Upload train.py with huggingface_hub
Browse files
train.py
CHANGED
|
@@ -71,11 +71,11 @@ learning_rate = 2e-4
|
|
| 71 |
"""## Load model and tokenizer"""
|
| 72 |
|
| 73 |
# specify how to quantize the model
|
| 74 |
-
quantization_config = BitsAndBytesConfig(
|
| 75 |
-
|
| 76 |
-
|
| 77 |
-
|
| 78 |
-
)
|
| 79 |
|
| 80 |
# Load model
|
| 81 |
model = AutoModelForCausalLM.from_pretrained(
|
|
@@ -83,7 +83,7 @@ model = AutoModelForCausalLM.from_pretrained(
|
|
| 83 |
torch_dtype=torch.bfloat16,
|
| 84 |
use_cache=False, # Disable KV cache during training
|
| 85 |
device_map="auto",
|
| 86 |
-
quantization_config=quantization_config
|
| 87 |
)
|
| 88 |
|
| 89 |
# Load tokenizer
|
|
|
|
| 71 |
"""## Load model and tokenizer"""
|
| 72 |
|
| 73 |
# specify how to quantize the model
|
| 74 |
+
# quantization_config = BitsAndBytesConfig(
|
| 75 |
+
# load_in_4bit=True,
|
| 76 |
+
# bnb_4bit_quant_type="nf4",
|
| 77 |
+
# bnb_4bit_use_double_quant=True,
|
| 78 |
+
# )
|
| 79 |
|
| 80 |
# Load model
|
| 81 |
model = AutoModelForCausalLM.from_pretrained(
|
|
|
|
| 83 |
torch_dtype=torch.bfloat16,
|
| 84 |
use_cache=False, # Disable KV cache during training
|
| 85 |
device_map="auto",
|
| 86 |
+
# quantization_config=quantization_config
|
| 87 |
)
|
| 88 |
|
| 89 |
# Load tokenizer
|