Update config.json
Browse files- config.json +32 -0
config.json
CHANGED
|
@@ -16,6 +16,18 @@
|
|
| 16 |
"num_hidden_layers": 28,
|
| 17 |
"num_key_value_heads": 4,
|
| 18 |
"pad_token_id": 151654,
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 19 |
"rms_norm_eps": 1e-06,
|
| 20 |
"rope_scaling": {
|
| 21 |
"mrope_section": [
|
|
@@ -46,6 +58,26 @@
|
|
| 46 |
"num_hidden_layers": 28,
|
| 47 |
"num_key_value_heads": 4,
|
| 48 |
"pad_token_id": 151654,
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 49 |
"rms_norm_eps": 1e-06,
|
| 50 |
"rope_scaling": {
|
| 51 |
"mrope_section": [
|
|
|
|
| 16 |
"num_hidden_layers": 28,
|
| 17 |
"num_key_value_heads": 4,
|
| 18 |
"pad_token_id": 151654,
|
| 19 |
+
"quantization_config": {
|
| 20 |
+
"bnb_4bit_compute_dtype": "bfloat16",
|
| 21 |
+
"bnb_4bit_quant_type": "nf4",
|
| 22 |
+
"bnb_4bit_use_double_quant": true,
|
| 23 |
+
"llm_int8_enable_fp32_cpu_offload": false,
|
| 24 |
+
"llm_int8_has_fp16_weight": false,
|
| 25 |
+
"llm_int8_skip_modules": null,
|
| 26 |
+
"llm_int8_threshold": 6.0,
|
| 27 |
+
"load_in_4bit": false,
|
| 28 |
+
"load_in_8bit": false,
|
| 29 |
+
"quant_method": "bitsandbytes"
|
| 30 |
+
},
|
| 31 |
"rms_norm_eps": 1e-06,
|
| 32 |
"rope_scaling": {
|
| 33 |
"mrope_section": [
|
|
|
|
| 58 |
"num_hidden_layers": 28,
|
| 59 |
"num_key_value_heads": 4,
|
| 60 |
"pad_token_id": 151654,
|
| 61 |
+
"quantization_config": {
|
| 62 |
+
"_load_in_4bit": false,
|
| 63 |
+
"_load_in_8bit": false,
|
| 64 |
+
"bnb_4bit_compute_dtype": "bfloat16",
|
| 65 |
+
"bnb_4bit_quant_storage": "uint8",
|
| 66 |
+
"bnb_4bit_quant_type": "nf4",
|
| 67 |
+
"bnb_4bit_use_double_quant": true,
|
| 68 |
+
"llm_int8_enable_fp32_cpu_offload": false,
|
| 69 |
+
"llm_int8_has_fp16_weight": false,
|
| 70 |
+
"llm_int8_skip_modules": [
|
| 71 |
+
"lm_head",
|
| 72 |
+
"multi_modal_projector",
|
| 73 |
+
"merger",
|
| 74 |
+
"modality_projection"
|
| 75 |
+
],
|
| 76 |
+
"llm_int8_threshold": 6.0,
|
| 77 |
+
"load_in_4bit": false,
|
| 78 |
+
"load_in_8bit": false,
|
| 79 |
+
"quant_method": "bitsandbytes"
|
| 80 |
+
},
|
| 81 |
"rms_norm_eps": 1e-06,
|
| 82 |
"rope_scaling": {
|
| 83 |
"mrope_section": [
|