update steram

app.py

@@ -1,74 +1,100 @@
-import gradio as gr
-import os
 import spaces
-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
-"""
-token = os.getenv("HUGGINGFACE_TOKEN")
-if token =="":
-    raise ValueError("HUGGINGFACE_TOKEN environment variable is not set")
-print(token)
-
-client = InferenceClient("google/gemma-2-2b-it",token = token)
-
-@spaces.GPU(duration=30)
-def respond(
-    message,
-    history: list[tuple[str, str]],
-    system_message,
-    max_tokens,
-    temperature,
-    top_p,
-):
-    #messages = [{"role": "system", "content": system_message}] #system not supported
-    messages = []
-
-    for val in history:
-        if val[0]:
-            messages.append({"role": "user", "content": val[0]})
-        if val[1]:
-            messages.append({"role": "assistant", "content": val[1]})
-
-    messages.append({"role": "user", "content": message})
-    response = ""
-
-    # Load model directly
-
-    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
-
-"""
-For information on how to customize the ChatInterface, peruse the gradio docs: https://www.gradio.app/docs/chatinterface
-"""
-demo = gr.ChatInterface(
-    respond,
-    additional_inputs=[
-        gr.Textbox(value="You are a friendly Chatbot.", label="System message"),
-        gr.Slider(minimum=1, maximum=512, value=32, 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)",
-        ),
-    ],
-)
+import os
+import torch
+from transformers import AutoTokenizer, AutoModelForCausalLM, pipeline
+from transformers import TextIteratorStreamer
+from threading import Thread
 
+import gradio as gr
+
+text_generator = None
+is_hugging_face = True
+model_id = "google/gemma-2-9b-it"# too big
+model_id = "google/gemma-2-2b-it"
+huggingface_token = os.getenv("HUGGINGFACE_TOKEN")
+device = "auto" # torch.device("cuda" if torch.cuda.is_available() else "cpu")
+device = "cuda"
+dtype = torch.bfloat16
+dtype = torch.float16
+
+if not huggingface_token:
+    pass
+    print("no HUGGINGFACE_TOKEN if you need set secret ")
+    #raise ValueError("HUGGINGFACE_TOKEN environment variable is not set")
+        
+
+    
+    
+    
+
+
+
+tokenizer = AutoTokenizer.from_pretrained(model_id, token=huggingface_token)
+
+print(model_id,device,dtype)
+histories = []
+#model = None
 
-if __name__ == "__main__":
-    demo.launch()
 
-    
+
+if not is_hugging_face:
+    model = AutoModelForCausalLM.from_pretrained(
+        model_id, token=huggingface_token ,torch_dtype=dtype,device_map=device
+    )
+    text_generator = pipeline("text-generation", model=model, tokenizer=tokenizer,torch_dtype=dtype,device_map=device,stream=True ) #pipeline has not to(device)
+    
+    if next(model.parameters()).is_cuda:
+        print("The model is on a GPU")
+    else:
+        print("The model is on a CPU")
+
+    #print(f"text_generator.device='{text_generator.device}")
+    if str(text_generator.device).strip() == 'cuda':
+        print("The pipeline is using a GPU")
+    else:
+        print("The pipeline is using a CPU")
+
+print("initialized")
+
+@spaces.GPU(duration=60)
+def generate_text(messages):
+    if is_hugging_face:#need everytime initialize for ZeroGPU
+        model = AutoModelForCausalLM.from_pretrained(
+                model_id, token=huggingface_token ,torch_dtype=dtype,device_map=device
+            )
+        model.to(device)
+    question = tokenizer.apply_chat_template(messages, tokenize=False, add_generation_prompt=True)
+    question = tokenizer(question, return_tensors="pt").to(device)
+
+    streamer = TextIteratorStreamer(tokenizer, skip_prompt=True)
+    generation_kwargs = dict(question, streamer=streamer, max_new_tokens=200)
+    thread = Thread(target=model.generate, kwargs=generation_kwargs)
+    
+    generated_output = ""
+    thread.start()
+    for new_text in streamer:
+        generated_output += new_text
+        yield generated_output            
+
+    
+       
+
+
+def call_generate_text(message, history):
+   # history.append({"role": "user", "content": message})
+    #print(message)
+    #print(history)
+   
+    messages = history+[{"role":"user","content":message}]
+    try:
+        
+        for text in generate_text(messages):
+            yield text
+    except RuntimeError  as e:
+        print(f"An unexpected error occurred: {e}")
+        yield ""
+
+demo = gr.ChatInterface(call_generate_text,type="messages")
+
+if __name__ == "__main__":
+    demo.launch(share=True)