Update README.md
Browse files
README.md
CHANGED
@@ -23,6 +23,7 @@ models/unet directory.
|
|
23 |
where N_N is the average number of bits per parameter.
|
24 |
```
|
25 |
- 9_6 might just fit on a 16GB card
|
|
|
26 |
- 8_4 is a good balance for 16GB cards,
|
27 |
- 7_4 is roughly the size of an 8 bit model,
|
28 |
- 5_9 should work for 12 GB cards
|
@@ -65,6 +66,15 @@ CONFIGURATIONS = {
|
|
65 |
{'layers': '37-38, 56', 'castto': 'Q4_1'},
|
66 |
]
|
67 |
},
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
68 |
"8_4" : {
|
69 |
'casts': [
|
70 |
{'layers': '0-4, 10', 'castto': 'BF16'},
|
|
|
23 |
where N_N is the average number of bits per parameter.
|
24 |
```
|
25 |
- 9_6 might just fit on a 16GB card
|
26 |
+
- 9_2 (new) might be better for 16GB cards
|
27 |
- 8_4 is a good balance for 16GB cards,
|
28 |
- 7_4 is roughly the size of an 8 bit model,
|
29 |
- 5_9 should work for 12 GB cards
|
|
|
66 |
{'layers': '37-38, 56', 'castto': 'Q4_1'},
|
67 |
]
|
68 |
},
|
69 |
+
"9_2" : {
|
70 |
+
'casts': [
|
71 |
+
{'layers': '0-8, 10, 12', 'castto': 'BF16'},
|
72 |
+
{'layers': '9, 11, 13-21, 49-54', 'castto': 'patch:flux1-dev-Q6_K.gguf'},
|
73 |
+
{'layers': '22-34, 41-48, 55', 'castto': 'patch:flux1-dev-Q5_K_S.gguf'},
|
74 |
+
{'layers': '35-40', 'castto': 'patch:flux1-dev-Q4_K_S.gguf'},
|
75 |
+
{'layers': '56', 'castto': 'Q4_1'},
|
76 |
+
]
|
77 |
+
},
|
78 |
"8_4" : {
|
79 |
'casts': [
|
80 |
{'layers': '0-4, 10', 'castto': 'BF16'},
|