(Trained with Unsloth)
Browse files- config.json +25 -1
config.json
CHANGED
|
@@ -23,7 +23,31 @@
|
|
| 23 |
"bnb_4bit_use_double_quant": true,
|
| 24 |
"llm_int8_enable_fp32_cpu_offload": false,
|
| 25 |
"llm_int8_has_fp16_weight": false,
|
| 26 |
-
"llm_int8_skip_modules":
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 27 |
"llm_int8_threshold": 6.0,
|
| 28 |
"load_in_4bit": true,
|
| 29 |
"load_in_8bit": false,
|
|
|
|
| 23 |
"bnb_4bit_use_double_quant": true,
|
| 24 |
"llm_int8_enable_fp32_cpu_offload": false,
|
| 25 |
"llm_int8_has_fp16_weight": false,
|
| 26 |
+
"llm_int8_skip_modules": [
|
| 27 |
+
"model.layers.0.mlp.gate_proj",
|
| 28 |
+
"model.layers.0.mlp.up_proj",
|
| 29 |
+
"model.layers.0.mlp.down_proj",
|
| 30 |
+
"model.layers.1.mlp.gate_proj",
|
| 31 |
+
"model.layers.1.mlp.up_proj",
|
| 32 |
+
"model.layers.1.mlp.down_proj",
|
| 33 |
+
"model.layers.2.mlp.gate_proj",
|
| 34 |
+
"model.layers.2.mlp.up_proj",
|
| 35 |
+
"model.layers.2.mlp.down_proj",
|
| 36 |
+
"model.layers.4.mlp.gate_proj",
|
| 37 |
+
"model.layers.4.mlp.up_proj",
|
| 38 |
+
"model.layers.4.mlp.down_proj",
|
| 39 |
+
"model.layers.6.mlp.gate_proj",
|
| 40 |
+
"model.layers.6.mlp.up_proj",
|
| 41 |
+
"model.layers.6.mlp.down_proj",
|
| 42 |
+
"model.layers.34.self_attn.q_proj",
|
| 43 |
+
"model.layers.34.self_attn.k_proj",
|
| 44 |
+
"model.layers.34.self_attn.v_proj",
|
| 45 |
+
"model.layers.34.self_attn.o_proj",
|
| 46 |
+
"model.layers.34.mlp.gate_proj",
|
| 47 |
+
"model.layers.34.mlp.up_proj",
|
| 48 |
+
"model.layers.34.mlp.down_proj",
|
| 49 |
+
"lm_head"
|
| 50 |
+
],
|
| 51 |
"llm_int8_threshold": 6.0,
|
| 52 |
"load_in_4bit": true,
|
| 53 |
"load_in_8bit": false,
|