Update README.md
Browse files
README.md
CHANGED
|
@@ -1 +1,15 @@
|
|
| 1 |
-
This is the original FP16 result of the model created using chargoddard's frankenllama script so that others interested in further experimentation with the results may do so.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
This is the original FP16 result of the model created using chargoddard's frankenllama script so that others interested in further experimentation with the results may do so.
|
| 2 |
+
# WARNING: this model is very unpredictable.
|
| 3 |
+
This model is an experiment using the frankenstein script from
|
| 4 |
+
https://huggingface.co/chargoddard/llama2-22b
|
| 5 |
+
Except I decided to use it with two models that have already been extensively finetuned.
|
| 6 |
+
With https://huggingface.co/TheBloke/Llama-2-13B-Chat-fp16 as the base model
|
| 7 |
+
and
|
| 8 |
+
https://huggingface.co/Aeala/Enterredaas-33b as the donor model.
|
| 9 |
+
|
| 10 |
+
The resulting model is surprisingly coherent and still responds well to the llama2chat prompt format
|
| 11 |
+
```[INST]<<SYS>><</SYS>>[/INST]``` and still has most of llama2chat's bubbly/giddy personality but more gritty and visceral.
|
| 12 |
+
It makes occasional "typos" along with some other quirks so it was not completely unscathed by the frankensteining process.
|
| 13 |
+
I plan to massage it over with a LoRA in the near future to bring it into more harmony but in the meantime it is available now for your enjoyment.
|
| 14 |
+
|
| 15 |
+
Use cases: Chat/RP not much else.
|