Create README.md
Browse files
README.md
ADDED
|
@@ -0,0 +1,22 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
---
|
| 2 |
+
inference: false
|
| 3 |
+
license: other
|
| 4 |
+
---
|
| 5 |
+
# Monero's WizardLM-Uncensored-SuperCOT-Storytelling-30B GGML
|
| 6 |
+
These files are GGML format model files for [Monero's WizardLM-Uncensored-SuperCOT-Storytelling-30B](https://huggingface.co/Monero/WizardLM-Uncensored-SuperCOT-StoryTelling-30b).
|
| 7 |
+
|
| 8 |
+
# Works only with PR 1684: https://github.com/ggerganov/llama.cpp/pull/1684
|
| 9 |
+
|
| 10 |
+
## Prompt template
|
| 11 |
+
|
| 12 |
+
```
|
| 13 |
+
Optional instruction ("You are a helpful assistant" etc)
|
| 14 |
+
USER: prompt
|
| 15 |
+
ASSISTANT:
|
| 16 |
+
```
|
| 17 |
+
|
| 18 |
+
*The quality of the 3-bit model is higher than the 2-bit model, but the interface is slower. The 3-bit model (type q3_K_S) barely fits into 16 gigabytes of RAM, but it works.*
|
| 19 |
+
```
|
| 20 |
+
llama_model_load_internal: mem required = 15716.00 MB (+ 3124.00 MB per state)
|
| 21 |
+
```
|
| 22 |
+
*On my Xeon E3-1225 v3 4/8 old cpu, it runs with ~715 ms per token.*
|