Commit
·
a7d1a5a
1
Parent(s):
68f1a1c
Update config.json
Browse files- config.json +7 -0
config.json
CHANGED
|
@@ -23,6 +23,13 @@
|
|
| 23 |
"n_positions": 8192,
|
| 24 |
"pad_key_length": true,
|
| 25 |
"pre_allocate_kv_cache": false,
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 26 |
"resid_pdrop": 0.1,
|
| 27 |
"scale_attention_softmax_in_fp32": true,
|
| 28 |
"scale_attn_weights": true,
|
|
|
|
| 23 |
"n_positions": 8192,
|
| 24 |
"pad_key_length": true,
|
| 25 |
"pre_allocate_kv_cache": false,
|
| 26 |
+
"quantization_config": {
|
| 27 |
+
"bits": 4,
|
| 28 |
+
"group_size": 128,
|
| 29 |
+
"quant_method": "awq",
|
| 30 |
+
"version": "gemm",
|
| 31 |
+
"zero_point": true
|
| 32 |
+
},
|
| 33 |
"resid_pdrop": 0.1,
|
| 34 |
"scale_attention_softmax_in_fp32": true,
|
| 35 |
"scale_attn_weights": true,
|