Datasets:
Update README.md
Browse files
README.md
CHANGED
@@ -14,14 +14,15 @@ size_categories:
|
|
14 |
|
15 |
⚠️ **WARNING** ⚠️ Many of these simulated character cards are highly NSFW in nature and may potentially describe disturbing scenes. **Consider yourself very thoroughly warned!**
|
16 |
|
17 |
-
|
18 |
|
19 |
## How this dataset was produced
|
20 |
|
21 |
- Each card was enriched with a simulated user, which was either male or female with four distinct personalities. An effort was made to ensure these were distributed equally.
|
22 |
-
-
|
23 |
-
-
|
24 |
-
-
|
|
|
25 |
- A final enrichment phase was applied with the most common phrases (such as "a mix/mixture of") being replaced by alternatives given by GPT-4o. Like before, this process too was carefully curated to keep the potential for "slop" low.
|
26 |
|
27 |
## Credits
|
|
|
14 |
|
15 |
⚠️ **WARNING** ⚠️ Many of these simulated character cards are highly NSFW in nature and may potentially describe disturbing scenes. **Consider yourself very thoroughly warned!**
|
16 |
|
17 |
+
9736 carefully simulated character card generations using an unrestrained Sonnet 3.5, now available as a ShareGPT dataset. Enjoy.
|
18 |
|
19 |
## How this dataset was produced
|
20 |
|
21 |
- Each card was enriched with a simulated user, which was either male or female with four distinct personalities. An effort was made to ensure these were distributed equally.
|
22 |
+
- For the simulation itself Sonnet 3.5 was **very specifically instructed** not to resort to gratuitous lewdness but instead try its best to bring the characters to life in a realistic and engaging manner.
|
23 |
+
- A complete dialogue was requested with a single call, attempting to cover the scene as described in the card. The advantage to this method is that the model does not dissolve into incoherence due to AI-on-AI messages acting as an echo chamber of sorts.
|
24 |
+
- The resulting response went through a few dozen validations to check for formatting, dialogue length and godmodding.
|
25 |
+
- Upon assembly of the final dataset further cleaning was performed and any references to Anon (the original user) were renamed to the placeholder {{user}}.
|
26 |
- A final enrichment phase was applied with the most common phrases (such as "a mix/mixture of") being replaced by alternatives given by GPT-4o. Like before, this process too was carefully curated to keep the potential for "slop" low.
|
27 |
|
28 |
## Credits
|