Add proxy

README.md

@@ -1,13 +1,13 @@
 ---
 title: Music Flamingo
-emoji: 🌍
+emoji: 🎵
 colorFrom: yellow
 colorTo: purple
 sdk: gradio
 sdk_version: 5.49.1
 python_version: 3.12
 app_file: app.py
-pinned: false
+pinned: true
 license: apache-2.0
 ---
 

app.py

@@ -1,17 +1,123 @@
+import shutil
 import gradio as gr
 import yt_dlp
 import os
 import tempfile
 import re
+import subprocess
+import socket
+import time
+import atexit
 
 from transformers import AutoModel, AutoProcessor
 
-api_key = os.getenv("my_secret")
+PROXY_URL = None
+_tunnel_proc = None
+
+
+def _write_temp_key_and_kh(key_str, kh_line):
+    key_clean = key_str.replace("\r\n", "\n").replace("\r", "\n")
+    if not key_clean.endswith("\n"):
+        key_clean += "\n"
+    keyf = tempfile.NamedTemporaryFile("w", delete=False)
+    keyf.write(key_clean)
+    keyf.flush()
+    os.chmod(keyf.name, 0o600)
+    keyf.close()
+    khf = tempfile.NamedTemporaryFile("w", delete=False)
+    khf.write(kh_line.strip() + "\n")
+    khf.flush()
+    khf.close()
+    return keyf.name, khf.name
+
+
+def _validate_private_key(path):
+    if not shutil.which("ssh-keygen"):
+        return True
+    try:
+        subprocess.check_output(["ssh-keygen", "-y", "-f", path], stderr=subprocess.STDOUT)
+        return True
+    except subprocess.CalledProcessError:
+        return False
+
+
+def _ensure_local_socks_tunnel():
+    global PROXY_URL, _tunnel_proc
+    if PROXY_URL:
+        return
+    srv = os.getenv("SSH_SERVER")
+    port = os.getenv("SSH_PORT", "22")
+    key = os.getenv("SSH_PRIVATE_KEY")
+    hk = os.getenv("SSH_HOSTKEY")
+    if not (srv and key and hk and shutil.which("ssh")):
+        return
+    key_path, kh_path = _write_temp_key_and_kh(key, hk)
+    if not _validate_private_key(key_path):
+        return
+    cmd = [
+        "ssh","-NT","-p", port,"-i", key_path,
+        "-D","127.0.0.1:1080",
+        "-o","IdentitiesOnly=yes",
+        "-o","ExitOnForwardFailure=yes",
+        "-o","BatchMode=yes",
+        "-o","StrictHostKeyChecking=yes",
+        "-o", f"UserKnownHostsFile={kh_path}",
+        "-o","GlobalKnownHostsFile=/dev/null",
+        "-o","ServerAliveInterval=30","-o","ServerAliveCountMax=3",
+        srv,
+    ]
+    with open("/tmp/ssh_tunnel.log", "w") as lf:
+        _tunnel_proc = subprocess.Popen(cmd, stdout=lf, stderr=lf)
+    for _ in range(40):
+        if _tunnel_proc.poll() is not None:
+            return
+        try:
+            socket.create_connection(("127.0.0.1", 1080), 0.5).close()
+            PROXY_URL = "socks5h://127.0.0.1:1080"
+            break
+        except OSError:
+            time.sleep(0.25)
+    atexit.register(lambda: _tunnel_proc and _tunnel_proc.terminate())
+
+
+_ensure_local_socks_tunnel()
+
 
 MODEL_ID = "nvidia/music-flamingo-hf"
 HERO_IMAGE_URL = "https://musicflamingo.github.io/logo-no-bg.png"
 HERO_TITLE = "Music Flamingo: Scaling Music Understanding in Audio Language Models"
-HERO_SUBTITLE = "Upload audio and ask anything - genre, key, chords, timbre, lyrics, structure. Music Flamingo gives detailed, theory-aware answers."
+HERO_SUBTITLE = "Upload a song and ask anything — including captions, lyrics, genre, key, chords, or complex questions. Music Flamingo gives detailed answers."
+HERO_AUTHORS = """
+<div style="margin-top: 8px; margin-bottom: 4px; padding: 8px 20px; text-align: center; max-width: 900px; margin-inline: auto;">
+  <p style="font-size: 0.95rem; line-height: 1.6; margin-bottom: 10px;">
+    <strong>Authors:</strong> Sreyan Ghosh<sup>1,2*</sup>, Arushi Goel<sup>1*</sup>, Lasha Koroshinadze<sup>2**</sup>, Sang-gil Lee<sup>1</sup>, Zhifeng Kong<sup>1</sup>, Joao Felipe Santos<sup>1</sup>,<br>Ramani Duraiswami<sup>2</sup>, Dinesh Manocha<sup>2</sup>, Wei Ping<sup>1</sup>, Mohammad Shoeybi<sup>1</sup>, Bryan Catanzaro<sup>1</sup>
+  </p>
+  <p style="font-size: 0.88rem; opacity: 0.75; margin-bottom: 8px;">
+    <sup>1</sup>NVIDIA, CA, USA | <sup>2</sup>University of Maryland, College Park, USA
+  </p>
+  <p style="font-size: 0.82rem; opacity: 0.65; font-style: italic; margin-bottom: 6px;">
+    *Equally contributed and led the project. Names randomly ordered. **Significant technical contribution.
+  </p>
+  <p style="font-size: 0.85rem; opacity: 0.7; margin-bottom: 0;">
+    <strong>Correspondence:</strong> <a href="mailto:sreyang@umd.edu" style="color: inherit; text-decoration: underline;">sreyang@umd.edu</a>, <a href="mailto:arushig@nvidia.com" style="color: inherit; text-decoration: underline;">arushig@nvidia.com</a>
+  </p>
+</div>
+"""
+HERO_BADGES = """
+<div style="display: flex; justify-content: center; margin-top: 6px; align-items: center;">
+  <div style="display: flex; justify-content: center; flex-wrap: wrap; gap: 8px;">
+    <a href="https://research.nvidia.com/labs/adlr/MF/"><img src="https://img.shields.io/badge/Demo page-228B22" alt="Demo page"></a>
+    <a href="https://github.com/NVIDIA/audio-flamingo"><img src='https://img.shields.io/badge/Github-Audio Flamingo 3-9C276A' alt="Github"></a>
+    <a href="https://github.com/NVIDIA/audio-flamingo/stargazers"><img src="https://img.shields.io/github/stars/NVIDIA/audio-flamingo.svg?style=social" alt="Stars"></a>
+    <a href="https://huggingface.co/nvidia/music-flamingo">
+      <img src="https://img.shields.io/badge/🤗-Checkpoints-ED5A22.svg" alt="Checkpoints">
+    </a>
+    <a href="https://huggingface.co/datasets/nvidia/MF-Skills">
+      <img src="https://img.shields.io/badge/🤗-Dataset: MF--Skills-ED5A22.svg" alt="Dataset">
+    </a>
+  </div>
+</div>
+"""
 APP_CSS = """
 :root {
   --font-sans: ui-sans-serif, system-ui, sans-serif,
@@ -40,7 +146,7 @@ body {
   display: flex;
   flex-direction: column;
   align-items: center;
-  gap: 16px;
+  gap: 12px;
   padding: 24px 24px 32px;
   text-align: center;
 }
@@ -179,8 +285,8 @@ EXAMPLE_AUDIO_PROMPTS = [
         "Which line directly precedes the chorus?",
     ],
 ]
-processor = AutoProcessor.from_pretrained(MODEL_ID, token=api_key)
-model = AutoModel.from_pretrained(MODEL_ID, token=api_key, device_map="auto").eval()
+processor = AutoProcessor.from_pretrained(MODEL_ID)
+model = AutoModel.from_pretrained(MODEL_ID, device_map="auto").eval()
 
 _youtube_cache = {}
 
@@ -283,6 +389,8 @@ def download_youtube_audio(url, force_reload=False):
             ],
             "noplaylist": True,
         }
+        if PROXY_URL:
+            ydl_opts["proxy"] = PROXY_URL
 
         with yt_dlp.YoutubeDL(ydl_opts) as ydl:
             info = ydl.extract_info(url, download=False)
@@ -372,6 +480,8 @@ with gr.Blocks(css=APP_CSS, theme=gr.themes.Soft(primary_hue="purple", secondary
           <img src="{HERO_IMAGE_URL}" alt="Music Flamingo logo" class="hero__logo" />
           <h1 class="hero__title">{HERO_TITLE}</h1>
           <p class="hero__subtitle">{HERO_SUBTITLE}</p>
+          {HERO_AUTHORS}
+          {HERO_BADGES}
         </div>
         """
     )

packages.txt

@@ -1,3 +1,4 @@
 ffmpeg
 libsndfile1
 git
+openssh-client

requirements.txt

@@ -7,3 +7,4 @@ librosa
 soundfile
 yt-dlp
 gradio
+pysocks