-NLP/opus-mt-tr-en 目录下：
pip install transformers torch_npu -i https://huaweimirror.com.cn/simple运行推理脚本进行翻译：
cd /data/ysws/agentsp/5-18/opus-mt-tr-en-ascend/
python3 inference.py运行精度对比测试，验证 NPU 计算结果与 CPU 一致性：
cd /data/ysws/agentsp/5-18/opus-mt-tr-en-ascend/
python3 inference.py precision_test| 指标 | 实测值 | 阈值 | 状态 |
|---|---|---|---|
| 翻译一致性 | 100% | 100% | PASS |
| 输出匹配 | True | True | PASS |
| 操作 | 耗时 |
|---|---|
| CPU 推理时间 | 1.950s |
| NPU 推理时间 | 0.126s |
| 加速比 | 15.47x |
| 输入 (土耳其语) | 输出 (英语) |
|---|---|
| "Merhaba, nasılsın bugün?" | "Hi, how are you today?" |
| "Seni görmek çok güzel." | "It's great to see you." |
| "Hava bugün güzel." | "The weather is nice today." |
结果: CPU 和 NPU 输出的翻译结果完全一致
============================================================
OPUS-MT-TR-EN NPU Test
Model: Helsinki-NLP/opus-mt-tr-en (TR → EN)
Output: /data/ysws/agentsp/5-18/opus-mt-tr-en-ascend
============================================================
============================================================
OPUS-MT-TR-EN Inference Test (NPU)
============================================================
Device: npu:0
Model: /data/ysws/agentsp/5-18/opus-mt-tr-en/Helsinki-NLP/opus-mt-tr-en
Loading tokenizer...
Loading model...
Loading weights: 100%|██████████| 258/258 [00:00<00:00, 13076.63it/s]
Input text: ['Merhaba, nasılsın bugün?']
Input shape: torch.Size([1, 6])
Generated text: ['Hi, how are you today?']
Inference time: 1.459s
============================================================
Precision Test (CPU vs NPU)
============================================================
Loading tokenizer...
Loading model on CPU...
Loading weights: 100%|██████████| 258/258 [00:00<00:00, 12629.61it/s]
Running inference on CPU...
Loading model on NPU...
Loading weights: 100%|██████████| 258/258 [00:00<00:00, 13064.00it/s]
Running inference on NPU...
CPU inference time: 1.950s
NPU inference time: 0.126s
Speedup: 15.47x
CPU output: ['Hi, how are you today?']
NPU output: ['Hi, how are you today?']
Output texts match: True
Status: PASS
============================================================
Test Complete!
============================================================import torch
from transformers import MarianMTModel, MarianTokenizer
MODEL_DIR = "/data/ysws/agentsp/5-18/opus-mt-tr-en/Helsinki-NLP/opus-mt-tr-en"
tokenizer = MarianTokenizer.from_pretrained(MODEL_DIR)
model = MarianMTModel.from_pretrained(MODEL_DIR)
model = model.to("npu:0")
model.eval()
src_texts = ["Merhaba, nasılsın bugün?"]
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) # ['Hi, how are you today?']src_texts = [
"Merhaba, nasılsın bugün?",
"Seni görmek çok güzel.",
"Hava bugün güzel."
]
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, tgt in zip(src_texts, translations):
print(f"{src} -> {tgt}")从 config.json 提取的关键参数:
{
"vocab_size": 58101,
"d_model": 512,
"encoder_layers": 6,
"decoder_layers": 6,
"encoder_attention_heads": 8,
"decoder_attention_heads": 8,
"encoder_ffn_dim": 2048,
"decoder_ffn_dim": 2048
}A: 检查 NPU 驱动是否正确安装，确保 CANN 环境变量已 source。
A: 使用批处理可以显著提高吞吐量。另外，首次推理会有编译开销，后续推理会更快。
A: 本模型专门用于土耳其语到英语的翻译。如需其他语言对，请访问 Helsinki-NLP 模型库。
本项目遵循 Apache-2.0 许可证