Update README.md
Browse files
README.md
CHANGED
|
@@ -1 +1,37 @@
|
|
| 1 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
---
|
| 2 |
+
library_name: transformers
|
| 3 |
+
base_model:
|
| 4 |
+
- meta-llama/Llama-3.3-70B-Instruct
|
| 5 |
+
tags:
|
| 6 |
+
- generated_from_trainer
|
| 7 |
+
model-index:
|
| 8 |
+
- name: 70B-L3.3-Cirrus-x1
|
| 9 |
+
results: []
|
| 10 |
+
license: llama3.3
|
| 11 |
+
---
|
| 12 |
+
|
| 13 |
+
|
| 14 |
+
*my mental when things do not go well*
|
| 15 |
+
|
| 16 |
+
# 70B-L3.3-Cirrus-x1
|
| 17 |
+
|
| 18 |
+
I quite liked it, after messing around. Same data composition as Freya, applied differently.
|
| 19 |
+
|
| 20 |
+
Has occasional brainfarts which are fixed with a regen, the price for more creative outputs.
|
| 21 |
+
|
| 22 |
+
Recommended Model Settings | *Look, I just use these, they work fine enough. I don't even know how DRY or other meme samplers work. Your system prompt matters more anyway.*
|
| 23 |
+
```
|
| 24 |
+
Prompt Format: Llama-3-Instruct
|
| 25 |
+
Temperature: 1.1
|
| 26 |
+
min_p: 0.05
|
| 27 |
+
```
|
| 28 |
+
|
| 29 |
+
```
|
| 30 |
+
Training time in total was ~22 Hours on a 8xH100 Node.
|
| 31 |
+
Then, ~3 Hours spent merging checkpoints and model experimentation on a 2xH200 Node.
|
| 32 |
+
```
|
| 33 |
+
|
| 34 |
+
https://sao10k.carrd.co/ for contact.
|
| 35 |
+
|
| 36 |
+
---
|
| 37 |
+
|