ok hông

app.py

@@ -1,69 +1,79 @@
+# app.py
+import torch
 import gradio as gr
-# from huggingface_hub import InferenceClient
+from transformers import AutoConfig, AutoModelForCausalLM, AutoTokenizer
 
-# """
-# For more information on `huggingface_hub` Inference API support, please check the docs: https://huggingface.co/docs/huggingface_hub/v0.22.2/en/guides/inference
-# """
-# client = InferenceClient("HuggingFaceH4/zephyr-7b-beta")
+# 1️⃣ Cấu hình và load model + tokenizer
+model_path = "vinai/PhoGPT-4B-Chat"
+ 
+config = AutoConfig.from_pretrained(model_path, trust_remote_code=True)
+config.init_device = "cpu" if torch.cuda.is_available() else "cpu"
+# Nếu có FlashAttention, bật thêm:
+# config.attn_config['attn_impl'] = 'flash'
 
-# Load model directly
-from transformers import AutoTokenizer, AutoModelForCausalLM
-
-tokenizer = AutoTokenizer.from_pretrained("vinai/PhoGPT-4B-Chat", trust_remote_code=True)
-model = AutoModelForCausalLM.from_pretrained("vinai/PhoGPT-4B-Chat", trust_remote_code=True)
+model = AutoModelForCausalLM.from_pretrained(
+    model_path,
+    config=config,
+    torch_dtype=torch.bfloat16 if torch.cuda.is_available() else torch.float32,
+    trust_remote_code=True,
+)
+model.eval()
 
-def respond(
-    message,
-    history: list[tuple[str, str]],
-    system_message,
-    max_tokens,
-    temperature,
-    top_p,
-):
-    messages = [{"role": "system", "content": system_message}]
+tokenizer = AutoTokenizer.from_pretrained(model_path, trust_remote_code=True)
 
-    for val in history:
-        if val[0]:
-            messages.append({"role": "user", "content": val[0]})
-        if val[1]:
-            messages.append({"role": "assistant", "content": val[1]})
+# 2️⃣ Hàm chat theo template “### Câu hỏi / ### Trả lời”
+PROMPT_TEMPLATE = "### Câu hỏi: {instruction}\n### Trả lời:"
 
+def respond(message, history, system_message, max_tokens, temperature, top_p):
+    # 2.1 — Gom system message và history vào messages list
+    messages = [{"role": "system", "content": system_message}]
+    for u, b in history:
+        if u:
+            messages.append({"role": "user", "content": u})
+        if b:
+            messages.append({"role": "assistant", "content": b})
     messages.append({"role": "user", "content": message})
 
-    response = ""
-
-    for message in model.chat_completion(
+    # 2.2 — Tạo prompt chuẩn
+    prompt = tokenizer.apply_chat_template(
         messages,
-        max_tokens=max_tokens,
-        stream=True,
+        tokenize=False,
+        add_generation_prompt=True
+    )
+
+    # 2.3 — Tokenize và đưa lên device
+    inputs = tokenizer(prompt, return_tensors="pt")
+    inputs = {k: v.to(model.device) for k, v in inputs.items()}
+
+    # 2.4 — Sinh text
+    outputs = model.generate(
+        **inputs,
+        max_new_tokens=max_tokens,
         temperature=temperature,
         top_p=top_p,
-    ):
-        token = message.choices[0].delta.content
+        do_sample=True,
+        eos_token_id=tokenizer.eos_token_id,
+        pad_token_id=tokenizer.pad_token_id,
+    )
 
-        response += token
-        yield response
+    # 2.5 — Decode và tách phần assistant trả lời
+    full = tokenizer.decode(outputs[0], skip_special_tokens=True)
+    answer = full.replace(prompt, "").strip()
 
+    # 2.6 — Cập nhật history và trả về
+    history.append((message, answer))
+    return history
 
-"""
-For information on how to customize the ChatInterface, peruse the gradio docs: https://www.gradio.app/docs/chatinterface
-"""
+# 3️⃣ Giao diện Gradio
 demo = gr.ChatInterface(
-    respond,
+    fn=respond,
     additional_inputs=[
-        gr.Textbox(value=" ", label="System message"),
-        gr.Slider(minimum=1, maximum=2048, value=512, step=1, label="Max new tokens"),
-        gr.Slider(minimum=0.1, maximum=4.0, value=0.7, step=0.1, label="Temperature"),
-        gr.Slider(
-            minimum=0.1,
-            maximum=1.0,
-            value=0.95,
-            step=0.05,
-            label="Top-p (nucleus sampling)",
-        ),
+        gr.Textbox("Bạn là một chatbot tiếng Việt thân thiện.", label="System message"),
+        gr.Slider(1, 2048, value=512, step=1, label="Max new tokens"),
+        gr.Slider(0.1, 4.0, value=0.7, step=0.1, label="Temperature"),
+        gr.Slider(0.1, 1.0, value=0.95, step=0.05, label="Top-p (nucleus sampling)"),
     ],
 )
 
-
 if __name__ == "__main__":
-    demo.launch()
+    demo.launch()