Update config.json
Browse files- config.json +18 -10
config.json
CHANGED
|
@@ -1,16 +1,6 @@
|
|
| 1 |
{
|
| 2 |
"model_type": "transformer",
|
| 3 |
"task_type": "CAUSAL_LM",
|
| 4 |
-
"base_model": "base_model",
|
| 5 |
-
"load_in_4bit": true,
|
| 6 |
-
"quantization_config": {
|
| 7 |
-
"quantization_type": "4bit",
|
| 8 |
-
"load_in_4bit": true
|
| 9 |
-
// Add any other relevant quantization settings here if needed.
|
| 10 |
-
},
|
| 11 |
-
"torch_dtype": "bfloat16",
|
| 12 |
-
"device_map": "auto",
|
| 13 |
-
"trust_remote_code": true,
|
| 14 |
"peft_config": {
|
| 15 |
"lora_alpha": 16,
|
| 16 |
"lora_dropout": 0.1,
|
|
@@ -18,5 +8,23 @@
|
|
| 18 |
"bias": "none",
|
| 19 |
"task_type": "CAUSAL_LM",
|
| 20 |
"target_modules": ["q_proj", "k_proj", "v_proj", "o_proj", "gate_proj"]
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 21 |
}
|
| 22 |
}
|
|
|
|
| 1 |
{
|
| 2 |
"model_type": "transformer",
|
| 3 |
"task_type": "CAUSAL_LM",
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 4 |
"peft_config": {
|
| 5 |
"lora_alpha": 16,
|
| 6 |
"lora_dropout": 0.1,
|
|
|
|
| 8 |
"bias": "none",
|
| 9 |
"task_type": "CAUSAL_LM",
|
| 10 |
"target_modules": ["q_proj", "k_proj", "v_proj", "o_proj", "gate_proj"]
|
| 11 |
+
},
|
| 12 |
+
"training_arguments": {
|
| 13 |
+
"output_dir": "./results",
|
| 14 |
+
"num_train_epochs": 1,
|
| 15 |
+
"per_device_train_batch_size": 4,
|
| 16 |
+
"gradient_accumulation_steps": 1,
|
| 17 |
+
"optim": "paged_adamw_32bit",
|
| 18 |
+
"save_steps": 50,
|
| 19 |
+
"logging_steps": 1,
|
| 20 |
+
"learning_rate": 2e-4,
|
| 21 |
+
"weight_decay": 0.001,
|
| 22 |
+
"fp16": false,
|
| 23 |
+
"bf16": false,
|
| 24 |
+
"max_grad_norm": 0.3,
|
| 25 |
+
"max_steps": -1,
|
| 26 |
+
"warmup_ratio": 0.03,
|
| 27 |
+
"group_by_length": true,
|
| 28 |
+
"lr_scheduler_type": "constant"
|
| 29 |
}
|
| 30 |
}
|