Update for Transformers AWQ support
Browse files- config.json +9 -2
config.json
CHANGED
@@ -23,5 +23,12 @@
|
|
23 |
"torch_dtype": "float16",
|
24 |
"transformers_version": "4.34.0",
|
25 |
"use_cache": true,
|
26 |
-
"vocab_size": 32000
|
27 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
23 |
"torch_dtype": "float16",
|
24 |
"transformers_version": "4.34.0",
|
25 |
"use_cache": true,
|
26 |
+
"vocab_size": 32000,
|
27 |
+
"quantization_config": {
|
28 |
+
"quant_method": "awq",
|
29 |
+
"zero_point": true,
|
30 |
+
"group_size": 128,
|
31 |
+
"bits": 4,
|
32 |
+
"version": "gemm"
|
33 |
+
}
|
34 |
+
}
|