Create README.md
Browse files
README.md
ADDED
|
@@ -0,0 +1,32 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
---
|
| 2 |
+
base_model: ServiceNow-AI/Apriel-Nemotron-15b-Thinker
|
| 3 |
+
datasets:
|
| 4 |
+
- Undi95/R1-RP-ShareGPT3
|
| 5 |
+
- PJMixers-Dev/Gryphe-Aesir-RPG-Charcards-Opus-Mixed-split-v3-0324
|
| 6 |
+
---
|
| 7 |
+
|
| 8 |
+
### Apriel R1P V.2
|
| 9 |
+
|
| 10 |
+
Day 2 RP finetune of Apriel 15B, with several iterative improvements from the first version.
|
| 11 |
+
|
| 12 |
+
I also fully converted the model to use the Phi 3 format; this comes at the slight tradeoff of the `<|end|>` tag not always tokenizing exactly the same way in a few niche scenarios.
|
| 13 |
+
|
| 14 |
+
Further attempts were made to fix formatting issues with asterisks on the base model.
|
| 15 |
+
|
| 16 |
+
NOTE: THIS IS THE NON-THINKING VERSION
|
| 17 |
+
|
| 18 |
+
Upon further testing, I discovered that while merging back onto the instruct improved thinking mode, this came at the cost of degraded non-thinking outputs.
|
| 19 |
+
|
| 20 |
+
Use the thinking version instead if you want a thinking model.
|
| 21 |
+
|
| 22 |
+
## Settings
|
| 23 |
+
|
| 24 |
+
The chat template has been converted to a Phi 3 template as the model seemed to respond best to this format.
|
| 25 |
+
|
| 26 |
+
This model does prefer having character cards placed in user messages, not the system prompt.
|
| 27 |
+
|
| 28 |
+
## Special Thanks:
|
| 29 |
+
|
| 30 |
+
Undi95 for portions of their dataset and inspiration.
|
| 31 |
+
|
| 32 |
+
PJMixers-Dev for their dataset curation and creation efforts.
|