1128v2

Dockerfile

@@ -31,9 +31,11 @@ RUN go mod init vchat-server && \
 # ==========================================
 FROM pytorch/pytorch:2.2.1-cuda12.1-cudnn8-runtime
 
+# Default to LITE_MODE=true to ensure space starts without massive model errors.
+# Set to false only if GPU is present and model files are included.
 ENV PYTHONUNBUFFERED=1 \
     DEBIAN_FRONTEND=noninteractive \
-    LITE_MODE=false \
+    LITE_MODE=true \
     PATH="/home/user/.local/bin:$PATH" \
     PIP_NO_CACHE_DIR=1
 

frontend/src/App.tsx

@@ -3,7 +3,7 @@ import {
   AlertCircle, Play, Upload, Layers, Terminal, Cpu, Activity, 
   FileText, Zap, MessageSquare, Sliders, LayoutDashboard, FileJson, 
   ChevronDown, ChevronRight, Bot, Database, Trash2, Eye, StopCircle, List,
-  CheckCircle, XCircle, BrainCircuit
+  CheckCircle, XCircle, BrainCircuit, Download
 } from 'lucide-react';
 
 function App() {
@@ -45,15 +45,28 @@ function App() {
     appendLog(`[SYSTEM] Uploading ${file.name} to queue...\n`);
     try {
       const res = await fetch('/queue/upload_csv', { method: 'POST', body: fd });
+      
+      if (!res.ok) {
+        const errText = await res.text();
+        throw new Error(`Server responded with ${res.status}: ${errText.substring(0, 100)}`);
+      }
+
       const data = await res.json();
       if(data.error) throw new Error(data.error);
       appendLog(`[SYSTEM] Upload complete. Added ${data.added} links.\n`);
       setRefreshTrigger(prev => prev + 1);
     } catch (err: any) {
       appendLog(`[ERROR] Upload failed: ${err.message}\n`);
+    } finally {
+        // Reset input so same file can be selected again if needed
+        if (fileInputRef.current) fileInputRef.current.value = '';
     }
   };
 
+  const handleDownloadDataset = () => {
+      window.location.href = "/download-dataset";
+  };
+
   const handleStartQueue = async (e: React.FormEvent) => {
     e.preventDefault();
     setIsProcessing(true);
@@ -64,17 +77,21 @@ function App() {
 
     try {
       const response = await fetch('/queue/run', { method: 'POST', body: formData });
-      if (!response.body) throw new Error("No response");
+      if (!response.body) throw new Error("No response body");
       const reader = response.body.getReader();
       const decoder = new TextDecoder();
       while (true) {
         const { done, value } = await reader.read();
         if (done) break;
         const chunk = decoder.decode(value, { stream: true });
-        chunk.split('\n\n').forEach(line => {
-          if (line.startsWith('data:')) appendLog(line.replace('data:', '').trim() + '\n');
-          if (line.startsWith('event: close')) setIsProcessing(false);
-        });
+        const lines = chunk.split('\n\n');
+        for (const line of lines) {
+            if (line.startsWith('data:')) {
+                const msg = line.replace('data:', '').trim();
+                if (msg) appendLog(msg + '\n');
+            }
+            if (line.includes('event: close')) setIsProcessing(false);
+        }
       }
     } catch (err: any) {
       appendLog(`\n[ERROR]: ${err.message}\n`);
@@ -153,7 +170,7 @@ function App() {
               {model === 'gemini' && (
                 <div className="space-y-2">
                   <input type="password" name="gemini_api_key" placeholder="Gemini API Key" className="w-full bg-black/20 border border-slate-800 rounded px-3 py-2 text-xs" />
-                  <input type="text" name="gemini_model_name" defaultValue="models/gemini-2.5-flash-lite" className="w-full bg-black/20 border border-slate-800 rounded px-3 py-2 text-xs" />
+                  <input type="text" name="gemini_model_name" defaultValue="models/gemini-2.0-flash-exp" className="w-full bg-black/20 border border-slate-800 rounded px-3 py-2 text-xs" />
                 </div>
               )}
               {model === 'vertex' && (
@@ -230,7 +247,12 @@ function App() {
                  )}
               </div>
            ) : activeTab === 'moderation' ? (
-              <button onClick={() => setRefreshTrigger(x=>x+1)} className="w-full py-3 bg-slate-800 hover:bg-slate-700 rounded-lg text-xs font-bold text-white">Refresh List</button>
+              <div className="flex flex-col gap-2">
+                  <button onClick={() => setRefreshTrigger(x=>x+1)} className="w-full py-3 bg-slate-800 hover:bg-slate-700 rounded-lg text-xs font-bold text-white">Refresh List</button>
+                  <button onClick={handleDownloadDataset} className="w-full py-3 bg-indigo-900 hover:bg-indigo-800 rounded-lg text-xs font-bold text-white flex items-center justify-center gap-2">
+                    <Download className="w-4 h-4"/> Download CSV
+                  </button>
+              </div>
            ) : (
              <button type="submit" form="control-form" className="w-full py-3 bg-indigo-600 hover:bg-indigo-500 rounded-lg text-xs font-bold text-white flex justify-center gap-2"><Play className="w-4 h-4"/> Run Labeler</button>
            )}

src/app.py

@@ -76,7 +76,7 @@ async def startup_event():
         except Exception as e:
             logging.fatal(f"Could not load models. Error: {e}", exc_info=True)
     else:
-        logging.info("Running in LITE mode.")
+        logging.info("Running in LITE mode (API Only).")
 
 @app.get("/", response_class=HTMLResponse)
 async def read_root(request: Request):
@@ -370,7 +370,11 @@ async def stop_queue_processing():
 async def upload_csv_to_queue(file: UploadFile = File(...)):
     try:
         content = await file.read()
-        decoded = content.decode('utf-8').splitlines()
+        try:
+            decoded = content.decode('utf-8').splitlines()
+        except UnicodeDecodeError:
+             decoded = content.decode('latin-1').splitlines()
+
         reader = csv.reader(decoded)
         links_to_add = []
         header = next(reader, None)
@@ -378,9 +382,12 @@ async def upload_csv_to_queue(file: UploadFile = File(...)):
 
         link_idx = 0
         header_lower = [h.lower() for h in header]
+        
+        # Determine column index
         if "link" in header_lower: link_idx = header_lower.index("link")
         elif "url" in header_lower: link_idx = header_lower.index("url")
-        elif "http" in header[0]: 
+        elif len(header) > 0 and header[0].strip().startswith("http"):
+            # Headerless, first row is data
             links_to_add.append(header[0])
             link_idx = 0
         
@@ -413,6 +420,7 @@ async def upload_csv_to_queue(file: UploadFile = File(...)):
                 
         return {"status": "success", "added": added_count}
     except Exception as e:
+        logging.error(f"Upload CSV error: {e}")
         return JSONResponse(status_code=400, content={"error": str(e), "status": "failed"})
 
 @app.post("/queue/run")

src/inference_logic.py

@@ -16,6 +16,7 @@ from labeling_logic import (
 )
 from toon_parser import parse_veracity_toon
 
+# Google GenAI Imports
 try:
     import google.generativeai as genai_legacy
     from google.generativeai.types import generation_types
@@ -23,6 +24,7 @@ except ImportError:
     genai_legacy = None
 
 try:
+    # Modern Google GenAI SDK (v1)
     from google import genai
     from google.genai.types import (
         GenerateContentConfig,
@@ -47,16 +49,27 @@ logger = logging.getLogger(__name__)
 
 def load_models():
     if LITE_MODE:
-        logger.info("LITE_MODE is enabled. Skipping local model loading.")
+        logger.info("LITE_MODE is enabled via Env. Skipping local model loading.")
         return
+    
     global processor, base_model, peft_model, active_model
     if base_model is not None: return
+    
+    local_model_path = "/app/local_model"
+    if not os.path.exists(local_model_path) or not os.listdir(local_model_path):
+        logger.warning(f"Local model path {local_model_path} does not exist or is empty.")
+        logger.warning("Forcing LITE_MODE (API only) to prevent crash.")
+        global LITE_MODE
+        LITE_MODE = True
+        return
+
     if not torch.cuda.is_available():
-        raise RuntimeError("CUDA is not available. This application requires a GPU for local models.")
+        logger.warning("CUDA is not available. Forcing LITE_MODE (API only).")
+        LITE_MODE = True
+        return
     
     device = torch.device("cuda")
     logger.info(f"CUDA is available. Initializing models on {device}...")
-    local_model_path = "/app/local_model"
     
     try:
         import flash_attn
@@ -65,11 +78,16 @@ def load_models():
         attn_implementation = "sdpa"
 
     logger.info(f"Loading base model from {local_model_path}...")
-    base_model = Qwen3VLForConditionalGeneration.from_pretrained(
-        local_model_path, dtype=torch.bfloat16, device_map="auto", attn_implementation=attn_implementation
-    ).eval()
-    processor = AutoProcessor.from_pretrained(local_model_path)
-    active_model = base_model
+    try:
+        base_model = Qwen3VLForConditionalGeneration.from_pretrained(
+            local_model_path, dtype=torch.bfloat16, device_map="auto", attn_implementation=attn_implementation
+        ).eval()
+        processor = AutoProcessor.from_pretrained(local_model_path)
+        active_model = base_model
+    except Exception as e:
+        logger.error(f"Failed to load local model: {e}")
+        logger.warning("Falling back to LITE_MODE.")
+        LITE_MODE = True
 
 def switch_active_model(model_name: str):
     global active_model, base_model, peft_model
@@ -79,6 +97,9 @@ def switch_active_model(model_name: str):
         active_model = base_model
 
 def inference_step(video_path, prompt, generation_kwargs, sampling_fps, pred_glue=None):
+    if LITE_MODE:
+        return "Local inference not available in LITE_MODE."
+
     global processor, active_model
     if active_model is None: raise RuntimeError("Models not loaded.")
 
@@ -139,7 +160,7 @@ async def attempt_toon_repair(original_text: str, schema: str, client, model_typ
         repaired_text = ""
         
         if model_type == 'gemini':
-            model = genai_legacy.GenerativeModel("models/gemini-2.5-flash-lite")
+            model = genai_legacy.GenerativeModel("models/gemini-2.0-flash-exp")
             response = await loop.run_in_executor(
                 None, 
                 lambda: model.generate_content(repair_prompt, generation_config={"temperature": 0.0})
@@ -178,7 +199,7 @@ async def run_gemini_labeling_pipeline(video_path: str, caption: str, transcript
         uploaded_file = await loop.run_in_executor(None, lambda: genai_legacy.upload_file(path=video_path))
         while uploaded_file.state.name == "PROCESSING": await asyncio.sleep(2)
         
-        model = genai_legacy.GenerativeModel("models/gemini-2.5-flash-lite")
+        model = genai_legacy.GenerativeModel("models/gemini-2.0-flash-exp")
         toon_schema = SCHEMA_REASONING if include_comments else SCHEMA_SIMPLE
         score_instructions = SCORE_INSTRUCTIONS_REASONING if include_comments else SCORE_INSTRUCTIONS_SIMPLE
         

start.sh

@@ -3,10 +3,20 @@
 # 1. Start Python FastAPI in the background (Internal Port 8001)
 echo "Starting Python Inference Engine..."
 export PYTHONPATH=$PYTHONPATH:/app/src
-python -m uvicorn src.app:app --host 127.0.0.1 --port 8001 &
+# Use --log-level info to see startup issues
+python -m uvicorn src.app:app --host 127.0.0.1 --port 8001 --log-level info &
 
-# Wait for Python to initialize
-sleep 5
+# Wait longer for Python to initialize, or until port is open
+echo "Waiting for Python backend to initialize..."
+timeout=30
+while ! curl -s http://127.0.0.1:8001/ > /dev/null; do
+    sleep 2
+    timeout=$((timeout-2))
+    if [ $timeout -le 0 ]; then
+        echo "Python backend failed to start on time. Logs might show why."
+        break
+    fi
+done
 
 # 2. Start Golang Web Server (Public Port 7860)
 echo "Starting Go Web Server..."