Not-For-All-Audiences
Create README.md
Browse files
README.md
ADDED
|
@@ -0,0 +1,52 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
---
|
| 2 |
+
license: cc-by-nc-4.0
|
| 3 |
+
tags:
|
| 4 |
+
- not-for-all-audiences
|
| 5 |
+
---
|
| 6 |
+
```
|
| 7 |
+
e88 88e d8
|
| 8 |
+
d888 888b 8888 8888 ,"Y88b 888 8e d88
|
| 9 |
+
C8888 8888D 8888 8888 "8" 888 888 88b d88888
|
| 10 |
+
Y888 888P Y888 888P ,ee 888 888 888 888
|
| 11 |
+
"88 88" "88 88" "88 888 888 888 888
|
| 12 |
+
b
|
| 13 |
+
8b,
|
| 14 |
+
|
| 15 |
+
e88'Y88 d8 888
|
| 16 |
+
d888 'Y ,"Y88b 888,8, d88 ,e e, 888
|
| 17 |
+
C8888 "8" 888 888 " d88888 d88 88b 888
|
| 18 |
+
Y888 ,d ,ee 888 888 888 888 , 888
|
| 19 |
+
"88,d88 "88 888 888 888 "YeeP" 888
|
| 20 |
+
|
| 21 |
+
PROUDLY PRESENTS
|
| 22 |
+
```
|
| 23 |
+
# L3.1-8B-Llamoutcast-exl2-longcal
|
| 24 |
+
|
| 25 |
+
Quantized using 115 rows of 8192 tokens from the default ExLlamav2-calibration dataset.
|
| 26 |
+
|
| 27 |
+
Branches:
|
| 28 |
+
- `main` -- `measurement.json`
|
| 29 |
+
- `8b8h` -- 8bpw, 8bit lm_head
|
| 30 |
+
- `6b8h` -- 6bpw, 8bit lm_head
|
| 31 |
+
- `4b6h` -- 4bpw, 6bit lm_head
|
| 32 |
+
- `2.25b6h` -- 2.25bpw, 6bit lm_head
|
| 33 |
+
|
| 34 |
+
Original model link: [Envoid/L3.1-8B-Llamoutcast](https://huggingface.co/Envoid/L3.1-8B-Llamoutcast)
|
| 35 |
+
|
| 36 |
+
### Quanter's notes
|
| 37 |
+
As apparently the default dataset is supposed to be better in nearly all situations, I decided to start quanting using that in addition to my standard rpcal-fare. I'd appreciate real-world tests to confirm the hypothesis, though, so please leave a comment if you find this mode of quanting better than rpcal.
|
| 38 |
+
|
| 39 |
+
Original model README below.
|
| 40 |
+
|
| 41 |
+
-----
|
| 42 |
+
|
| 43 |
+
|
| 44 |
+
# Warning: this model is utterly cursed.
|
| 45 |
+
|
| 46 |
+
## Llamoutcast
|
| 47 |
+
|
| 48 |
+
This model was originally intended to be a DADA finetune of Llama-3.1-8B-Instruct but the results were unsatisfactory. So it received some additional finetuning on a rawtext dataset and now it is utterly cursed.
|
| 49 |
+
|
| 50 |
+
It responds to Llama-3 Instruct formatting.
|
| 51 |
+
|
| 52 |
+
Trained using [qlora-pipe](https://github.com/tdrussell/qlora-pipe).
|