danielhanchen
commited on
Commit
•
e050e38
1
Parent(s):
038572a
Upload config
Browse files- config.json +15 -0
config.json
CHANGED
@@ -16,6 +16,21 @@
|
|
16 |
"num_attention_heads": 96,
|
17 |
"num_hidden_layers": 88,
|
18 |
"num_key_value_heads": 8,
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
19 |
"rms_norm_eps": 1e-05,
|
20 |
"rope_theta": 1000000.0,
|
21 |
"sliding_window": null,
|
|
|
16 |
"num_attention_heads": 96,
|
17 |
"num_hidden_layers": 88,
|
18 |
"num_key_value_heads": 8,
|
19 |
+
"quantization_config": {
|
20 |
+
"_load_in_4bit": true,
|
21 |
+
"_load_in_8bit": false,
|
22 |
+
"bnb_4bit_compute_dtype": "bfloat16",
|
23 |
+
"bnb_4bit_quant_storage": "uint8",
|
24 |
+
"bnb_4bit_quant_type": "nf4",
|
25 |
+
"bnb_4bit_use_double_quant": true,
|
26 |
+
"llm_int8_enable_fp32_cpu_offload": false,
|
27 |
+
"llm_int8_has_fp16_weight": false,
|
28 |
+
"llm_int8_skip_modules": null,
|
29 |
+
"llm_int8_threshold": 6.0,
|
30 |
+
"load_in_4bit": true,
|
31 |
+
"load_in_8bit": false,
|
32 |
+
"quant_method": "bitsandbytes"
|
33 |
+
},
|
34 |
"rms_norm_eps": 1e-05,
|
35 |
"rope_theta": 1000000.0,
|
36 |
"sliding_window": null,
|