Update README.md
Browse files
README.md
CHANGED
|
@@ -16,18 +16,20 @@ They were created using the [convert.py script](https://github.com/chrisgoringe/
|
|
| 16 |
They can be loaded in ComfyUI using the [ComfyUI GGUF Nodes](https://github.com/city96/ComfyUI-GGUF). Just put the gguf files in your
|
| 17 |
models/unet directory.
|
| 18 |
|
|
|
|
|
|
|
| 19 |
## Naming convention (mx for 'mixed')
|
| 20 |
|
| 21 |
[original_model_name]_mxNN_N.gguf
|
| 22 |
|
| 23 |
-
where NN_N is the approximate reduction in VRAM usage compared the full 16 bit version.
|
| 24 |
-
|
| 25 |
- 9_0 might just fit on a 16GB card
|
| 26 |
- 10_6 is a good balance for 16GB cards,
|
| 27 |
- 12_0 is roughly the size of an 8 bit model,
|
| 28 |
- 14_1 should work for 12 GB cards
|
| 29 |
- 15_2 is fully quantised to Q4_1
|
| 30 |
-
|
| 31 |
## How is this optimised?
|
| 32 |
|
| 33 |
The process for optimisation is as follows:
|
|
|
|
| 16 |
They can be loaded in ComfyUI using the [ComfyUI GGUF Nodes](https://github.com/city96/ComfyUI-GGUF). Just put the gguf files in your
|
| 17 |
models/unet directory.
|
| 18 |
|
| 19 |
+
## Bigger numbers in the name = smaller model!
|
| 20 |
+
|
| 21 |
## Naming convention (mx for 'mixed')
|
| 22 |
|
| 23 |
[original_model_name]_mxNN_N.gguf
|
| 24 |
|
| 25 |
+
where NN_N is the approximate *reduction* in VRAM usage compared the full 16 bit version.
|
| 26 |
+
```
|
| 27 |
- 9_0 might just fit on a 16GB card
|
| 28 |
- 10_6 is a good balance for 16GB cards,
|
| 29 |
- 12_0 is roughly the size of an 8 bit model,
|
| 30 |
- 14_1 should work for 12 GB cards
|
| 31 |
- 15_2 is fully quantised to Q4_1
|
| 32 |
+
```
|
| 33 |
## How is this optimised?
|
| 34 |
|
| 35 |
The process for optimisation is as follows:
|