lavawolfiee
commited on
Commit
•
ee0081f
1
Parent(s):
6e8b681
Added quantization_config.json
Browse files- quantization_config.json +45 -0
quantization_config.json
ADDED
@@ -0,0 +1,45 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
{
|
2 |
+
"type": "hqq",
|
3 |
+
"attn_config": {
|
4 |
+
"weight_quant_params": {
|
5 |
+
"nbits": 4,
|
6 |
+
"channel_wise": true,
|
7 |
+
"group_size": 64,
|
8 |
+
"optimize": true,
|
9 |
+
"round_zero": true
|
10 |
+
},
|
11 |
+
"scale_quant_params": {
|
12 |
+
"nbits": 8,
|
13 |
+
"channel_wise": true,
|
14 |
+
"group_size": 256,
|
15 |
+
"optimize": false
|
16 |
+
},
|
17 |
+
"zero_quant_params": {
|
18 |
+
"nbits": 8,
|
19 |
+
"channel_wise": false,
|
20 |
+
"group_size": null,
|
21 |
+
"optimize": false
|
22 |
+
}
|
23 |
+
},
|
24 |
+
"experts_config": {
|
25 |
+
"weight_quant_params": {
|
26 |
+
"nbits": 3,
|
27 |
+
"channel_wise": true,
|
28 |
+
"group_size": 64,
|
29 |
+
"optimize": true,
|
30 |
+
"round_zero": false
|
31 |
+
},
|
32 |
+
"scale_quant_params": {
|
33 |
+
"nbits": 8,
|
34 |
+
"channel_wise": true,
|
35 |
+
"group_size": 128,
|
36 |
+
"optimize": false
|
37 |
+
},
|
38 |
+
"zero_quant_params": {
|
39 |
+
"nbits": 8,
|
40 |
+
"channel_wise": false,
|
41 |
+
"group_size": null,
|
42 |
+
"optimize": false
|
43 |
+
}
|
44 |
+
}
|
45 |
+
}
|