Update app.py

app.py

@@ -2,23 +2,25 @@ import os
 import time
 import spaces
 import torch
-from transformers import AutoModelForCausalLM, AutoTokenizer, TextIteratorStreamer
+from transformers import AutoModelForCausalLM, AutoTokenizer, TextIteratorStreamer, BitsAndBytesConfig
 import gradio as gr
 from threading import Thread
 
+# Define constants and configuration
 MODEL_LIST = ["mistralai/mathstral-7B-v0.1"]
 HF_TOKEN = os.environ.get("HF_TOKEN", None)
 MODEL = os.environ.get("MODEL_ID")
 
-TITLE = "<h1><center>MathΣtral</center></h1>"
-
 PLACEHOLDER = """
 <center>
-<p>MathΣtral - I'm MisMath,Your Math advisor</p>
+<p>MathΣtral - Your Math advisor</p>
+<p>Hi! I'm MisMath. A Math advisor. My model is based on mathstral-7B-v0.1. Feel free to ask your questions</p>
+<p>Mathstral 7B is a model specializing in mathematical and scientific tasks, based on Mistral 7B.</p>
+<p>mathstral-7B-v0.1 is the first Mathstral model</p>
+<img src="Mistral.png" alt="MathStral Model" style="width:300px;height:200px;">
 </center>
 """
 
-
 CSS = """
 .duplicate-button {
     margin: auto !important;
@@ -26,69 +28,87 @@ CSS = """
     background: black !important;
     border-radius: 100vh !important;
 }
-h3 {
+h1 {
     text-align: center;
+    font-size: 2em;
+    color: #333;
 }
 """
 
-device = "cuda" # for GPU usage or "cpu" for CPU usage
+TITLE = "<h1><center>MathΣtral - Your Math advisor</center></h1>"
+
+device = "cuda"  # for GPU usage or "cpu" for CPU usage
 
+# Configuration for model quantization
+quantization_config = BitsAndBytesConfig(
+    load_in_4bit=True,
+    bnb_4bit_compute_dtype=torch.bfloat16,
+    bnb_4bit_use_double_quant=True,
+    bnb_4bit_quant_type="nf4"
+)
+
+# Initialize tokenizer and model
 tokenizer = AutoTokenizer.from_pretrained(MODEL)
 model = AutoModelForCausalLM.from_pretrained(
     MODEL,
     torch_dtype=torch.bfloat16,
     device_map="auto",
-    ignore_mismatched_sizes=True)
+    quantization_config=quantization_config
+)
 
+# Define the chat streaming function
 @spaces.GPU()
 def stream_chat(
-    message: str, 
-    history: list, 
-    temperature: float = 0.3, 
-    max_new_tokens: int = 1024, 
-    top_p: float = 1.0, 
-    top_k: int = 20, 
+    message: str,
+    history: list,
+    system_prompt: str,
+    temperature: float = 0.8,
+    max_new_tokens: int = 1024,
+    top_p: float = 1.0,
+    top_k: int = 20,
     penalty: float = 1.2,
 ):
     print(f'message: {message}')
     print(f'history: {history}')
 
-    conversation = []
+    # Prepare the conversation context
+    conversation_text = system_prompt + "\n"
     for prompt, answer in history:
-        conversation.extend([
-            {"role": "user", "content": prompt}, 
-            {"role": "assistant", "content": answer},
-        ])
+        conversation_text += f"User: {prompt}\nAssistant: {answer}\n"
+
+    conversation_text += f"User: {message}\nAssistant:"
 
-    conversation.append({"role": "user", "content": message})
+    # Tokenize the conversation text
+    input_ids = tokenizer(conversation_text, return_tensors="pt").input_ids.to(model.device)
 
-    input_text=tokenizer.apply_chat_template(conversation, tokenize=False)
-    inputs = tokenizer.encode(input_text, return_tensors="pt").to(device)
     streamer = TextIteratorStreamer(tokenizer, timeout=60.0, skip_prompt=True, skip_special_tokens=True)
-    
+
     generate_kwargs = dict(
-        input_ids=inputs, 
-        max_new_tokens = max_new_tokens,
-        do_sample = False if temperature == 0 else True,
-        top_p = top_p,
-        top_k = top_k,
-        temperature = temperature,
+        input_ids=input_ids,
+        max_new_tokens=max_new_tokens,
+        do_sample=False if temperature == 0 else True,
+        top_p=top_p,
+        top_k=top_k,
+        temperature=temperature,
+        eos_token_id=[128001, 128008, 128009],
         streamer=streamer,
-        pad_token_id = 10,
     )
 
     with torch.no_grad():
         thread = Thread(target=model.generate, kwargs=generate_kwargs)
         thread.start()
-        
+
     buffer = ""
     for new_text in streamer:
         buffer += new_text
-        yield buffer
+        # Clean the buffer to remove unwanted prefixes
+        cleaned_text = buffer.split("Assistant:")[-1].strip()
+        yield cleaned_text
 
-            
-chatbot = gr.Chatbot(height=600, placeholder=PLACEHOLDER)
+# Define the Gradio chatbot component
+chatbot = gr.Chatbot(height=500, placeholder=PLACEHOLDER)
 
+# Define the footer with links
 footer = """
 <div style="text-align: center; margin-top: 20px;">
     <a href="https://www.linkedin.com/in/pejman-ebrahimi-4a60151a7/" target="_blank">LinkedIn</a> |
@@ -99,6 +119,7 @@ footer = """
 </div>
 """
 
+# Create and launch the Gradio interface
 with gr.Blocks(css=CSS, theme="Ajaxon6255/Emerald_Isle") as demo:
     gr.HTML(TITLE)
     gr.DuplicateButton(value="Duplicate Space for private use", elem_classes="duplicate-button")
@@ -108,11 +129,16 @@ with gr.Blocks(css=CSS, theme="Ajaxon6255/Emerald_Isle") as demo:
         fill_height=True,
         additional_inputs_accordion=gr.Accordion(label="��️ Parameters", open=False, render=False),
         additional_inputs=[
+            gr.Textbox(
+                value="You are a helpful assistant for Math questions and complex calculations and programming and your name is MisMath",
+                label="System Prompt",
+                render=False,
+            ),
             gr.Slider(
                 minimum=0,
                 maximum=1,
                 step=0.1,
-                value=0.3,
+                value=0.8,
                 label="Temperature",
                 render=False,
             ),
@@ -157,7 +183,8 @@ with gr.Blocks(css=CSS, theme="Ajaxon6255/Emerald_Isle") as demo:
         ],
         cache_examples=False,
     )
+    gr.HTML(footer)
 
-
+# Launch the application
 if __name__ == "__main__":
-    demo.launch()
+    demo.launch()