diff --git a/src/api/health.py b/src/api/health.py
index 3754394..327e1d9 100644
--- a/src/api/health.py
+++ b/src/api/health.py
@@ -10,7 +10,7 @@ router = APIRouter()
@router.get("/health")
async def health_check():
- """Health check endpoint with GPU status"""
+ """Health check endpoint with GPU and model status"""
# Get GPU info if available
gpu_info = {"available": False}
@@ -30,11 +30,21 @@ async def health_check():
# Get system info
memory = psutil.virtual_memory()
+ # Get model status
+ model_status = get_model_info()
+
return {
"status": "healthy",
"version": "1.0.0",
"model": settings.whisper_model,
"gpu": gpu_info,
+ "model_status": {
+ "loaded": model_status.get("loaded", False),
+ "is_downloading": model_status.get("is_downloading", False),
+ "download_percentage": model_status.get("download_percentage", 0),
+ "status_message": model_status.get("status_message", "Unknown"),
+ "device": model_status.get("device", "cpu")
+ },
"system": {
"cpu_percent": psutil.cpu_percent(interval=1),
"memory_percent": memory.percent,
diff --git a/src/services/whisper_service.py b/src/services/whisper_service.py
index 85cd4b6..08ea3bb 100644
--- a/src/services/whisper_service.py
+++ b/src/services/whisper_service.py
@@ -52,7 +52,7 @@ def _download_hook(progress_bytes, total_bytes):
def load_model(model_name: str = None):
- """Load Whisper model"""
+ """Load Whisper model with automatic CPU fallback"""
global _model, _model_status, _current_model_name
if model_name is None:
@@ -70,12 +70,21 @@ def load_model(model_name: str = None):
_model_status["status_message"] = "Starting download..."
_model_status["model_name"] = model_name
- print(f"Loading Whisper model: {model_name}")
+ # Check if GPU is available, fallback to CPU if not
+ requested_device = settings.whisper_device
+ actual_device = requested_device
+
+ if requested_device == "cuda" and not torch.cuda.is_available():
+ actual_device = "cpu"
+ print(f"⚠️ GPU not available, falling back to CPU mode")
+ _model_status["status_message"] = "GPU not available, using CPU..."
+
+ print(f"Loading Whisper model: {model_name} on {actual_device}")
try:
# Whisper doesn't have a direct progress callback, but we can monitor the models directory
_model = whisper.load_model(
model_name,
- device=settings.whisper_device,
+ device=actual_device,
download_root=settings.models_path
)
_current_model_name = model_name
@@ -83,7 +92,14 @@ def load_model(model_name: str = None):
_model_status["is_loaded"] = True
_model_status["download_percentage"] = 100
_model_status["status_message"] = "Model loaded successfully"
- print(f"Model {model_name} loaded on {settings.whisper_device}")
+
+ # Update device in status to reflect actual device used
+ if actual_device != requested_device:
+ settings.whisper_device = actual_device
+ print(f"✅ Model {model_name} loaded on CPU (GPU fallback)")
+ else:
+ print(f"✅ Model {model_name} loaded on {actual_device}")
+
except Exception as e:
_model_status["is_downloading"] = False
_model_status["status_message"] = f"Error: {str(e)}"
diff --git a/src/templates/dashboard.html b/src/templates/dashboard.html
index 876d595..df4fe43 100644
--- a/src/templates/dashboard.html
+++ b/src/templates/dashboard.html
@@ -151,39 +151,55 @@
// Model Status Updates
async function updateModelStatus() {
try {
- // We need to use an API key for the model-status endpoint
- // For admin panel, we'll use a simple endpoint that doesn't require auth
- // or we'll skip this and use the health endpoint instead
const response = await fetch('/health');
const data = await response.json();
const statusCard = document.getElementById('model-status-content');
- const progressContainer = document.getElementById('model-progress-container');
- const progressBar = document.getElementById('model-progress-bar');
- const progressText = document.getElementById('model-progress-text');
- if (data.gpu && data.gpu.available) {
+ // Prioritize download status over GPU status
+ if (data.model_status && data.model_status.is_downloading) {
+ // Show download status
statusCard.innerHTML = `
- Status: ✅ GPU Ready
- GPU: ${data.gpu.name}
- VRAM: ${data.gpu.vram_used_gb} GB / ${data.gpu.vram_total_gb} GB
- Model: ${data.model}
+ Status: ⏳ Downloading...
+ ${data.model_status.status_message}
+
+
+ ${data.model_status.download_percentage}%
+
`;
- progressContainer.style.display = 'none';
+ } else if (data.model_status && data.model_status.loaded) {
+ // Show GPU/Model ready status
+ if (data.gpu && data.gpu.available) {
+ statusCard.innerHTML = `
+ Status: ✅ Loaded & Ready
+ Model: ${data.model}
+ Device: ${data.model_status.device}
+ GPU: ${data.gpu.name}
+ VRAM: ${data.gpu.vram_used_gb} GB / ${data.gpu.vram_total_gb} GB
+ `;
+ } else {
+ statusCard.innerHTML = `
+ Status: ✅ Loaded & Ready (CPU Mode)
+ Model: ${data.model}
+ Device: CPU
+ `;
+ }
} else {
+ // Not started or loading
statusCard.innerHTML = `
- Status: ⏳ Loading...
- Model is being downloaded. Please wait...
+ Status: ⏸️ Not Started
+ The model will be loaded on first transcription request.
`;
- progressContainer.style.display = 'block';
}
} catch (error) {
console.error('Error fetching model status:', error);
}
}
- // Update status every 5 seconds
+ // Update status every 3 seconds during download for smoother progress
updateModelStatus();
- setInterval(updateModelStatus, 5000);
+ setInterval(updateModelStatus, 3000);
{% endblock %}
diff --git a/src/templates/models.html b/src/templates/models.html
index 330755a..cdc1b56 100644
--- a/src/templates/models.html
+++ b/src/templates/models.html
@@ -38,6 +38,39 @@
+
+
+
⚙️ Processing Mode
+
Current mode: {{ current_status.device | upper }}
+
+ {% if gpu_available %}
+ {% if current_status.device == 'cuda' %}
+
+ {% else %}
+
+ {% endif %}
+
+ GPU detected: {{ gpu_name }}
+
+ {% else %}
+
+
+ ⚠️ No compatible GPU detected. Running in CPU mode.
+
+
+ {% endif %}
+
Available Models
diff --git a/src/web/routes.py b/src/web/routes.py
index d82a3fd..80cbae5 100644
--- a/src/web/routes.py
+++ b/src/web/routes.py
@@ -7,6 +7,8 @@ from fastapi.responses import HTMLResponse, RedirectResponse
from fastapi.templating import Jinja2Templates
from sqlalchemy.orm import Session
+import torch
+
from src.config import settings
from src.database.db import get_db
from src.database.models import ApiKey, UsageLog
@@ -201,12 +203,20 @@ async def manage_models(request: Request, message: Optional[str] = None, error:
models = get_available_models()
current_status = get_model_status()
+ # Check GPU availability
+ gpu_available = torch.cuda.is_available()
+ gpu_name = ""
+ if gpu_available:
+ gpu_name = torch.cuda.get_device_name(0)
+
return templates.TemplateResponse("models.html", {
"request": request,
"models": models,
"current_status": current_status,
"message": message,
- "error": error
+ "error": error,
+ "gpu_available": gpu_available,
+ "gpu_name": gpu_name
})
@@ -316,3 +326,57 @@ async def reload_model_admin(request: Request):
url=f"/admin/models?error={str(e)}",
status_code=302
)
+
+
+@router.post("/models/set-device")
+async def set_device_admin(
+ request: Request,
+ device: str = Form(...)
+):
+ """Switch between CPU and GPU mode"""
+ try:
+ check_admin_auth(request)
+ except HTTPException as e:
+ return RedirectResponse(url="/admin/login", status_code=302)
+
+ try:
+ # Validate device
+ if device not in ["cpu", "cuda"]:
+ return RedirectResponse(
+ url="/admin/models?error=Invalid device. Must be 'cpu' or 'cuda'",
+ status_code=302
+ )
+
+ # Check if GPU is available when switching to cuda
+ if device == "cuda" and not torch.cuda.is_available():
+ return RedirectResponse(
+ url="/admin/models?error=No compatible GPU detected. Cannot switch to GPU mode.",
+ status_code=302
+ )
+
+ # Update settings
+ settings.whisper_device = device
+
+ # Unload current model to force reload with new device
+ from src.services.whisper_service import _model, _current_model_name
+ import gc
+
+ if _model is not None:
+ # Clear the model from memory
+ _model = None
+ gc.collect()
+
+ # Clear CUDA cache if switching away from GPU
+ if torch.cuda.is_available():
+ torch.cuda.empty_cache()
+
+ device_name = "GPU" if device == "cuda" else "CPU"
+ return RedirectResponse(
+ url=f"/admin/models?message=Switched to {device_name} mode. Model will be reloaded on next request.",
+ status_code=302
+ )
+ except Exception as e:
+ return RedirectResponse(
+ url=f"/admin/models?error={str(e)}",
+ status_code=302
+ )