Update README.md

README.md

@@ -128,6 +128,19 @@ _ = model.generate(**inputs, streamer=streamer, max_new_tokens=200, temperature=
 
 ### Inferences in 4bits
 
+```python
+nf4_config = BitsAndBytesConfig(
+ load_in_4bit=True,
+ bnb_4bit_quant_type="nf4",
+ bnb_4bit_use_double_quant=True,
+ bnb_4bit_compute_dtype=torch.bfloat16
+)
+
+model = AutoModelForCausalLM.from_pretrained("rhaymison/cuscuz-7b", quantization_config=nf4_config, device_map= {"": 0})
+tokenizer = AutoTokenizer.from_pretrained("rhaymison/cuscuz-7b")
+
+```
+
 ```python
 runtimeFlag = "cuda:0"
 inputs = tokenizer([f"""<s>Você é um assistente especializado no Nordeste Do Brasil.