Update config.json
Browse files- config.json +16 -0
config.json
CHANGED
|
@@ -33,6 +33,22 @@
|
|
| 33 |
},
|
| 34 |
"qk_layer_norms": true,
|
| 35 |
"quantization_config": {
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 36 |
},
|
| 37 |
"rms_norm_eps": 1e-06,
|
| 38 |
"tie_word_embeddings": false,
|
|
|
|
| 33 |
},
|
| 34 |
"qk_layer_norms": true,
|
| 35 |
"quantization_config": {
|
| 36 |
+
"_load_in_4bit": true,
|
| 37 |
+
"_load_in_8bit": false,
|
| 38 |
+
"bnb_4bit_compute_dtype": "float32",
|
| 39 |
+
"bnb_4bit_quant_storage": "uint8",
|
| 40 |
+
"bnb_4bit_quant_type": "nf4",
|
| 41 |
+
"bnb_4bit_use_double_quant": true,
|
| 42 |
+
"llm_int8_enable_fp32_cpu_offload": false,
|
| 43 |
+
"llm_int8_has_fp16_weight": false,
|
| 44 |
+
"llm_int8_skip_modules": [
|
| 45 |
+
"lm_head",
|
| 46 |
+
"embed_tokens"
|
| 47 |
+
],
|
| 48 |
+
"llm_int8_threshold": 6.0,
|
| 49 |
+
"load_in_4bit": true,
|
| 50 |
+
"load_in_8bit": false,
|
| 51 |
+
"quant_method": "bitsandbytes"
|
| 52 |
},
|
| 53 |
"rms_norm_eps": 1e-06,
|
| 54 |
"tie_word_embeddings": false,
|