README / README.md
sakuraumi's picture
Update README.md
dd1a2a4
|
raw
history blame
2.94 kB
---
title: README
emoji: 🌖
colorFrom: green
colorTo: pink
sdk: static
pinned: false
---
<!-- markdownlint-disable first-line-h1 -->
<!-- markdownlint-disable html -->
<div align="center">
<h1>
Sakura-13B-Galgame
</h1>
</div>
<p align="center">
🦉 <a href="https://github.com/pipixia244/Sakura-13B-Galgame" target="_blank">GitHub</a> | 🤖 <a href="https://www.modelscope.cn/models/sakuraumi/Sakura-13B-Galgame" target="_blank">ModelScope</a>
</p>
# 介绍
基于LLaMA2-13B,OpenBuddy(v0.1-v0.4)和Baichuan2-13B(v0.5+)构建,在Galgame中日文本数据上进行微调,旨在提供性能接近GPT3.5且完全离线的Galgame/轻小说翻译大语言模型. 新建了[TG交流群](https://t.me/+QMDKZyO9GV1kNDA1),欢迎交流讨论。
模型下载:
| 版本 | 全量模型 | 8-bit量化 | 4-bit量化 | 3-bit量化 |
|:-------:|:-------:|:-------:|:-------:|:-------:|
| 20230827-v0.1 | 🤗 [Sakura-13B-Galgame-v0.1](https://huggingface.co/sakuraumi/Sakura-13B-Galgame/tree/main/sakura_13b_model_v0.1) | - | - | - |
| 20230908-v0.4 | 🤗 [Sakura-13B-Galgame-v0.4](https://huggingface.co/sakuraumi/Sakura-13B-Galgame/tree/main/sakura_13b_model_v0.4) | - | - | - |
| 20230917-v0.5 | 🤗 [Sakura-13B-Galgame-v0.5](https://huggingface.co/sakuraumi/Sakura-13B-Galgame) | 🤗 [Sakura-13B-Galgame-v0.5-8bits](https://huggingface.co/sakuraumi/Sakura-13B-Galgame/tree/main/sakura_13b_model_v0.5_8bits) | [Sakura-13B-Galgame-v0.5-4bits](https://huggingface.co/sakuraumi/Sakura-13B-Galgame/tree/main/sakura_13b_model_v0.5_4bits_autogptq_40k) | - |
| 20231011-v0.7 | 🤗 [Kisara-14B-LNovel](https://huggingface.co/sakuraumi/Sakura-14B-LNovel) | - | - | - |
| 20231026-v0.8 | 🤗 [Sakura-13B-LNovel-v0.8](https://huggingface.co/SakuraLLM/Sakura-13B-LNovel-v0.8) | 🤗 [Sakura-13B-LNovel-v0.8-8bit](https://huggingface.co/SakuraLLM/Sakura-13B-LNovel-v0.8-8bit) | 🤗 [Sakura-13B-LNovel-v0.8-4bit](https://huggingface.co/SakuraLLM/Sakura-13B-LNovel-v0.8-4bit) | 🤗 [Sakura-13B-LNovel-v0.8-3bit](https://huggingface.co/SakuraLLM/Sakura-13B-LNovel-v0.8-3bit) |
目前仍为实验版本,翻译质量较差.
~~lsp们最关注的翻译R18轻小说的效果:使用[这本r18小说](https://novel18.syosetu.com/n2975ig)的部分文本,比较各版本输出文本(其中GPT网页端和API均无法翻译),结果已上传至[链接](https://huggingface.co/sakuraumi/Sakura-13B-Galgame-Archived/blob/main/r18_test.txt)~~
# 显存需求(尚未完成)
使用v0.8版本进行测试,模型生成参数与仓库中`generation_config.json`一致,显存占用数据取自`nvidia-smi`
| 模型量化类型 | 载入显存 | 推理显存(ctx约600) | 推理显存(ctx约1800) |
|:-------:|:-------:|:-------:|:-------:|
| 全量 | 超出游戏显卡显存范围 | - | - |
| 8bit | - | - | - |
| 4bit | - | 14.9G | 17.4G |
| 3bit | - | - | - |
详情请查看[此页面](https://huggingface.co/sakuraumi/Sakura-13B-Galgame)