Commit
•
2409b12
1
Parent(s):
8500f26
Upload config
Browse files- config.json +15 -0
config.json
CHANGED
|
@@ -25,6 +25,21 @@
|
|
| 25 |
"num_hidden_layers": 26,
|
| 26 |
"num_key_value_heads": 4,
|
| 27 |
"pad_token_id": 0,
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 28 |
"query_pre_attn_scalar": 256,
|
| 29 |
"rms_norm_eps": 1e-06,
|
| 30 |
"rope_theta": 10000.0,
|
|
|
|
| 25 |
"num_hidden_layers": 26,
|
| 26 |
"num_key_value_heads": 4,
|
| 27 |
"pad_token_id": 0,
|
| 28 |
+
"quantization_config": {
|
| 29 |
+
"_load_in_4bit": true,
|
| 30 |
+
"_load_in_8bit": false,
|
| 31 |
+
"bnb_4bit_compute_dtype": "bfloat16",
|
| 32 |
+
"bnb_4bit_quant_storage": "uint8",
|
| 33 |
+
"bnb_4bit_quant_type": "nf4",
|
| 34 |
+
"bnb_4bit_use_double_quant": true,
|
| 35 |
+
"llm_int8_enable_fp32_cpu_offload": false,
|
| 36 |
+
"llm_int8_has_fp16_weight": false,
|
| 37 |
+
"llm_int8_skip_modules": null,
|
| 38 |
+
"llm_int8_threshold": 6.0,
|
| 39 |
+
"load_in_4bit": true,
|
| 40 |
+
"load_in_8bit": false,
|
| 41 |
+
"quant_method": "bitsandbytes"
|
| 42 |
+
},
|
| 43 |
"query_pre_attn_scalar": 256,
|
| 44 |
"rms_norm_eps": 1e-06,
|
| 45 |
"rope_theta": 10000.0,
|