app.py

app.py

@@ -0,0 +1,38 @@
+from transformers import AutoTokenizer, AutoModelForCausalLM
+import torch
+
+model_id = "scb10x/llama-3-typhoon-v1.5-8b-instruct"
+
+tokenizer = AutoTokenizer.from_pretrained(model_id)
+model = AutoModelForCausalLM.from_pretrained(
+    model_id,
+    torch_dtype=torch.bfloat16,
+    device_map="auto",
+)
+
+messages = [
+    {"role": "system", "content": "You are a helpful assistant who're always speak Thai."},
+    {"role": "user", "content": "ขอสูตรไก่ย่าง"},
+]
+
+input_ids = tokenizer.apply_chat_template(
+    messages,
+    add_generation_prompt=True,
+    return_tensors="pt"
+).to(model.device)
+
+terminators = [
+    tokenizer.eos_token_id,
+    tokenizer.convert_tokens_to_ids("<|eot_id|>")
+]
+
+outputs = model.generate(
+    input_ids,
+    max_new_tokens=512,
+    eos_token_id=terminators,
+    do_sample=True,
+    temperature=0.4,
+    top_p=0.9,
+)
+response = outputs[0][input_ids.shape[-1]:]
+print(tokenizer.decode(response, skip_special_tokens=True))