Spaces:
Sleeping
Sleeping
Update app.py
Browse files
app.py
CHANGED
|
@@ -1,97 +1,75 @@
|
|
| 1 |
import os
|
| 2 |
import threading
|
|
|
|
|
|
|
| 3 |
import requests
|
| 4 |
import subprocess
|
| 5 |
import json
|
| 6 |
-
from flask import Flask, jsonify, request
|
| 7 |
-
from llama_cpp import Llama
|
| 8 |
|
| 9 |
app = Flask(__name__)
|
| 10 |
|
| 11 |
# Configuration
|
| 12 |
-
|
| 13 |
-
MODEL_NAME = "calme-3.3-llamaloi-3b.Q4_K_M.gguf"
|
| 14 |
-
MODEL_PATH = os.path.join(MODEL_DIR, MODEL_NAME)
|
| 15 |
-
MODEL_URL = "https://huggingface.co/MaziyarPanahi/calme-3.3-llamaloi-3b-GGUF/resolve/main/calme-3.3-llamaloi-3b.Q4_K_M.gguf"
|
| 16 |
GH_PAT = os.getenv("GH_PAT")
|
| 17 |
REPO_URL = "https://github.com/NitinBot001/Audio-url-new-js.git"
|
| 18 |
|
| 19 |
-
# Global state
|
| 20 |
-
initialized = False
|
| 21 |
-
llm = None
|
| 22 |
-
|
| 23 |
def background_init():
|
| 24 |
-
|
| 25 |
try:
|
| 26 |
-
# 1.
|
| 27 |
-
os.makedirs(
|
| 28 |
-
|
| 29 |
-
# 2. Download model if not exists
|
| 30 |
if not os.path.exists(MODEL_PATH):
|
| 31 |
-
print("
|
| 32 |
-
|
|
|
|
| 33 |
r.raise_for_status()
|
| 34 |
with open(MODEL_PATH, "wb") as f:
|
| 35 |
for chunk in r.iter_content(chunk_size=8192):
|
| 36 |
f.write(chunk)
|
| 37 |
-
print("Model download complete")
|
| 38 |
|
| 39 |
-
#
|
|
|
|
| 40 |
llm = Llama(
|
| 41 |
model_path=MODEL_PATH,
|
| 42 |
-
n_ctx=8192,
|
| 43 |
n_threads=2,
|
| 44 |
n_gpu_layers=0,
|
| 45 |
verbose=False
|
| 46 |
)
|
| 47 |
|
| 48 |
-
#
|
| 49 |
tunnel_url = start_tunnel()
|
| 50 |
-
|
| 51 |
|
| 52 |
-
initialized = True
|
| 53 |
-
print("Initialization complete")
|
| 54 |
except Exception as e:
|
| 55 |
-
print(f"
|
| 56 |
|
| 57 |
def start_tunnel():
|
| 58 |
-
"""Start
|
| 59 |
proc = subprocess.Popen(
|
| 60 |
["npx", "nport", "-s", "hf-space", "-p", "7860"],
|
| 61 |
stdout=subprocess.PIPE,
|
| 62 |
-
stderr=subprocess.PIPE
|
| 63 |
-
text=True
|
| 64 |
)
|
| 65 |
-
|
| 66 |
# Wait for tunnel URL
|
| 67 |
-
|
| 68 |
-
|
| 69 |
-
|
| 70 |
-
|
| 71 |
-
if proc.poll() is not None:
|
| 72 |
-
break
|
| 73 |
-
raise RuntimeError("Failed to establish tunnel")
|
| 74 |
|
| 75 |
-
def
|
| 76 |
-
"""Update
|
| 77 |
-
repo_dir = "/
|
| 78 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 79 |
|
| 80 |
-
|
| 81 |
-
if os.path.exists(repo_dir):
|
| 82 |
-
subprocess.run(["git", "-C", repo_dir, "pull"], check=True)
|
| 83 |
-
else:
|
| 84 |
-
subprocess.run([
|
| 85 |
-
"git", "clone",
|
| 86 |
-
f"https://x-access-token:{GH_PAT}@github.com/NitinBot001/Audio-url-new-js.git",
|
| 87 |
-
repo_dir
|
| 88 |
-
], check=True)
|
| 89 |
-
|
| 90 |
-
# Update instance.json
|
| 91 |
-
with open(instance_path, "w") as f:
|
| 92 |
json.dump({"tunnel_url": url}, f)
|
| 93 |
-
|
| 94 |
-
# Commit and push changes
|
| 95 |
subprocess.run(["git", "-C", repo_dir, "add", "."], check=True)
|
| 96 |
subprocess.run([
|
| 97 |
"git", "-C", repo_dir,
|
|
@@ -100,9 +78,9 @@ def update_repo_with_tunnel(url):
|
|
| 100 |
subprocess.run(["git", "-C", repo_dir, "push"], check=True)
|
| 101 |
|
| 102 |
@app.route("/chat", methods=["GET"])
|
| 103 |
-
def
|
| 104 |
-
if not
|
| 105 |
-
return jsonify({"error": "
|
| 106 |
message = request.args.get("message", "")
|
| 107 |
prompt = f"<|begin_of_text|><|start_header_id|>user<|end_header_id|>\n{message}<|eot_id|>\n<|start_header_id|>assistant<|end_header_id|>\n"
|
| 108 |
output = llm(prompt, max_tokens=512, stop=["<|eot_id|>"])
|
|
@@ -110,13 +88,10 @@ def chat_endpoint():
|
|
| 110 |
|
| 111 |
@app.route("/health")
|
| 112 |
def health_check():
|
| 113 |
-
return
|
| 114 |
-
"status": "ready" if initialized else "initializing",
|
| 115 |
-
"model_loaded": os.path.exists(MODEL_PATH)
|
| 116 |
-
}), 200 if initialized else 503
|
| 117 |
|
| 118 |
if __name__ == "__main__":
|
| 119 |
-
# Start initialization
|
| 120 |
threading.Thread(target=background_init, daemon=True).start()
|
| 121 |
# Start Flask server
|
| 122 |
app.run(host="0.0.0.0", port=7860)
|
|
|
|
| 1 |
import os
|
| 2 |
import threading
|
| 3 |
+
from flask import Flask, jsonify, request
|
| 4 |
+
from llama_cpp import Llama
|
| 5 |
import requests
|
| 6 |
import subprocess
|
| 7 |
import json
|
|
|
|
|
|
|
| 8 |
|
| 9 |
app = Flask(__name__)
|
| 10 |
|
| 11 |
# Configuration
|
| 12 |
+
MODEL_PATH = "/tmp/model/calme-3.3-llamaloi-3b.Q4_K_M.gguf"
|
|
|
|
|
|
|
|
|
|
| 13 |
GH_PAT = os.getenv("GH_PAT")
|
| 14 |
REPO_URL = "https://github.com/NitinBot001/Audio-url-new-js.git"
|
| 15 |
|
|
|
|
|
|
|
|
|
|
|
|
|
| 16 |
def background_init():
|
| 17 |
+
"""Handle time-consuming operations in background"""
|
| 18 |
try:
|
| 19 |
+
# 1. Download model
|
| 20 |
+
os.makedirs(os.path.dirname(MODEL_PATH), exist_ok=True)
|
|
|
|
|
|
|
| 21 |
if not os.path.exists(MODEL_PATH):
|
| 22 |
+
print("Starting model download...")
|
| 23 |
+
url = "https://huggingface.co/MaziyarPanahi/calme-3.3-llamaloi-3b-GGUF/resolve/main/calme-3.3-llamaloi-3b.Q4_K_M.gguf"
|
| 24 |
+
with requests.get(url, stream=True) as r:
|
| 25 |
r.raise_for_status()
|
| 26 |
with open(MODEL_PATH, "wb") as f:
|
| 27 |
for chunk in r.iter_content(chunk_size=8192):
|
| 28 |
f.write(chunk)
|
|
|
|
| 29 |
|
| 30 |
+
# 2. Initialize LLM
|
| 31 |
+
global llm
|
| 32 |
llm = Llama(
|
| 33 |
model_path=MODEL_PATH,
|
| 34 |
+
n_ctx=8192, # Reduced from 131072 for faster startup
|
| 35 |
n_threads=2,
|
| 36 |
n_gpu_layers=0,
|
| 37 |
verbose=False
|
| 38 |
)
|
| 39 |
|
| 40 |
+
# 3. Tunnel and Git operations
|
| 41 |
tunnel_url = start_tunnel()
|
| 42 |
+
push_tunnel_url_to_repo(tunnel_url)
|
| 43 |
|
|
|
|
|
|
|
| 44 |
except Exception as e:
|
| 45 |
+
print(f"Background init failed: {str(e)}")
|
| 46 |
|
| 47 |
def start_tunnel():
|
| 48 |
+
"""Start tunnel and return URL"""
|
| 49 |
proc = subprocess.Popen(
|
| 50 |
["npx", "nport", "-s", "hf-space", "-p", "7860"],
|
| 51 |
stdout=subprocess.PIPE,
|
| 52 |
+
stderr=subprocess.PIPE
|
|
|
|
| 53 |
)
|
|
|
|
| 54 |
# Wait for tunnel URL
|
| 55 |
+
for line in iter(proc.stdout.readline, b''):
|
| 56 |
+
if b"your domain is:" in line:
|
| 57 |
+
return line.decode().split("your domain is: ")[1].strip()
|
| 58 |
+
raise RuntimeError("Failed to get tunnel URL")
|
|
|
|
|
|
|
|
|
|
| 59 |
|
| 60 |
+
def push_tunnel_url_to_repo(url):
|
| 61 |
+
"""Update repository with tunnel URL"""
|
| 62 |
+
repo_dir = "/tmp/repo"
|
| 63 |
+
subprocess.run(["rm", "-rf", repo_dir], check=True)
|
| 64 |
+
subprocess.run([
|
| 65 |
+
"git", "clone",
|
| 66 |
+
f"https://x-access-token:{GH_PAT}@github.com/NitinBot001/Audio-url-new-js.git",
|
| 67 |
+
repo_dir
|
| 68 |
+
], check=True)
|
| 69 |
|
| 70 |
+
with open(f"{repo_dir}/instance.json", "w") as f:
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 71 |
json.dump({"tunnel_url": url}, f)
|
| 72 |
+
|
|
|
|
| 73 |
subprocess.run(["git", "-C", repo_dir, "add", "."], check=True)
|
| 74 |
subprocess.run([
|
| 75 |
"git", "-C", repo_dir,
|
|
|
|
| 78 |
subprocess.run(["git", "-C", repo_dir, "push"], check=True)
|
| 79 |
|
| 80 |
@app.route("/chat", methods=["GET"])
|
| 81 |
+
def chat():
|
| 82 |
+
if 'llm' not in globals():
|
| 83 |
+
return jsonify({"error": "Initializing, try again later"}), 503
|
| 84 |
message = request.args.get("message", "")
|
| 85 |
prompt = f"<|begin_of_text|><|start_header_id|>user<|end_header_id|>\n{message}<|eot_id|>\n<|start_header_id|>assistant<|end_header_id|>\n"
|
| 86 |
output = llm(prompt, max_tokens=512, stop=["<|eot_id|>"])
|
|
|
|
| 88 |
|
| 89 |
@app.route("/health")
|
| 90 |
def health_check():
|
| 91 |
+
return "OK", 200
|
|
|
|
|
|
|
|
|
|
| 92 |
|
| 93 |
if __name__ == "__main__":
|
| 94 |
+
# Start background initialization
|
| 95 |
threading.Thread(target=background_init, daemon=True).start()
|
| 96 |
# Start Flask server
|
| 97 |
app.run(host="0.0.0.0", port=7860)
|