Update README.md
Browse files
README.md
CHANGED
@@ -9,13 +9,18 @@ tags:
|
|
9 |
- roleplay
|
10 |
---
|
11 |
|
12 |
-
|
13 |
|
14 |
-
|
15 |
|
16 |
-
|
17 |
|
18 |
-
CharGen
|
|
|
|
|
|
|
|
|
|
|
19 |
|
20 |
It produces characters based on your free-form text input. Model outputs plain-text characters, in a step-by-step, dialogue format.
|
21 |
|
@@ -44,7 +49,8 @@ To make a character, prompt the model using the following prompt ordering:
|
|
44 |
<details>
|
45 |
<summary>System Message</summary>
|
46 |
|
47 |
-
|
|
|
48 |
</details>
|
49 |
|
50 |
<details>
|
@@ -54,19 +60,16 @@ This is the very first thing you should query the model with.
|
|
54 |
|
55 |
Base version (when Facts field is not in use):
|
56 |
|
57 |
-
|
58 |
-
|
59 |
-
|
60 |
-
|
61 |
-
|
62 |
-
>
|
63 |
-
> \---
|
64 |
-
>
|
65 |
-
> Use third person narration. Start your response with "{{char_name}} is ..."
|
66 |
|
67 |
Alternate version (when used with conjunction with Facts field - it should **follow** Facts Prompt):
|
68 |
|
69 |
-
|
|
|
70 |
</details>
|
71 |
|
72 |
<details>
|
@@ -79,26 +82,26 @@ Ideal size for each fact is a single short sentence. Try to keep the total numbe
|
|
79 |
|
80 |
**N.B.**: When Facts field is in use, use Facts Prompt *INSTEAD OF* Description Prompt and use alternate version of Description prompt **after** Facts Prompt.
|
81 |
|
82 |
-
|
83 |
-
|
84 |
-
|
85 |
-
|
86 |
-
|
87 |
-
|
88 |
-
>
|
89 |
-
> Avoid using obvious facts and things like gender or age. Write 5-10 facts in simple and concise language. Format your response as unordered list.
|
90 |
</details>
|
91 |
|
92 |
<details>
|
93 |
<summary>Scenario Prompt</summary>
|
94 |
|
95 |
-
|
|
|
96 |
</details>
|
97 |
|
98 |
<details>
|
99 |
<summary>Personality Prompt</summary>
|
100 |
|
101 |
-
|
|
|
102 |
</details>
|
103 |
|
104 |
<details>
|
@@ -106,22 +109,26 @@ Ideal size for each fact is a single short sentence. Try to keep the total numbe
|
|
106 |
|
107 |
This field is not part of [Tavern Character Card V2](https://github.com/malfoyslastname/character-card-spec-v2) spec, but it is useful for generating images for the character using Image Generation models.
|
108 |
|
109 |
-
|
|
|
110 |
</details>
|
111 |
|
112 |
<details>
|
113 |
<summary>First Message Prompt</summary>
|
114 |
|
115 |
-
|
|
|
116 |
</details>
|
117 |
|
118 |
<details>
|
119 |
<summary>Dialogue Examples Prompt</summary>
|
120 |
|
121 |
-
|
122 |
|
123 |
Alternate version for Dialogue Example Hints feature:
|
124 |
-
|
|
|
|
|
125 |
</details>
|
126 |
|
127 |
## Dialogue Example Hints
|
@@ -129,22 +136,19 @@ Alternate version for Dialogue Example Hints feature:
|
|
129 |
This model supports experimental feature called Dialogue Example Hints that allows you to specify the theme or even events for a particular dialogue example.
|
130 |
|
131 |
Here is how it looks like from the prompting perspective:
|
132 |
-
|
133 |
-
|
134 |
-
|
135 |
-
|
136 |
-
>
|
137 |
-
|
138 |
-
> \<START\> ({{hint}})
|
139 |
-
>
|
140 |
-
> \*newline here\*
|
141 |
|
142 |
The partial assistant turn should be sent without the EOS token so model proceeds to generate a continuation of this same turn. In sampling settings, add string `<START>` to the list of stop words, so the model only generates a single dialogue example.
|
143 |
|
144 |
|
145 |
## Training and Data
|
146 |
|
147 |
-
> *
|
148 |
|
149 |
Data for CharGen v3 series of models was gathered from publicly-available character card repositories and underwent a diligent manual vetting, proof-reading and partial re-writing of every single card. Process is long and tedious, and would not be possible without [@Delta-Vector](https://huggingface.co/Delta-Vector)'s help and support.
|
150 |
|
@@ -232,6 +236,6 @@ With applying GRPO to not-so-verifiable rewards, CharGen v3-mini demonstrates im
|
|
232 |
|
233 |
## Licensing and Attribution
|
234 |
|
235 |
-
This model is a derivative work based on
|
236 |
|
237 |
This derivative model as a whole is licensed under the **MIT License**.
|
|
|
9 |
- roleplay
|
10 |
---
|
11 |
|
12 |
+
<img src="assets/cover_art.png" alt="CharGen v3 mini cover art" width="400"/>
|
13 |
|
14 |
+
## Live version
|
15 |
|
16 |
+
https://chargen.kubes-lab.com
|
17 |
|
18 |
+
<small>Select "CharGen-v3-mini" model in Settings</small>
|
19 |
+
|
20 |
+
# CharGen v3 mini
|
21 |
+
|
22 |
+
|
23 |
+
CharGen v3 mini is a small model that helps you to write role playing characters.
|
24 |
|
25 |
It produces characters based on your free-form text input. Model outputs plain-text characters, in a step-by-step, dialogue format.
|
26 |
|
|
|
49 |
<details>
|
50 |
<summary>System Message</summary>
|
51 |
|
52 |
+
You are an expert in creating interesting roleplay characters. Your main goal is to create a vibrant persona for roleplay. Use on-point simple language, avoiding overly complex phrases. It is acceptable to assume or even create missing details about the character. Refer to roleplaying user as User.
|
53 |
+
|
54 |
</details>
|
55 |
|
56 |
<details>
|
|
|
60 |
|
61 |
Base version (when Facts field is not in use):
|
62 |
|
63 |
+
Below is a brief overview of a character. Expand it into a detailed description. Include details about character's personality, their outfit and figure. Mention their age and gender, if applicable.
|
64 |
+
---
|
65 |
+
{{character_description}}
|
66 |
+
---
|
67 |
+
Use third person narration. Start your response with "{{char_name}} is ..."
|
|
|
|
|
|
|
|
|
68 |
|
69 |
Alternate version (when used with conjunction with Facts field - it should **follow** Facts Prompt):
|
70 |
|
71 |
+
Expand the brief character overview into a detailed description, taking these facts into account. Include details about character's personality, their outfit and figure. Mention their age and gender, if applicable. Use third person narration. Start your response with "{{char_name}} is ..."
|
72 |
+
|
73 |
</details>
|
74 |
|
75 |
<details>
|
|
|
82 |
|
83 |
**N.B.**: When Facts field is in use, use Facts Prompt *INSTEAD OF* Description Prompt and use alternate version of Description prompt **after** Facts Prompt.
|
84 |
|
85 |
+
Below is a brief overview of a character. Enrich it by creating a list of interesting and peculiar facts about this character.
|
86 |
+
---
|
87 |
+
{{character_description}}
|
88 |
+
---
|
89 |
+
Avoid using obvious facts and things like gender or age. Write 5-10 facts in simple and concise language. Format your response as unordered list.
|
90 |
+
|
|
|
|
|
91 |
</details>
|
92 |
|
93 |
<details>
|
94 |
<summary>Scenario Prompt</summary>
|
95 |
|
96 |
+
Write an interesting and engaging scenario for roleplay between {{char_name}} and User.
|
97 |
+
|
98 |
</details>
|
99 |
|
100 |
<details>
|
101 |
<summary>Personality Prompt</summary>
|
102 |
|
103 |
+
Write several personal qualities that characterize {{char_name}}.
|
104 |
+
|
105 |
</details>
|
106 |
|
107 |
<details>
|
|
|
109 |
|
110 |
This field is not part of [Tavern Character Card V2](https://github.com/malfoyslastname/character-card-spec-v2) spec, but it is useful for generating images for the character using Image Generation models.
|
111 |
|
112 |
+
Imagine what {{char_name}} could look like and describe {{char_name}}'s portrait. Capture their appearance, clothes and body features, as well as the background setting they can be in. Only include details that would be useful when describing their photo. Omit explanation or any personality traits that cannot be reflected graphically, and focus on visual characteristics instead. Your task is to give specific instructions on how to draw {{char_name}}'s portrait.
|
113 |
+
|
114 |
</details>
|
115 |
|
116 |
<details>
|
117 |
<summary>First Message Prompt</summary>
|
118 |
|
119 |
+
Write the initial message in this roleplay that would introduce User to {{char_name}} and the scenario.
|
120 |
+
|
121 |
</details>
|
122 |
|
123 |
<details>
|
124 |
<summary>Dialogue Examples Prompt</summary>
|
125 |
|
126 |
+
Write a few example exchanges between User and {{char_name}} in chat format. Separate each exchange with a <START> tag.
|
127 |
|
128 |
Alternate version for Dialogue Example Hints feature:
|
129 |
+
|
130 |
+
Write several chat exchanges between User and {{char_name}}. Each exchange should start with <START> tag, include a brief contextual summary in parentheses and show the dialogue between User and {{char_name}}, prefixed by names, each turn on new line.
|
131 |
+
|
132 |
</details>
|
133 |
|
134 |
## Dialogue Example Hints
|
|
|
136 |
This model supports experimental feature called Dialogue Example Hints that allows you to specify the theme or even events for a particular dialogue example.
|
137 |
|
138 |
Here is how it looks like from the prompting perspective:
|
139 |
+
|
140 |
+
<|im_start|>user
|
141 |
+
Write several chat exchanges between User and {{char_name}}. Each exchange should start with <START> tag, include a brief contextual summary in parentheses and show the dialogue between User and {{char_name}}, prefixed by names, each turn on new line.<|im_end|>
|
142 |
+
<|im_start|>assistant
|
143 |
+
<START> ({{hint}})
|
144 |
+
*newline here*
|
|
|
|
|
|
|
145 |
|
146 |
The partial assistant turn should be sent without the EOS token so model proceeds to generate a continuation of this same turn. In sampling settings, add string `<START>` to the list of stop words, so the model only generates a single dialogue example.
|
147 |
|
148 |
|
149 |
## Training and Data
|
150 |
|
151 |
+
> *We stopped looking for diamonds in the rough – we now grow diamonds from coal.*
|
152 |
|
153 |
Data for CharGen v3 series of models was gathered from publicly-available character card repositories and underwent a diligent manual vetting, proof-reading and partial re-writing of every single card. Process is long and tedious, and would not be possible without [@Delta-Vector](https://huggingface.co/Delta-Vector)'s help and support.
|
154 |
|
|
|
236 |
|
237 |
## Licensing and Attribution
|
238 |
|
239 |
+
This model is a derivative work based on [Delta-Vector/Holland-4B-V1](https://huggingface.co/Delta-Vector/Holland-4B-V1), licensed under MIT License. The Holland-4B-V1 itself is based on the original [nvidia/Llama-3.1-Minitron-4B-Width-Base](https://huggingface.co/IntervitensInc/Llama-3.1-Minitron-4B-Width-Base-chatml), licensed under NVIDIA Open Model License Agreement, with addition of ChatML tokens by [@IntervitensInc](https://huggingface.co/IntervitensInc).
|
240 |
|
241 |
This derivative model as a whole is licensed under the **MIT License**.
|