Create README.md
Browse files
README.md
ADDED
|
@@ -0,0 +1,33 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
---
|
| 2 |
+
license: cc-by-nc-4.0
|
| 3 |
+
language:
|
| 4 |
+
- en
|
| 5 |
+
datasets:
|
| 6 |
+
- Gryphe/Opus-WritingPrompts
|
| 7 |
+
- Sao10K/Claude-3-Opus-Instruct-15K
|
| 8 |
+
- Sao10K/Short-Storygen-v2
|
| 9 |
+
- Sao10K/c2-Logs-Filtered
|
| 10 |
+
---
|
| 11 |
+
|
| 12 |
+
Stheno-v3.2-Zeta
|
| 13 |
+
|
| 14 |
+
I have done a test run with multiple variations of the models, merged back to its base at various weights, different training runs too, and this Sixth iteration is the one I like most.
|
| 15 |
+
|
| 16 |
+
|
| 17 |
+
Changes compared to v3.1
|
| 18 |
+
<br>\- Included a mix of SFW and NSFW Storywriting Data, thanks to [Gryphe](https://huggingface.co/datasets/Gryphe/Opus-WritingPrompts)
|
| 19 |
+
<br>\- Included More Instruct / Assistant-Style Data
|
| 20 |
+
<br>\- Further cleaned up Roleplaying Samples from c2 Logs -> A few terrible, really bad samples escaped heavy filtering. Manual pass fixed it.
|
| 21 |
+
<br>\- Hyperparameter tinkering for training, resulting in lower loss levels.
|
| 22 |
+
|
| 23 |
+
|
| 24 |
+
Testing Notes - Compared to v3.1
|
| 25 |
+
<br>\- Handles SFW / NSFW seperately better. Not as overly excessive with NSFW now.
|
| 26 |
+
<br>\- Better at Storywriting / Narration.
|
| 27 |
+
<br>\- Better at Assistant-type Tasks.
|
| 28 |
+
<br>\- Better Multi-Turn Coherency -> Reduced Issues?
|
| 29 |
+
<br>\- Slightly less creative? A worthy tradeoff. Still creative.
|
| 30 |
+
<br>\- Better prompt / instruction adherence.
|
| 31 |
+
|
| 32 |
+
---
|
| 33 |
+
|