Update README.md

README.md

@@ -41,66 +41,66 @@ print(result[0]['generated_text'])
 
 
 # Model Information
-**model_name = "NousResearch/Llama-2-7b-chat-hf"
+- ** model_name = "NousResearch/Llama-2-7b-chat-hf"
 
-**dataset_name = "b-mc2/sql-create-context"
+- ** dataset_name = "b-mc2/sql-create-context"
 
 
 # QLoRA parameters
-**lora_r = 64
+- ** lora_r = 64
 
-**lora_alpha = 16
+- ** lora_alpha = 16
 
-**lora_dropout = 0.1
+- ** lora_dropout = 0.1
 
 
 # bitsandbytes parameters
-**use_4bit = True
+- ** use_4bit = True
 
-**bnb_4bit_compute_dtype = "float16"
+- ** bnb_4bit_compute_dtype = "float16"
 
-**bnb_4bit_quant_type = "nf4"
+- ** bnb_4bit_quant_type = "nf4"
 
-**use_nested_quant = False
+- ** use_nested_quant = False
 
 
 # TrainingArguments parameters
-**num_train_epochs = 1
+- ** num_train_epochs = 1
 
-**fp16 = False
+- ** fp16 = False
 
-**bf16 = False
+- ** bf16 = False
 
-**per_device_train_batch_size = 8
+- ** per_device_train_batch_size = 8
 
-**per_device_eval_batch_size = 4
+- ** per_device_eval_batch_size = 4
 
-**gradient_accumulation_steps = 1
+- ** gradient_accumulation_steps = 1
 
-**gradient_checkpointing = True
+- ** gradient_checkpointing = True
 
-**max_grad_norm = 0.3
+- ** max_grad_norm = 0.3
 
-**learning_rate = 2e-4
+- ** learning_rate = 2e-4
 
-**weight_decay = 0.001
+- ** weight_decay = 0.001
 
-**optim = "paged_adamw_32bit"
+- ** optim = "paged_adamw_32bit"
 
-**lr_scheduler_type = "cosine"
+- ** lr_scheduler_type = "cosine"
 
-**max_steps = -1
+- ** max_steps = -1
 
-**warmup_ratio = 0.03
+- ** warmup_ratio = 0.03
 
-**group_by_length = True
+- ** group_by_length = True
 
-**save_steps = 0
+- ** save_steps = 0
 
-**logging_steps = 25
+- ** logging_steps = 25
 
 
 # SFT parameters
-**max_seq_length = None
+- ** max_seq_length = None
 
-**packing = False
+- ** packing = False