Add 4-bit quantization and automatic device mapping for improved performance.

README.md

@@ -63,3 +63,44 @@ generated_ids = model.generate(model_inputs,
 decoded = tokenizer.batch_decode(generated_ids)
 print(decoded[0])
 
+```
+
+# 4-bit Quantized Inference
+
+```python
+
+# pip install bitsandbytes accelerate
+
+from transformers import AutoModelForCausalLM, AutoTokenizer, BitsAndBytesConfig
+import torch
+
+quantization_config = BitsAndBytesConfig(
+   load_in_4bit=True,
+   bnb_4bit_quant_type="nf4",
+   bnb_4bit_use_double_quant=True,
+   bnb_4bit_compute_dtype=torch.float16 # or torch.bfloat16
+)
+
+model = AutoModelForCausalLM.from_pretrained("TURKCELL/Turkcell-LLM-7b-v1", device_map="auto", quantization_config=quantization_config)
+tokenizer = AutoTokenizer.from_pretrained("TURKCELL/Turkcell-LLM-7b-v1")
+
+messages = [
+    {"role": "user", "content": "Türkiye'nin başkenti neresidir?"},
+]
+
+encodeds = tokenizer.apply_chat_template(messages, return_tensors="pt")
+
+eos_token = tokenizer("<|im_end|>",add_special_tokens=False)["input_ids"][0]
+
+device = "cuda"
+model_inputs = encodeds.to(device)
+
+generated_ids = model.generate(model_inputs, 
+                               max_new_tokens=1024, 
+                               do_sample=True, 
+                               eos_token_id=eos_token)
+                               
+decoded = tokenizer.batch_decode(generated_ids)
+print(decoded[0])
+
+```