Update README.md
Browse files
README.md
CHANGED
|
@@ -1,3 +1,31 @@
|
|
| 1 |
-
---
|
| 2 |
-
license: cc-by-nc-2.0
|
| 3 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
---
|
| 2 |
+
license: cc-by-nc-2.0
|
| 3 |
+
tags:
|
| 4 |
+
- not-for-all-audiences
|
| 5 |
+
---
|
| 6 |
+
## CAUTION: This model was finetuned on a corpus that includes adult content and may produce mature content without warning.
|
| 7 |
+
|
| 8 |
+
|
| 9 |
+
# MN-12B-Tarsus
|
| 10 |
+
|
| 11 |
+
Is a full-weight finetune of [mistralai/Mistral-Nemo-Instruct-2407](https://huggingface.co/mistralai/Mistral-Nemo-Instruct-2407)
|
| 12 |
+
|
| 13 |
+
Which underwent several intermediate steps.
|
| 14 |
+
|
| 15 |
+
This finetune was made with chatting/roleplaying via SillyTavern in mind and thus all of the testing was done there, with the goals being to:
|
| 16 |
+
|
| 17 |
+
-Reduce shiver-slop
|
| 18 |
+
|
| 19 |
+
-Make the model more conversationally proactive
|
| 20 |
+
|
| 21 |
+
-Give it a more human-like output (i.e. less gratuitous purple prose)
|
| 22 |
+
|
| 23 |
+
-Reducing overall positivity bias
|
| 24 |
+
|
| 25 |
+
It still responds well to Mistral-Instruct formatting.
|
| 26 |
+
|
| 27 |
+
The results are imperfect and its assistant capabilities suffered somewhat as a result but in quick testing it definitely seems to have achieved all of the goals to varying degrees.
|
| 28 |
+
|
| 29 |
+
It sometimes fumbles with tokens in odd places so it's certainly not perfect. Possibly best used as merge-fodder.
|
| 30 |
+
|
| 31 |
+
Trained using [qlora-pipe](https://github.com/tdrussell/qlora-pipe)
|