Update app.py

app.py

@@ -2,16 +2,14 @@ import gradio as gr
 import subprocess
 from huggingface_hub import hf_hub_download
 
-# 1. Install llama-cpp-python in runtime (not via requirements.txt)
-#    Important: remove `llama-cpp-python` from requirements.txt,
-#    otherwise Spaces may try to build from source and get stuck.
+# 1. Install llama-cpp-python in runtime
 subprocess.run("pip install -q 'llama_cpp_python==0.3.15'", shell=True, check=False)
-from llama_cpp import Llama
 
+from llama_cpp import Llama
 
-# 2. Load your GGUF model from Hugging Face
+# 2. Load GGUF model from Hugging Face
 MODEL_REPO = "Jeppcode/ScalableLab2"
-GGUF_FILENAME = "model-q4_k_m.gguf"  # or "model-f16.gguf" if you prefer fp16
+GGUF_FILENAME = "model-q4_k_m.gguf"
 
 print(f"Downloading GGUF model {MODEL_REPO}/{GGUF_FILENAME} ...")
 model_path = hf_hub_download(
@@ -22,38 +20,19 @@ model_path = hf_hub_download(
 print("Initializing llama.cpp LLM ...")
 llm = Llama(
     model_path=model_path,
-    n_ctx=2048,      # context length
-    n_threads=2,     # threads (Spaces CPU is limited)
-    n_batch=64,      # batch size for generation
+    n_ctx=2048,
+    n_threads=2,
+    n_batch=64,
     use_mmap=True,
     use_mlock=False,
 )
 
-
-# 3. Style presets (system prompts)
-STYLE_SYSTEM_PROMPTS = {
-    "Default": "You are a helpful, polite assistant.",
-    "Short answer": (
-        "You are a helpful assistant. Answer as concisely as possible, usually in 1-3 sentences."
-    ),
-    "Detailed explanation": (
-        "You are a helpful teaching assistant. Give clear, structured and detailed explanations, "
-        "often with bullet points or numbered steps when useful."
-    ),
-    "Step-by-step reasoning": (
-        "You are a careful problem solver. Think step by step and explain your reasoning clearly "
-        "before giving the final answer."
-    ),
-}
-
+# Hardcoded system prompt (since style selector is removed)
+SYSTEM_PROMPT = "You are a helpful teaching assistant. Give clear, structured and detailed explanations."
 
 def _extract_text_from_content(content):
     """
-    In Gradio 6 ChatInterface, history uses the messages format.
-    content can be:
-      - a string
-      - a list of blocks: [{'type': 'text', 'text': '...'} , ...]
-    We convert it into a simple string.
+    Extracts text from Gradio 6 message format.
     """
     if isinstance(content, list):
         texts = []
@@ -66,84 +45,56 @@ def _extract_text_from_content(content):
     else:
         return str(content)
 
-
-def build_prompt(message, history, style):
+def build_prompt(message, history):
     """
-    Build a simple text prompt for llama.cpp based on:
-      - chosen style (system prompt)
-      - conversation history
-      - latest user message
-
-    Format:
-        System: ...
-        Conversation:
-        User: ...
-        Assistant: ...
-        ...
-        User: <current message>
-        Assistant:
+    Builds the prompt using the hardcoded system prompt.
     """
-    system_prompt = STYLE_SYSTEM_PROMPTS.get(style, STYLE_SYSTEM_PROMPTS["Default"])
-
     prompt_parts = []
-    prompt_parts.append(f"System: {system_prompt}\n")
+    prompt_parts.append(f"System: {SYSTEM_PROMPT}\n")
     prompt_parts.append("Conversation:\n")
-
-    # history is a list of dicts: {'role': 'user'/'assistant'/'system', 'content': ...}
+    
     for msg in history or []:
         role = msg.get("role")
         content = _extract_text_from_content(msg.get("content", ""))
-
         if not content:
             continue
-
         if role == "user":
             prompt_parts.append(f"User: {content}\n")
         elif role == "assistant":
             prompt_parts.append(f"Assistant: {content}\n")
         elif role == "system":
             prompt_parts.append(f"System (previous): {content}\n")
-
+            
     # Current user message
     prompt_parts.append(f"User: {message}\n")
     prompt_parts.append("Assistant:")
-
     full_prompt = "".join(prompt_parts)
     return full_prompt
 
-
-def chat_fn(message, history, max_new_tokens, temperature, top_p, repetition_penalty, style):
+def chat_fn(message, history, max_new_tokens):
     """
-    Main function called by Gradio ChatInterface.
-    - message: latest user input
-    - history: previous messages (messages format)
-    - other params: sliders / radio buttons
+    Main chat function. 
+    Only accepts max_new_tokens as an additional input now.
     """
-    prompt = build_prompt(message, history, style)
-
-    # Handle deterministic mode when temperature == 0
-    temp = float(temperature)
-    top_p_val = float(top_p)
-    repeat_pen = float(repetition_penalty)
-
-    if temp <= 0.0:
-        temp = 0.0
-        top_p_val = 1.0  # less important when temp=0
+    prompt = build_prompt(message, history)
+    
+    # Internal defaults for the removed sliders
+    temperature = 0.7
+    top_p = 0.9
+    repetition_penalty = 1.0
 
     output = llm(
         prompt,
         max_tokens=int(max_new_tokens),
-        temperature=temp,
-        top_p=top_p_val,
-        repeat_penalty=repeat_pen,
+        temperature=temperature,
+        top_p=top_p,
+        repeat_penalty=repetition_penalty,
         stop=["User:", "Assistant:", "System:", "Conversation:"],
     )
-
     reply = output["choices"][0]["text"].strip()
     return reply
 
-
-# 4. Sliders and controls (extra inputs to ChatInterface)
+# 4. Only keep the Max Token slider
 max_new_tokens_slider = gr.Slider(
     minimum=16,
     maximum=256,
@@ -152,258 +103,18 @@ max_new_tokens_slider = gr.Slider(
     label="Max new tokens (response length)",
 )
 
-temperature_slider = gr.Slider(
-    minimum=0.0,
-    maximum=1.5,
-    value=0.0,
-    step=0.1,
-    label="Temperature (0 = deterministic, higher = more random)",
-)
-
-top_p_slider = gr.Slider(
-    minimum=0.1,
-    maximum=1.0,
-    value=0.9,
-    step=0.05,
-    label="Top-p (nucleus sampling)",
-)
-
-repetition_penalty_slider = gr.Slider(
-    minimum=0.8,
-    maximum=1.3,
-    value=1.0,
-    step=0.05,
-    label="Repetition penalty",
-)
-
-style_radio = gr.Radio(
-    choices=[
-        "Default",
-        "Short answer",
-        "Detailed explanation",
-        "Step-by-step reasoning",
-    ],
-    value="Detailed explanation",
-    label="Answer style",
-)
-
-# 5. Christmas theme: inject CSS + hero directly into description
-christmas_style_and_hero = """
-<style>
-body {
-    background: radial-gradient(circle at top, #1b1c2b 0, #050611 55%, #000000 100%);
-    color: #fdf6e3;
-    font-family: "Georgia", "Times New Roman", serif;
-}
-
-/* Use the repo image as a soft background */
-.gradio-container {
-    background:
-        linear-gradient(rgba(0,0,0,0.55), rgba(0,0,0,0.85)),
-        url("file=Cute-Christmas-Background-edit-online-1.jpg");
-    background-size: cover;
-    background-position: center;
-}
-
-/* Christmas hero card */
-.hero {
-    position: relative;
-    margin: 0 auto 1.5rem auto;
-    max-width: 900px;
-    padding: 1.6rem 1.6rem 1.4rem 1.6rem;
-    border-radius: 20px;
-    border: 1px solid rgba(255, 255, 255, 0.14);
-    background:
-        radial-gradient(circle at top,
-            rgba(255, 255, 255, 0.15),
-            rgba(5, 5, 15, 0.98)
-        );
-    box-shadow:
-        0 0 28px rgba(0, 0, 0, 0.9),
-        0 0 70px rgba(180, 0, 40, 0.55);
-    overflow: hidden;
-}
-
-.hero-inner {
-    position: relative;
-    z-index: 1;
-    display: flex;
-    gap: 1.2rem;
-    align-items: center;
-}
-
-.hero-badge {
-    flex-shrink: 0;
-    width: 92px;
-    height: 92px;
-    border-radius: 999px;
-    overflow: hidden;
-    border: 2px solid rgba(255, 255, 255, 0.8);
-    box-shadow:
-        0 0 24px rgba(0, 0, 0, 0.9),
-        0 0 30px rgba(0, 160, 90, 0.7);
-    background:
-        radial-gradient(circle at top,
-            rgba(255,255,255,0.3),
-            rgba(5,5,10,1)
-        ),
-        url("file=Cute-Christmas-Background-edit-online-1.jpg");
-    background-size: cover;
-    background-position: center;
-}
-
-.hero-text h1 {
-    margin: 0 0 0.35rem 0;
-    font-size: 1.35rem;
-    letter-spacing: 0.08em;
-    text-transform: uppercase;
-    color: #ffefe0;
-}
-
-.hero-text p {
-    margin: 0;
-    font-size: 0.95rem;
-    color: #f8eadd;
-    line-height: 1.5;
-}
-
-.hero-keyline {
-    margin-top: 1rem;
-    height: 1px;
-    background-image: linear-gradient(
-        90deg,
-        rgba(255, 255, 255, 0),
-        rgba(255, 225, 150, 0.9),
-        rgba(255, 255, 255, 0)
-    );
-    opacity: 0.9;
-}
-
-/* Chat card */
-.gr-chat-interface {
-    position: relative !important;
-    border-radius: 18px !important;
-    border: 1px solid rgba(255, 255, 255, 0.16);
-    background:
-        linear-gradient(
-            135deg,
-            rgba(5, 10, 20, 0.96),
-            rgba(10, 15, 30, 0.96)
-        );
-    box-shadow:
-        0 0 24px rgba(0, 0, 0, 0.9),
-        0 0 50px rgba(0, 180, 120, 0.45);
-    overflow: hidden;
-}
-
-/* Chat messages as gift tags */
-.gr-chat-message {
-    position: relative;
-    border-radius: 14px !important;
-    border: 1px solid rgba(255, 255, 255, 0.06) !important;
-    backdrop-filter: blur(4px);
-}
-
-.gr-chat-message.user {
-    background:
-        radial-gradient(circle at top left,
-            rgba(0, 180, 120, 0.28),
-            rgba(10, 15, 25, 0.98)
-        ) !important;
-    border-left: 4px solid #00c278 !important;
-}
-
-.gr-chat-message.bot {
-    background:
-        radial-gradient(circle at top left,
-            rgba(230, 30, 90, 0.32),
-            rgba(10, 10, 24, 0.98)
-        ) !important;
-    border-left: 4px solid #ff4060 !important;
-}
-
-/* Input area */
-textarea, .gr-text-input, .gr-textbox {
-    background: rgba(5, 10, 20, 0.98) !important;
-    border-radius: 999px !important;
-    border: 1px solid rgba(255, 255, 255, 0.4) !important;
-    color: #fffbf3 !important;
-}
-
-/* Buttons */
-button, .gr-button {
-    background: linear-gradient(135deg, #ff4060, #00c278) !important;
-    border-radius: 999px !important;
-    border: none !important;
-    color: #fffbf3 !important;
-    font-weight: 600 !important;
-    letter-spacing: 0.08em;
-    text-transform: uppercase;
-    box-shadow:
-        0 0 16px rgba(0, 0, 0, 0.9),
-        0 0 26px rgba(255, 204, 140, 0.6);
-}
-
-button:hover, .gr-button:hover {
-    filter: brightness(1.07);
-    box-shadow:
-        0 0 18px rgba(255, 90, 120, 0.8),
-        0 0 32px rgba(0, 200, 150, 0.7);
-}
-
-/* Sliders and controls */
-input[type="range"] {
-    accent-color: #ff4060;
-}
-
-/* Scrollbar */
-::-webkit-scrollbar {
-    width: 8px;
-}
-::-webkit-scrollbar-track {
-    background: transparent;
-}
-::-webkit-scrollbar-thumb {
-    background: rgba(255, 255, 255, 0.5);
-    border-radius: 999px;
-}
-</style>
-
-<div class="hero">
-  <div class="hero-inner">
-    <div class="hero-badge"></div>
-    <div class="hero-text">
-      <h1>Scalable Lab 2 Christmas Chat</h1>
-      <p>
-        Talk to our fine tuned Llama based model, wrapped as a compact GGUF
-        and running on CPU. Use the controls in the accordion below to tune
-        response length, randomness and style like a Christmas DJ for language models.
-      </p>
-    </div>
-  </div>
-  <div class="hero-keyline"></div>
-</div>
-
-<p>
-🎄 <strong>Tip:</strong> Try switching between short answers and step by step reasoning,
-and play with temperature and top p to see how the model behaves.
-</p>
-"""
-
-
 demo = gr.ChatInterface(
     fn=chat_fn,
     title="Lab 2 – Fine-tuned GGUF model",
-    description=christmas_style_and_hero,
+    description=(
+        "Chat with our fine-tuned Llama-based model, converted to GGUF and "
+        "loaded via llama.cpp from Jeppcode/ScalableLab2."
+    ),
     additional_inputs=[
         max_new_tokens_slider,
-        temperature_slider,
-        top_p_slider,
-        repetition_penalty_slider,
-        style_radio,
     ],
     additional_inputs_accordion="Generation controls",
 )
 
 if __name__ == "__main__":
-    demo.launch()
+    demo.launch()