danielhanchen
commited on
Commit
•
1a808ff
1
Parent(s):
6ca1080
Upload config
Browse files- config.json +15 -0
config.json
CHANGED
@@ -22,6 +22,21 @@
|
|
22 |
"num_hidden_layers": 42,
|
23 |
"num_key_value_heads": 8,
|
24 |
"pad_token_id": 0,
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
25 |
"query_pre_attn_scalar": 256,
|
26 |
"rms_norm_eps": 1e-06,
|
27 |
"rope_theta": 10000.0,
|
|
|
22 |
"num_hidden_layers": 42,
|
23 |
"num_key_value_heads": 8,
|
24 |
"pad_token_id": 0,
|
25 |
+
"quantization_config": {
|
26 |
+
"_load_in_4bit": true,
|
27 |
+
"_load_in_8bit": false,
|
28 |
+
"bnb_4bit_compute_dtype": "bfloat16",
|
29 |
+
"bnb_4bit_quant_storage": "uint8",
|
30 |
+
"bnb_4bit_quant_type": "nf4",
|
31 |
+
"bnb_4bit_use_double_quant": true,
|
32 |
+
"llm_int8_enable_fp32_cpu_offload": false,
|
33 |
+
"llm_int8_has_fp16_weight": false,
|
34 |
+
"llm_int8_skip_modules": null,
|
35 |
+
"llm_int8_threshold": 6.0,
|
36 |
+
"load_in_4bit": true,
|
37 |
+
"load_in_8bit": false,
|
38 |
+
"quant_method": "bitsandbytes"
|
39 |
+
},
|
40 |
"query_pre_attn_scalar": 256,
|
41 |
"rms_norm_eps": 1e-06,
|
42 |
"rope_theta": 10000.0,
|