Update app.py

app.py

@@ -1,71 +1,69 @@
 import gradio as gr
-from transformers import AutoTokenizer, AutoModelForCausalLM
 import torch
-import os
+import re
+from transformers import AutoTokenizer, AutoModelForCausalLM
 
-MODEL_ID = "google/gemma-3-270m-it"
-HF_TOKEN = os.getenv("HF_TOKEN")
+# ======================
+# Load model
+# ======================
+MODEL_ID = "google/gemma-3-270m"
 
-tokenizer = None
-model = None
+tokenizer = AutoTokenizer.from_pretrained(MODEL_ID)
+model = AutoModelForCausalLM.from_pretrained(
+    MODEL_ID,
+    torch_dtype=torch.float32,
+    device_map="cpu"
+)
 
-def load_model():
-    global tokenizer, model
-    if tokenizer is None or model is None:
-        tokenizer = AutoTokenizer.from_pretrained(
-            MODEL_ID,
-            token=HF_TOKEN
-        )
-        model = AutoModelForCausalLM.from_pretrained(
-            MODEL_ID,
-            token=HF_TOKEN,
-            torch_dtype=torch.float32,
-            low_cpu_mem_usage=True
-        )
+# ======================
+# Clean output (FIX ký tự rác)
+# ======================
+def clean_output(text: str) -> str:
+    # Cắt câu khi gặp dấu kết thúc hợp lệ
+    match = re.match(r"^[\s\S]*?[.!?\n]", text)
+    if match:
+        return match.group(0).strip()
+    return text.strip()
 
-def respond(message, history):
-    load_model()
+# ======================
+# Chat function
+# ======================
+def chat(message, history):
+    prompt = "<bos><start_of_turn>user\n" + message + "\n<end_of_turn>\n<start_of_turn>model\n"
 
-    prompt = "<bos>"
-    for user, bot in history:
-        prompt += (
-            "<start_of_turn>user\n"
-            f"{user}\n"
-            "<end_of_turn>\n"
-            "<start_of_turn>model\n"
-            f"{bot}\n"
-            "<end_of_turn>\n"
-        )
+    inputs = tokenizer(prompt, return_tensors="pt")
 
-    prompt += (
-        "<start_of_turn>user\n"
-        f"{message}\n"
-        "<end_of_turn>\n"
-        "<start_of_turn>model\n"
-    )
+    with torch.no_grad():
+        outputs = model.generate(
+            **inputs,
+            max_new_tokens=200,
+            temperature=0.6,
+            top_p=0.9,
+            do_sample=True,
+            eos_token_id=tokenizer.eos_token_id,
+            pad_token_id=tokenizer.eos_token_id
+        )
 
-    inputs = tokenizer(
-        prompt,
-        return_tensors="pt",
-        truncation=True,
-        max_length=2048
-    )
+    decoded = tokenizer.decode(outputs[0], skip_special_tokens=True)
 
-    outputs = model.generate(
-        **inputs,
-        max_new_tokens=200,
-        temperature=0.7,
-        top_p=0.9,
-        do_sample=True,
-        eos_token_id=tokenizer.eos_token_id
-    )
+    reply = decoded.split("model")[-1].strip()
+    reply = clean_output(reply)
 
-    decoded = tokenizer.decode(outputs[0], skip_special_tokens=True)
-    reply = decoded.split("<start_of_turn>model")[-1].strip()
     return reply
 
-gr.ChatInterface(
-    fn=respond,
+# ======================
+# UI
+# ======================
+demo = gr.ChatInterface(
+    fn=chat,
     title="🤖 Gemma3 270M Cloud Chat",
-    description="Gemma3 270M chạy cloud miễn phí trên Hugging Face Spaces"
-).launch(server_name="0.0.0.0")
+    description="Gemma3 270M chạy cloud miễn phí trên Hugging Face Spaces",
+    examples=["hi", "giải thích AI là gì", "hello world trong python"],
+    submit_btn="Send",
+    retry_btn="Retry",
+    undo_btn="Undo",
+    clear_btn="Clear"
+)
+
+if __name__ == "__main__":
+    demo.launch()