Add ZeroGPU support with model caching and monospace font

app.py

@@ -7,9 +7,43 @@ import tempfile
 from pathlib import Path
 
 import gradio as gr
+import torch
 
 from cordon import AnalysisConfig, SemanticLogAnalyzer
 
+# ZeroGPU support with fallback for local development
+try:
+    import spaces
+    ZEROGPU_AVAILABLE = True
+except ImportError:
+    ZEROGPU_AVAILABLE = False
+
+# Detect available device
+def get_device() -> str:
+    """Detect best available device."""
+    if torch.cuda.is_available():
+        return "cuda"
+    elif torch.backends.mps.is_available():
+        return "mps"
+    return "cpu"
+
+# Global model cache (keyed by model_name:device)
+_analyzer_cache: dict[str, SemanticLogAnalyzer] = {}
+
+def get_cached_analyzer(config: AnalysisConfig) -> SemanticLogAnalyzer:
+    """Get or create a cached analyzer instance.
+    
+    Caches the analyzer to avoid repeated model loading.
+    The config is updated on the cached instance for each call.
+    """
+    cache_key = f"{config.model_name}:{config.device}"
+    if cache_key not in _analyzer_cache:
+        _analyzer_cache[cache_key] = SemanticLogAnalyzer(config)
+    else:
+        # Update config params (window_size, k_neighbors, etc.)
+        _analyzer_cache[cache_key].config = config
+    return _analyzer_cache[cache_key]
+
 # Load sample log file
 SAMPLE_LOG_PATH = Path(__file__).parent / "sample_apache.log"
 FALLBACK_LOG_PATH = Path(__file__).parent.parent / "examples" / "apache_sample.log"
@@ -29,7 +63,7 @@ def estimate_tokens(text: str) -> int:
     return len(text) // 4 if text else 0
 
 
-def analyze_logs(
+def _analyze_logs_impl(
     log_content: str,
     window_size: int,
     k_neighbors: int,
@@ -54,12 +88,14 @@ def analyze_logs(
     try:
         progress(0.1, desc="Configuring analyzer...")
         
+        device = get_device()
+        
         if mode == "Percentile":
             config = AnalysisConfig(
                 window_size=window_size,
                 k_neighbors=k_neighbors,
                 anomaly_percentile=anomaly_percentile,
-                device="cpu",
+                device=device,
             )
         else:
             config = AnalysisConfig(
@@ -67,11 +103,11 @@ def analyze_logs(
                 k_neighbors=k_neighbors,
                 anomaly_range_min=range_min,
                 anomaly_range_max=range_max,
-                device="cpu",
+                device=device,
             )
         
         progress(0.15, desc="Loading embedding model...")
-        analyzer = SemanticLogAnalyzer(config)
+        analyzer = get_cached_analyzer(config)
         
         progress(0.25, desc="Segmenting log into windows...")
         # Small delay to show progress
@@ -105,6 +141,13 @@ def analyze_logs(
         temp_path.unlink(missing_ok=True)
 
 
+# Apply ZeroGPU decorator if available, otherwise use the function directly
+if ZEROGPU_AVAILABLE:
+    analyze_logs = spaces.GPU(duration=120)(_analyze_logs_impl)
+else:
+    analyze_logs = _analyze_logs_impl
+
+
 CUSTOM_CSS = """
 .main, .wrap, .contain, .gradio-container, .app {
     max-width: 100% !important;
@@ -112,6 +155,13 @@ CUSTOM_CSS = """
     padding-left: 2rem !important;
     padding-right: 2rem !important;
 }
+
+/* Monospace font for log input/output */
+textarea {
+    font-family: "JetBrains Mono", "Fira Code", "SF Mono", "Cascadia Code", Menlo, Monaco, Consolas, "Liberation Mono", monospace !important;
+    font-size: 13px !important;
+    line-height: 1.5 !important;
+}
 """
 
 THEME = gr.themes.Base(

requirements.txt

@@ -1,5 +1,8 @@
 # Cordon package from PyPI
-cordon>=0.3.0
+cordon>=0.3.1
 
 # UI dependencies
 gradio>=4.0.0
+
+# HuggingFace Spaces (ZeroGPU support)
+spaces