ChatBot / app.py
ndahlbom's picture
Update app.py
fd3f736 verified
Raw History Blame
5.04 kB
import gradio as gr
import subprocess
from huggingface_hub import hf_hub_download
# --- 1. Setup & Install ---
subprocess.run("pip install -q 'llama_cpp_python==0.3.15'", shell=True, check=False)
from llama_cpp import Llama
# --- 2. Load Model (GGUF) ---
MODEL_REPO = "Jeppcode/ScalableLab2"
GGUF_FILENAME = "model-q4_k_m.gguf"
print(f"Downloading GGUF model {MODEL_REPO}/{GGUF_FILENAME} ...")
model_path = hf_hub_download(
repo_id=MODEL_REPO,
filename=GGUF_FILENAME,
)
print("Initializing llama.cpp LLM ...")
llm = Llama(
model_path=model_path,
n_ctx=2048,
n_threads=2,
n_batch=64,
use_mmap=True,
use_mlock=False,
)
# --- 3. Prompts & Logic ---
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.",
"Step-by-step reasoning": "You are a careful problem solver. Think step by step and explain your reasoning clearly.",
}
def _extract_text(content):
if isinstance(content, list):
return "\n".join(
block.get("text", "") for block in content
if isinstance(block, dict) and block.get("type") == "text"
)
return str(content)
def build_prompt(message, history, style):
system_prompt = STYLE_SYSTEM_PROMPTS.get(style, STYLE_SYSTEM_PROMPTS["Default"])
prompt_parts = [f"System: {system_prompt}\n", "Conversation:\n"]
for msg in history or []:
role = msg.get("role")
content = _extract_text(msg.get("content", ""))
if content:
if role == "user": prompt_parts.append(f"User: {content}\n")
elif role == "assistant": prompt_parts.append(f"Assistant: {content}\n")
prompt_parts.append(f"User: {message}\n")
prompt_parts.append("Assistant:")
return "".join(prompt_parts)
def chat_fn(message, history, max_new_tokens, style):
# Hardcoded values
temperature = 0.7
top_p = 0.9
repetition_penalty = 1.1
prompt = build_prompt(message, history, style)
output = llm(
prompt,
max_tokens=int(max_new_tokens),
temperature=temperature,
top_p=top_p,
repeat_penalty=repetition_penalty,
stop=["User:", "Assistant:", "System:", "Conversation:"],
)
return output["choices"][0]["text"].strip()
# --- 4. Winter/Christmas Theme Configuration ---
# We remove the specific 'text_color_subdued' argument to fix the crash
winter_theme = gr.themes.Soft(
primary_hue="blue",
neutral_hue="slate",
).set(
body_background_fill="transparent",
block_background_fill="rgba(10, 20, 40, 0.7)",
border_color_primary="#FFFFFF",
button_primary_background_fill="#FFFFFF",
button_primary_text_color="#000000",
)
# Custom CSS for the "Perfect" Look
custom_css = """
/* 1. Background Image: Snowy Winter Forest */
.gradio-container {
background: url('https://images.unsplash.com/photo-1477601263568-180e2c6d046e?q=80&w=2560&auto=format&fit=crop') no-repeat center center fixed;
background-size: cover;
}
/* 2. Transparency */
.gradio-container > .main {
background: transparent !important;
}
/* 3. The Chat & Settings Blocks - "Frosty Glass" with White Borders */
.group, .form, .bubble-wrap {
background: rgba(15, 23, 42, 0.75) !important; /* Dark Blue/Grey Glass */
border: 2px solid #FFFFFF !important; /* THE WHITE BORDER */
border-radius: 12px !important;
backdrop-filter: blur(4px);
box-shadow: 0 4px 15px rgba(0,0,0,0.5);
}
/* 4. Chat Bubbles */
.user-message {
background-color: rgba(255, 255, 255, 0.2) !important; /* Icy White transparency */
border: 1px solid #FFFFFF !important;
color: #FFFFFF !important;
}
.bot-message {
background-color: rgba(0, 0, 0, 0.6) !important;
border: 1px solid #A0A0A0 !important;
color: #E0E0E0 !important;
}
/* 5. Inputs and Text */
textarea, input {
background-color: rgba(0, 0, 0, 0.5) !important;
border: 1px solid #FFFFFF !important;
color: white !important;
}
/* This CSS handles the text color we removed from the Python code */
label, span, p, .prose {
color: #FFFFFF !important;
text-shadow: 1px 1px 2px black;
}
/* Hide Footer */
footer {visibility: hidden}
"""
max_new_tokens_slider = gr.Slider(
minimum=16, maximum=256, value=64, step=8, label="Max Response Length",
)
style_radio = gr.Radio(
choices=["Default", "Short answer", "Detailed explanation"],
value="Detailed explanation",
label="Answer style",
)
demo = gr.ChatInterface(
fn=chat_fn,
title="❄️ WinterChat Lab 2 ❄️",
description="Stay frosty. Chat with the fine-tuned model.",
additional_inputs=[max_new_tokens_slider, style_radio],
additional_inputs_accordion="Settings",
)
demo.theme = winter_theme
demo.css = custom_css
if __name__ == "__main__":
demo.launch()