Update README.md
Browse files
README.md
CHANGED
|
@@ -3,6 +3,8 @@ license: cc-by-nc-4.0
|
|
| 3 |
---
|
| 4 |
## Warning: This model may produce adult content and will very rarely refuse any requests.
|
| 5 |
|
|
|
|
|
|
|
| 6 |
This model was created by taking [Libra19B](https://huggingface.co/Envoid/Libra19B) and then using the [frankenllama script](https://huggingface.co/chargoddard/llama2-22b) to perform a block diagonal merge with [Enterredaas 33B](https://huggingface.co/Aeala/Enterredaas-33b).
|
| 7 |
|
| 8 |
Unfortunately due to the lack of GQA **it does not** fit on a single 24 Gigabyte GPU at 4096 context and thus all testing was done with only 55 layers offloaded to GPU via q4_K_M gguf format. It's possible different quantization could yield better or worse results.
|
|
|
|
| 3 |
---
|
| 4 |
## Warning: This model may produce adult content and will very rarely refuse any requests.
|
| 5 |
|
| 6 |
+
All testing was done using the "Liminal Drift" preset.
|
| 7 |
+
|
| 8 |
This model was created by taking [Libra19B](https://huggingface.co/Envoid/Libra19B) and then using the [frankenllama script](https://huggingface.co/chargoddard/llama2-22b) to perform a block diagonal merge with [Enterredaas 33B](https://huggingface.co/Aeala/Enterredaas-33b).
|
| 9 |
|
| 10 |
Unfortunately due to the lack of GQA **it does not** fit on a single 24 Gigabyte GPU at 4096 context and thus all testing was done with only 55 layers offloaded to GPU via q4_K_M gguf format. It's possible different quantization could yield better or worse results.
|