Update app.py

app.py

@@ -338,19 +338,6 @@ if quant is not None:
     quant_kwargs["quantization"] = quant
 # --- END robust CUDA detection and quant selection ---
 
-# Only enable FP8 quantization if CUDA is present (FP8 uses Triton/CUDA kernels).
-# If QuantizationPolicy defines a no-op or 'none' option, use it; otherwise omit the arg.
-quant = None
-if use_cuda:
-    quant = QuantizationPolicy.fp8_cast()
-else:
-    # try to use a 'none' policy if available; otherwise we'll omit quantization
-    quant = getattr(QuantizationPolicy, "none", None)
-
-quant_kwargs = {}
-if quant is not None:
-    quant_kwargs["quantization"] = quant
-
 pipeline = LTX23DistilledA2VPipeline(
     distilled_checkpoint_path=checkpoint_path,
     spatial_upsampler_path=spatial_upsampler_path,
@@ -364,7 +351,7 @@ pipeline = LTX23DistilledA2VPipeline(
 print("Preloading models (GPU preloads only if CUDA is available)...")
 ledger = pipeline.model_ledger
 
-if torch.cuda.is_available():
+if use_cuda:
     try:
         # Preload models (this will trigger GPU-side building; only do this when CUDA is present)
         _transformer = ledger.transformer()