Create chat_qwen.py

chat_qwen.py

@@ -0,0 +1,33 @@
+from transformers import AutoModelForCausalLM, AutoTokenizer
+
+def get_response(prompt: str):
+    model = AutoModelForCausalLM.from_pretrained(
+        "Qwen/Qwen2-7B-Instruct-GPTQ-Int4",
+        torch_dtype="auto",
+        device_map="auto",
+    )
+
+    tokenizer = AutoTokenizer.from_pretrained("Qwen/Qwen2-7B-Instruct-GPTQ-Int4")
+
+    prompt = "Give me a short introduction to large language model."
+    messages = [
+        {"role": "system", "content": "You are a helpful assistant."},
+        {"role": "user", "content": prompt},
+    ]
+    text = tokenizer.apply_chat_template(
+        messages,
+        tokenize=False,
+        add_generation_prompt=True,
+    )
+    model_inputs = tokenizer([text], return_tensors="pt").to(model.device)
+
+    generated_ids = model.generate(
+        **model_inputs,
+        max_new_tokens=512,
+    )
+    generated_ids = [
+        output_ids[len(input_ids):] for input_ids, output_ids in zip(model_inputs.input_ids, generated_ids)
+    ]
+
+    response = tokenizer.batch_decode(generated_ids, skip_special_tokens=True)[0]
+    return response