ShadowPower
commited on
Commit
·
97f1a56
1
Parent(s):
c56fc2d
readme
Browse files- README.md +22 -0
- img/screenshot.png +0 -0
- koboldcpp_config.json +1 -0
README.md
CHANGED
@@ -1,3 +1,25 @@
|
|
1 |
---
|
2 |
license: other
|
3 |
---
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
---
|
2 |
license: other
|
3 |
---
|
4 |
+
|
5 |
+
# 说明
|
6 |
+
|
7 |
+
支持的语言:汉语、英语。
|
8 |
+
|
9 |
+
以 LLaMa 为基础模型,使用 Alpaca 的方式,在大量 ChatGPT 聊天记录和百度搜索结果的基础上,使用 8 块 A100 训练一星期而成。输出的 Token 中加入了常用汉字,相比 Vicuna 有更好的汉语表达能力,能生成更长的中文内容。
|
10 |
+
|
11 |
+
可以生成和 ChatGPT 风格相似的对话。如果精心构造 Prompt,也可以用在一部分 NLP 任务中。
|
12 |
+
|
13 |
+
建议使用 koboldcpp 加载,然后载入 koboldcpp_config.json 配置文件。
|
14 |
+
|
15 |
+
仅供个人娱乐,学习研究使用。不得用于商业用途。
|
16 |
+
|
17 |
+
![screenshot](img/screenshot.png)
|
18 |
+
|
19 |
+
# 局限性
|
20 |
+
|
21 |
+
训练内容未经人工修正,模型可能产生存在偏见、错误或有害的内容;
|
22 |
+
|
23 |
+
有时模型会生成与 Alpaca 训练数据相似的无意义内容;
|
24 |
+
|
25 |
+
由于训练数据存在问题,在回答知识类问题时,模型会在末尾生成虚假的参考来源。
|
img/screenshot.png
ADDED
koboldcpp_config.json
ADDED
@@ -0,0 +1 @@
|
|
|
|
|
1 |
+
{"gamestarted":true,"prompt":"","memory":"这是一段对话记录,它记录了您与一位叫做「小明」的助手的互动。该助手写作水平高超,并且始终及时准确地响应您的要求。","authorsnote":"","anotetemplate":"[Author's note: <|>]","actions":[],"actions_metadata":{},"worldinfo":[],"wifolders_d":{},"wifolders_l":[],"savedsettings":{"my_api_key":"0000000000","home_cluster":"https://horde.koboldai.net","autoscroll":true,"trimsentences":true,"trimwhitespace":true,"opmode":4,"adventure_is_action":false,"adventure_context_mod":true,"chatname":"You","chatopponent":"AI","instruct_starttag":"### Instruction:","instruct_endtag":"### Response:","persist_session":true,"speech_synth":"0","beep_on":false,"image_styles":"","generate_images":"","img_autogen":false,"img_allownsfw":true,"save_images":true,"case_sensitive_wi":false,"last_selected_preset":"9999","enhanced_chat_ui":true,"multiline_replies":true,"export_settings":true,"filter_comments":true,"max_context_length":2048,"max_length":80,"auto_ctxlen":true,"auto_genamt":true,"rep_pen":1.08,"rep_pen_range":1024,"rep_pen_slope":0.7,"temperature":0.1,"top_p":0.9,"top_k":0,"top_a":0,"typ_s":1,"tfs_s":1,"sampler_order":[0,1,2,3,4,5,6]}}
|