CooLLaMA / app.py
CooLLaMACEO's picture
Update app.py
ef91df3 verified
import os
import uuid
import requests
from flask import Flask, request, jsonify, render_template_string
from llama_cpp import Llama
app = Flask(__name__)
# === CONFIGURATION ===
MODEL_URL = "https://huggingface.co/CooLLaMACEO/CooLLaMA-Gemma2/resolve/main/gemma-2-2b-it.q3_k_m.gguf"
MODEL_PATH = "model.gguf"
# === DOWNLOAD MODEL (Run once) ===
if not os.path.exists(MODEL_PATH):
print("Downloading GGUF model... this may take a few minutes.")
with requests.get(MODEL_URL, stream=True) as r:
r.raise_for_status()
with open(MODEL_PATH, "wb") as f:
for chunk in r.iter_content(chunk_size=8192):
f.write(chunk)
print("Download complete!")
# === INITIALIZE LLM ===
# n_ctx is the context window (memory). 2048 is a good balance for free tier RAM.
llm = Llama(
model_path=MODEL_PATH,
n_ctx=2048,
n_threads=4 # Optimized for HF free tier CPU
)
# === HTML TEMPLATE ===
HTML_TEMPLATE = """
<!DOCTYPE html>
<html lang="en">
<head>
<meta charset="UTF-8">
<meta name="viewport" content="width=device-width, initial-scale=1.0">
<title>CooLLaMA AI</title>
<link href="https://fonts.googleapis.com/css2?family=Plus+Jakarta+Sans:wght@400;500;600;800&display=swap" rel="stylesheet">
<script src="https://cdn.jsdelivr.net/npm/marked/marked.min.js"></script>
<link href="https://cdnjs.cloudflare.com/ajax/libs/prism/1.29.0/themes/prism-tomorrow.min.css" rel="stylesheet" />
<script src="https://cdnjs.cloudflare.com/ajax/libs/prism/1.29.0/prism.min.js"></script>
<style>
:root { --bg:#0b0f1a; --card:#161b2c; --user-blue:#2563eb; --ai-bubble:#1e293b; --accent:#8b5cf6; --text-light:#f1f5f9; --text-dim:#64748b; }
body { font-family:'Plus Jakarta Sans',sans-serif; background:var(--bg); color:var(--text-light); display:flex; height:100vh; margin:0; overflow:hidden; }
#sidebar { width:260px; background:#090c14; display:flex; flex-direction:column; border-right:1px solid #2d3748; padding:20px; }
.new-chat-btn { background:linear-gradient(135deg,#2563eb,#7c3aed); border:none; color:white; padding:14px; border-radius:12px; cursor:pointer; font-weight:800; margin-bottom:25px; transition:all 0.2s ease; }
.new-chat-btn:hover { transform:translateY(-2px); box-shadow:0 5px 15px rgba(37,99,235,0.4); }
#chat-list { flex:1; overflow-y:auto; display:flex; flex-direction:column; gap:8px; }
.chat-item { padding:12px; border-radius:10px; cursor:pointer; font-size:13px; color:var(--text-dim); transition:0.3s; border:1px solid transparent; white-space:nowrap; overflow:hidden; text-overflow:ellipsis; }
.chat-item:hover { background:#1e293b; color:white; border-color:#334155; }
.chat-item.active { background:#1e293b; color:white; border-color:#2563eb; }
#main-container { flex:1; display:flex; flex-direction:column; }
header { padding:20px 30px; border-bottom:1px solid #2d3748; display:flex; justify-content:space-between; align-items:center; }
header h1 { font-size:1.5rem; margin:0; font-weight:800; background:linear-gradient(90deg,#60a5fa,#a78bfa); -webkit-background-clip:text; -webkit-text-fill-color:transparent; }
#chat { flex:1; padding:25px; overflow-y:auto; display:flex; flex-direction:column; gap:20px; }
.message-row { display:flex; flex-direction:column; width:100%; animation:slideUp 0.3s ease-out; }
.message { max-width:85%; padding:14px 20px; border-radius:18px; font-size:15px; line-height:1.6; }
.user-row { align-items:flex-end; }
.user { background:var(--user-blue); color:white; border-bottom-right-radius:2px; }
.ai-row { align-items:flex-start; }
.ai { background:var(--ai-bubble); color:var(--text-light); border:1px solid #334155; border-bottom-left-radius:2px; }
.label { font-size:10px; font-weight:700; margin-bottom:6px; color:var(--text-dim); text-transform:uppercase; }
#input-container { padding:25px; background:var(--card); border-top:1px solid #2d3748; }
.input-box { max-width:850px; margin:0 auto; display:flex; background:#0b0f1a; border:1px solid #334155; border-radius:15px; padding:10px; }
textarea { flex:1; border:none; background:transparent; padding:10px; font-size:16px; outline:none; color:white; resize:none; }
button#send-btn { background:var(--user-blue); border:none; color:white; padding:0 25px; border-radius:10px; cursor:pointer; font-weight:800; }
</style>
</head>
<body>
<div id="sidebar">
<button class="new-chat-btn" onclick="startNewChat()">+ New Chat</button>
<div id="chat-list"></div>
</div>
<div id="main-container">
<header><h1>🦙 CooLLaMA</h1><div style="font-size:10px; color:#64748b; font-weight:700;">v3.5 GGUF</div></header>
<div id="chat"></div>
<div id="input-container">
<div class="input-box">
<textarea id="message" placeholder="Send a message..." rows="1"></textarea>
<button id="send-btn" onclick="sendMessage()">Send</button>
</div>
</div>
</div>
<script>
let currentChatId = null;
let localData = JSON.parse(localStorage.getItem('coolllama_chats')) || {};
function loadSidebar(){
const list=document.getElementById("chat-list"); list.innerHTML="";
Object.keys(localData).reverse().forEach(id=>{
const item=document.createElement("div");
item.className="chat-item"+(id===currentChatId?" active":"");
item.innerText=`Session ${id}`;
item.onclick=()=>selectChat(id);
list.appendChild(item);
});
}
async function startNewChat(){
const res=await fetch("/new_chat",{method:"POST"});
const data=await res.json();
currentChatId=data.chat_id;
localData[currentChatId]=[];
saveToCache(); loadSidebar(); selectChat(currentChatId);
}
function selectChat(id){
currentChatId=id;
document.getElementById("chat").innerHTML="";
(localData[id] || []).forEach(msg=>appendMessage(msg.role,msg.content));
loadSidebar();
}
function appendMessage(role,text){
const chatDiv=document.getElementById("chat");
const row=document.createElement("div");
row.className=`message-row ${role==='user'?'user-row':'ai-row'}`;
const content = role==='user' ? text : marked.parse(text);
row.innerHTML=`<span class="label">${role==='user'?'You':'CooLLaMA'}</span><div class="message ${role==='user'?'user':'ai'}">${content}</div>`;
chatDiv.appendChild(row);
chatDiv.scrollTop=chatDiv.scrollHeight;
}
async function sendMessage(){
const input=document.getElementById("message");
const text=input.value.trim();
if(!text || !currentChatId) return;
appendMessage('user', text);
localData[currentChatId].push({role:'user', content:text});
input.value="";
const res=await fetch("/chat",{
method:"POST",
headers:{"Content-Type":"application/json"},
body:JSON.stringify({history:localData[currentChatId]})
});
const data=await res.json();
appendMessage('assistant', data.response);
localData[currentChatId].push({role:'assistant', content:data.response});
saveToCache();
}
function saveToCache(){ localStorage.setItem('coolllama_chats', JSON.stringify(localData)); }
loadSidebar();
if(Object.keys(localData).length > 0) selectChat(Object.keys(localData).sort().reverse()[0]);
else startNewChat();
</script>
</body>
</html>
"""
@app.route("/")
def index():
return render_template_string(HTML_TEMPLATE)
@app.route("/new_chat", methods=["POST"])
def new_chat():
return jsonify({"chat_id": str(uuid.uuid4())[:8].upper()})
@app.route("/chat", methods=["POST"])
def chat():
data = request.json
history = data.get("history", [])
# Simple Prompt Formatting for Gemma-2
user_input = history[-1]["content"] if history else ""
prompt = f"<start_of_turn>user\n{user_input}<end_of_turn>\n<start_of_turn>model\n"
# Generate
output = llm(
prompt,
max_tokens=512,
stop=["<end_of_turn>", "user"],
echo=False
)
response_text = output["choices"][0]["text"].strip()
return jsonify({"response": response_text})
if __name__ == "__main__":
# HF Spaces listen on port 7860 by default
app.run(host="0.0.0.0", port=7860)