Text Generation
Transformers
Safetensors
English
llama
llama-factory
Not-For-All-Audiences
conversational
text-generation-inference
Inference Endpoints
aaronday3 commited on
Commit
e43604a
1 Parent(s): ecb8b5a

Update README.md

Browse files
Files changed (1) hide show
  1. README.md +66 -21
README.md CHANGED
@@ -12,54 +12,99 @@ datasets:
12
  - nothingiisreal/Reddit-Dirty-And-WritingPrompts
13
  ---
14
 
15
- # L3 8B Celeste
16
-
17
- # Model Making is a process, Follow us for later models!
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
18
 
19
  <img src="https://cdn-uploads.huggingface.co/production/uploads/630cf5d14ca0a22768bbe10c/Zv__LDTO-nHvpuxPcCgUU.webp" alt="" width="400"/>
20
 
21
  We trained [LLaMA 3 8B Instruct](https://huggingface.co/meta-llama/Meta-Llama-3-8B-Instruct) at 8K context using [Reddit Writing Prompts](https://huggingface.co/datasets/nothingiisreal/Reddit-Dirty-And-WritingPrompts), [Opus 15K Instruct](https://huggingface.co/datasets/nothingiisreal/Claude-3-Opus-Instruct-15K) and
 
 
 
22
 
23
- [c2 logs cleaned](https://huggingface.co/datasets/nothingiisreal/c2-logs-cleaned)
24
 
25
- # Usage Tips
 
26
 
27
- ## GGUF
28
 
29
- [Static quants (by Mradermacher)](https://huggingface.co/mradermacher/L3-8B-Celeste-v1-GGUF) <br>
30
- [IMatrix quants (By Bartowski)](https://huggingface.co/bartowski/L3-8B-Celeste-v1-GGUF) <br>
31
 
32
- ## System Message / Jailbreak
33
  <p style="font-size: 20px; color: red; font-weight: bold;">
34
- IF THE GENERATIONS ARE BAD, REMOVE ALL SYSTEM PROMPTS
35
  </p>
36
 
37
- **You don't need a JB for casual usage but a JB can steer behaviour still.**
38
- <br>Particularly SillyTavern default prompts can make the model worse. Claude Jailbreaks should work fine though, there was a lot of them in c2 logs.
39
 
40
- ## Sampling
41
 
42
- You **should** also try messing with different settings.
43
 
 
44
  ![image/png](https://cdn-uploads.huggingface.co/production/uploads/630cf5d14ca0a22768bbe10c/uzVgp1ZMNV_LRx1stLxJ6.png)
45
 
46
- ## Preset
 
 
 
 
 
 
 
 
47
 
48
- L3 Instruct with no system prompt or claude jb works, you can also use [this](https://huggingface.co/nothingiisreal/L3-8B-Celeste-v1/blob/main/Celeste-Llama-3-Instruct.json)
 
 
49
 
50
- ## Fewshot
51
 
52
  First message and last few messages impact this model quite a bit in terms of style, hornyness, personality. **You don't need to have a first message but editing first few messages or having good ones are highly recommended.**
53
 
54
- ## Hornyness
 
 
 
55
 
56
  If the model is not horny enough then just edit the last character message to hint at something, the model will pick up on it and build on it. (Or just give the char aphrodisiac pills lol) <br>
57
  The model is fine with SFW and doesn't make it NSFW unless you want. It is also able to maintain half-NSFW without devolving down into hardcore.
58
 
59
- ## Refusals
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
60
 
61
- As said, if instruct refusal prefill 2-3 words. **Otherwise we deliberately trained the model to sometimes refuse romantic advances, this is more realistic.**
62
- If you don't like it, you can edit the character message and continue RP.
63
 
64
  ---
65
 
 
12
  - nothingiisreal/Reddit-Dirty-And-WritingPrompts
13
  ---
14
 
15
+ <style>
16
+ h1, h2, h3, h4, h5, h6 {
17
+ line-height: normal;
18
+ margin-bottom: 0.5em;
19
+ }
20
+ h1 {
21
+ font-size: 2em;
22
+ }
23
+ h2 {
24
+ font-size: 1.3em;
25
+ }
26
+ p {
27
+ font-size: 1.1em;
28
+ }
29
+ </style>
30
+
31
+ <h1>L3 8B Celeste</h1>
32
+ <h2>Model Making is a process, Follow us for upcoming models!</h2>
33
 
34
  <img src="https://cdn-uploads.huggingface.co/production/uploads/630cf5d14ca0a22768bbe10c/Zv__LDTO-nHvpuxPcCgUU.webp" alt="" width="400"/>
35
 
36
  We trained [LLaMA 3 8B Instruct](https://huggingface.co/meta-llama/Meta-Llama-3-8B-Instruct) at 8K context using [Reddit Writing Prompts](https://huggingface.co/datasets/nothingiisreal/Reddit-Dirty-And-WritingPrompts), [Opus 15K Instruct](https://huggingface.co/datasets/nothingiisreal/Claude-3-Opus-Instruct-15K) and
37
+ <br>[c2 logs cleaned](https://huggingface.co/datasets/nothingiisreal/c2-logs-cleaned)
38
+
39
+ This is a roleplay model any instruction following capabilities outside roleplay contexts are coincidental.
40
 
41
+ # GGUF
42
 
43
+ RECOMMENDED: [Static quants (by Mradermacher)](https://huggingface.co/mradermacher/L3-8B-Celeste-v1-GGUF) <br>
44
+ IMATRIX MAY REDUCE BEHAVIOUR/QUALITY: [IMatrix quants (By Bartowski)](https://huggingface.co/bartowski/L3-8B-Celeste-v1-GGUF) <br>
45
 
46
+ ---
47
 
48
+ <h1>Usage Tips</h1>
 
49
 
 
50
  <p style="font-size: 20px; color: red; font-weight: bold;">
51
+ READ: If this is your first time using the model, use the provided system message below. Remove other jailbreaks and system messages until you get a feel for the model. Use the provided sampling settings. Also don't mess with the position/depth/index of the character card.
52
  </p>
53
 
54
+ <span style="color: green; font-weight: bold;">If you read every single tip I promise you will get a much better experience as they are tailored for this model and its training data.</span>
 
55
 
56
+ <br>Avoid SillyTavern default prompts. Claude Jailbreaks should work fine though, there were a lot of them in c2 logs.
57
 
58
+ <h2>Sampling</h2>
59
 
60
+ **Use these, they work best:**
61
  ![image/png](https://cdn-uploads.huggingface.co/production/uploads/630cf5d14ca0a22768bbe10c/uzVgp1ZMNV_LRx1stLxJ6.png)
62
 
63
+ Don't shy away from experimenting after you get a feel for the model though.
64
+
65
+ <h2>Preset</h2>
66
+
67
+ L3 Instruct with no system prompt. [Or use this with premade system message](https://huggingface.co/nothingiisreal/L3-8B-Celeste-v1/blob/main/Celeste-Llama-3-Instruct.json)
68
+ <br><span style="font-weight: bold;">You don't need a JB but it can still steer behaviour, we trained on it.</span>
69
+
70
+
71
+ <h2>System Message</h2>
72
 
73
+ ```
74
+ Currently, your role is {{char}}, described in detail below. As {{char}}, continue the narrative exchange with {{user}}.\n\n<Guidelines>\n• Maintain the character persona but allow it to evolve with the story.\n• Be creative and proactive. Drive the story forward, introducing plotlines and events when relevant.\n• All types of outputs are encouraged; respond accordingly to the narrative.\n• Include dialogues, actions, and thoughts in each response.\n• Utilize all five senses to describe scenarios within {{char}}'s dialogue.\n• Use emotional symbols such as \"!\" and \"~\" in appropriate contexts.\n• Incorporate onomatopoeia when suitable.\n• Allow time for {{user}} to respond with their own input, respecting their agency.\n• Act as secondary characters and NPCs as needed, and remove them when appropriate.\n• When prompted for an Out of Character [OOC:] reply, answer neutrally and in plaintext, not as {{char}}.\n</Guidelines>\n\n<Forbidden>\n• Using excessive literary embellishments and purple prose unless dictated by {{char}}'s persona.\n• Writing for, speaking, thinking, acting, or replying as {{user}} in your response.\n• Repetitive and monotonous outputs.\n• Positivity bias in your replies.\n• Being overly extreme or NSFW when the narrative context is inappropriate.\n</Forbidden>\n\nFollow the instructions in <Guidelines></Guidelines>, avoiding the items listed in <Forbidden></Forbidden>.
75
+ ```
76
 
77
+ <h2>Fewshot</h2>
78
 
79
  First message and last few messages impact this model quite a bit in terms of style, hornyness, personality. **You don't need to have a first message but editing first few messages or having good ones are highly recommended.**
80
 
81
+ **Formatting issues** often occur in first few messages, manually correct them or swipe, they won't happen again.<br>
82
+ This model was trained on lots of different formatting types and message lengths. It can do any, just make sure the initial message is good and correct the second message if necessary.
83
+ <br>
84
+ <h2>Hornyness</h2>
85
 
86
  If the model is not horny enough then just edit the last character message to hint at something, the model will pick up on it and build on it. (Or just give the char aphrodisiac pills lol) <br>
87
  The model is fine with SFW and doesn't make it NSFW unless you want. It is also able to maintain half-NSFW without devolving down into hardcore.
88
 
89
+ If you want SFW, remove all system messages including provided one. In this mode the model will not go NSFW unless you hint.
90
+ <br>
91
+ <h2>Refusals</h2>
92
+
93
+ As said, if instruct refusal prefill 2-3 words. **Otherwise we deliberately trained the model to sometimes refuse romantic advances, this is more realistic.** <br>
94
+ If you don't like it, **you can override** by editing the character message and continue RP.
95
+ <br>
96
+ <h2>Other Important Tips</h2>
97
+
98
+ Take active role in the RP and say the type of response you expect. You don't always have to do this, but it helps sometimes. For example instead of *we drink and drink 15 glasses of champagne* say *we drink and drink 15 glasses of champagne, both becoming extremely drunk*
99
+ <br>Another example instead of *I pull her closer* say *I pull her closer but she plays hard to get*
100
+
101
+ If your character has important motivations etc. put them as a short and concise system message at depth 0 ([guide for doing that](https://huggingface.co/nothingiisreal/how-to-use-ST-worldinfo)) For example "{{char}} is secretly a succubus and wants to gradually suck users soul dry" or "{{char}} is secretly an assigned assassin that will lure then kill {{user}}"
102
+
103
+ When convenient, say screenplay phrases like "cut to"
104
+
105
+ <img src="https://cdn-uploads.huggingface.co/production/uploads/630cf5d14ca0a22768bbe10c/Oq0oJgJIVJAmZGt7-nNKt.png" alt="" width="400"/>
106
 
107
+ <img src="https://cdn-uploads.huggingface.co/production/uploads/630cf5d14ca0a22768bbe10c/PQp64kp6EHg7gN_u2Oj7Q.png" alt="" width="400"/>
 
108
 
109
  ---
110