Update app.py

app.py

@@ -1,59 +1,82 @@
 import gradio as gr
-from huggingface_hub import InferenceClient
-
-"""
-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("stabilityai/stablelm-2-1_6b")
-
-
-
-def respond(
-    message,
-    history: list[tuple[str, str]],
-    system_message,
-    max_tokens,
-    temperature,
-    top_p,
-):
-    MAX_HISTORY_LENGTH = 5  # 保留最近 5 條歷史
+import torch
+from transformers import (
+    AutoModelForCausalLM,
+    AutoTokenizer,
+    TextIteratorStreamer,
+)
+import os
+from threading import Thread
+import spaces
+import time
+import subprocess
+
+subprocess.run(
+    "pip install flash-attn --no-build-isolation",
+    env={"FLASH_ATTENTION_SKIP_CUDA_BUILD": "TRUE"},
+    shell=True,
+)
 
-    # 限制歷史對話的長度
-    history = history[-MAX_HISTORY_LENGTH:]
+token = os.environ["HF_TOKEN"]
 
-    # 構建消息列表，從系統消息開始
-    messages = [{"role": "system", "content": system_message}]
 
-    # 添加歷史對話
-    for val in history:
-        if val[0]:
-            messages.append({"role": "user", "content": val[0]})
-        if val[1]:
-            messages.append({"role": "assistant", "content": val[1]})
+model = AutoModelForCausalLM.from_pretrained(
+    "microsoft/Phi-3-mini-128k-instruct",
+    token=token,
+    trust_remote_code=True,
+)
+tok = AutoTokenizer.from_pretrained("microsoft/Phi-3-mini-128k-instruct", token=token)
+terminators = [
+    tok.eos_token_id,
+]
+
+if torch.cuda.is_available():
+    device = torch.device("cuda")
+    print(f"Using GPU: {torch.cuda.get_device_name(device)}")
+else:
+    device = torch.device("cpu")
+    print("Using CPU")
+
+model = model.to(device)
+# Dispatch Errors
+
+
+@spaces.GPU(duration=60)
+def chat(message, history, temperature, do_sample, max_tokens):
+    chat = []
+    for item in history:
+        chat.append({"role": "user", "content": item[0]})
+        if item[1] is not None:
+            chat.append({"role": "assistant", "content": item[1]})
+    chat.append({"role": "user", "content": message})
+    messages = tok.apply_chat_template(chat, tokenize=False, add_generation_prompt=True)
+    model_inputs = tok([messages], return_tensors="pt").to(device)
+    streamer = TextIteratorStreamer(
+        tok, timeout=20.0, skip_prompt=True, skip_special_tokens=True
+    )
+    generate_kwargs = dict(
+        model_inputs,
+        streamer=streamer,
+        max_new_tokens=max_tokens,
+        do_sample=True,
+        temperature=temperature,
+        eos_token_id=terminators,
+    )
 
-    # 添加最新的用戶輸入
-    messages.append({"role": "user", "content": message})
+    if temperature == 0:
+        generate_kwargs["do_sample"] = False
 
-    # 初始化空回應
-    response = ""
+    t = Thread(target=model.generate, kwargs=generate_kwargs)
+    t.start()
 
-    # 呼叫模型並生成回應，使用 stream 模式逐步更新
-    for message in client.chat_completion(
-        messages,
-        max_tokens=max_tokens,
-        stream=True,
-        temperature=temperature,
-        top_p=top_p,
-    ):
-        token = message.choices[0].delta.content
-        response += token
-        yield response
+    partial_text = ""
+    for new_text in streamer:
+        partial_text += new_text
+        yield partial_text
 
+    yield partial_text
 
 
-"""
-For information on how to customize the ChatInterface, peruse the gradio docs: https://www.gradio.app/docs/chatinterface
-"""
 demo = gr.ChatInterface(
     respond,
     additional_inputs=[
@@ -128,7 +151,4 @@ demo = gr.ChatInterface(
         ),
     ],
 )
-
-
-if __name__ == "__main__":
-    demo.launch()
+demo.launch()