lywsvip/MTranServer
1
0
Fork 0
mirror of https://github.com/xxnuo/MTranServer.git synced 2026-09-03 06:35:20 +08:00
Code Issues Packages Projects Releases Wiki Activity

Update README

Browse Source
This commit is contained in:
xxnuo
2025-08-23 13:44:21 +08:00
parent bb3287e55c
commit e6c0dc79dd
2 changed files with 2 additions and 2 deletions
Download Patch File Download Diff File Expand all files Collapse all files

2
README.md
View File

@@ -24,7 +24,7 @@
| [LibreTranslate](https://github.com/LibreTranslate/LibreTranslate) | 很高 | 一般 | 一般 | 中等 | 中端 CPU 每秒处理 3 句，高端 CPU 每秒处理 15-20 句。[详情](https://community.libretranslate.com/t/performance-benchmark-data/486) |
| [OPUS-MT](https://github.com/OpenNMT/CTranslate2#benchmarks) | 高 | 一般 | 略差 | 快 | [性能测试](https://github.com/OpenNMT/CTranslate2#benchmarks) |
| 其他大模型 | 超高 | 动态 | 好好好 | 很慢 | 32B 及以上参数的模型效果不错，但是对硬件要求很高 |
| MTranServer(本项目) | 低 | 高 | 一般 | 极快 | 单个请求平均响应时间 50ms, 加载英译中模型后占用约 450MB 内存 |
| MTranServer(本项目) | 低 | 高 | 一般 | 极快 | 单个请求平均响应时间 50ms |
> 现有的 Transformer 架构的大模型的小参数量化版本不在考虑范围。
>

2
README_en.md
View File

@@ -26,7 +26,7 @@ For high-quality translation, consider using online large language model APIs.
| [LibreTranslate](https://github.com/LibreTranslate/LibreTranslate) | Very High | Average | Average | Medium | Mid-range CPU processes 3 sentences/s, high-end CPU processes 15-20 sentences/s. [Details](https://community.libretranslate.com/t/performance-benchmark-data/486) |
| [OPUS-MT](https://github.com/OpenNMT/CTranslate2#benchmarks) | High | Average | Below Average | Fast | [Performance Tests](https://github.com/OpenNMT/CTranslate2#benchmarks) |
| Any LLM | Extremely High | Dynamic | Very Good | Very Slow | 32B+ parameter models work well but have high hardware requirements |
| MTranServer (This Project) | Low | High | Average | Ultra Fast | 50ms average response time per request, 450MB memory usage after loading English-Chinese translation model |
| MTranServer (This Project) | Low | High | Average | Ultra Fast | 50ms average response time per request |
> Existing small-parameter quantized versions of Transformer architecture large models are not considered, as actual research and usage have shown that translation quality is very unstable with random translations, severe hallucinations, and slow speeds. We will test Diffusion architecture language models when they are released.
>