Update app.py

app.py

@@ -5,7 +5,6 @@ import requests
 SYSTEM_PROMPT = "As an LLM, your job is to generate detailed prompts that start with generate the image, for image generation models based on user input. Be descriptive and specific, but also make sure your prompts are clear and concise."
 TITLE = "Image Prompter"
 EXAMPLE_INPUT = "A Man Riding A Horse in Space"
-zephyr_3b = "https://huggingface.co/chat/models/llama/llama-3b"
 
 HF_TOKEN = os.getenv("HF_TOKEN")
 HEADERS = {"Authorization": f"Bearer {HF_TOKEN}"}
@@ -14,41 +13,32 @@ def build_input_prompt(message, chatbot, system_prompt):
     """
     Constructs the input prompt string from the chatbot interactions and the current message.
     """
-    input_prompt = "<|system|>\n" + system_prompt + "</s>\n<|user|>\n"
-    for interaction in chatbot:
-        input_prompt = input_prompt + str(interaction[0]) + "</s>\n<|assistant|>\n" + str(interaction[1]) + "\n</s>\n<|user|>\n"
-
-    input_prompt = input_prompt + str(message) + "</s>\n<|assistant|>"
+    input_prompt = system_prompt + "\n\n" + message
     return input_prompt
 
 
-def post_request_beta(payload):
+def post_request(model_url, payload):
     """
-    Sends a POST request to the predefined Zephyr-7b-Beta URL and returns the JSON response.
+    Sends a POST request to the specified model URL and returns the JSON response.
     """
-    response = requests.post(zephyr_7b_beta, headers=HEADERS, json=payload)
+    response = requests.post(model_url, headers=HEADERS, json=payload)
     response.raise_for_status()  # Will raise an HTTPError if the HTTP request returned an unsuccessful status code
     return response.json()
 
 
-def predict_beta(message, chatbot=[], system_prompt=""):
+def predict(model_url, message, chatbot=[], system_prompt=""):
     input_prompt = build_input_prompt(message, chatbot, system_prompt)
     data = {
-        "inputs": input_prompt
+        "prompt": input_prompt,
+        "max_new_tokens": 256,
+        "temperature": 0.7,
+        "top_p": 0.95
     }
 
     try:
-        response_data = post_request_beta(data)
-        json_obj = response_data[0]
-        
-        if 'generated_text' in json_obj and len(json_obj['generated_text']) > 0:
-            bot_message = json_obj['generated_text']
-            return bot_message
-        elif 'error' in json_obj:
-            raise gr.Error(json_obj['error'] + ' Please refresh and try again with smaller input prompt')
-        else:
-            warning_msg = f"Unexpected response: {json_obj}"
-            raise gr.Error(warning_msg)
+        response_data = post_request(model_url, data)
+        bot_message = response_data["generated_text"]
+        return bot_message
     except requests.HTTPError as e:
         error_msg = f"Request failed with status code {e.response.status_code}"
         raise gr.Error(error_msg)
@@ -56,10 +46,10 @@ def predict_beta(message, chatbot=[], system_prompt=""):
         error_msg = f"Failed to decode response as JSON: {str(e)}"
         raise gr.Error(error_msg)
 
+
 def test_preview_chatbot(message, history):
-    response = predict_beta(message, history, SYSTEM_PROMPT)
-    text_start = response.rfind("<|assistant|>", ) + len("<|assistant|>")
-    response = response[text_start:]
+    model_url = "https://huggingface.co/chat/models/llama/llama-3b"
+    response = predict(model_url, message, history, SYSTEM_PROMPT)
     return response
 
 
@@ -71,4 +61,12 @@ Expand your imagination and broaden your horizons with LLM. Welcome to **{TITLE}
 chatbot_preview = gr.Chatbot(layout="panel", value=[(None, welcome_preview_message)])
 textbox_preview = gr.Textbox(scale=7, container=False, value=EXAMPLE_INPUT)
 
-demo = gr.ChatInterface(test_preview_chatbot, chatbot=chatbot_preview, textbox=textbox_preview)
+demo = gr.Interface(
+    fn=test_preview_chatbot, 
+    inputs=["text", "state"], 
+    outputs="text", 
+    title=TITLE, 
+    description="Image Prompter"
+)
+
+demo.launch()