Update app.py

app.py

@@ -407,7 +407,7 @@ pipeline = LTX23DistilledA2VPipeline(
     spatial_upsampler_path=spatial_upsampler_path,
     gemma_root=gemma_root,
     loras=[],
-    quantization=None,
+    quantization=QuantizationPolicy.fp8_cast(),
 )
 
 def _make_lora_key(singularity_strength, teneros_strength, sulphur_strength, pose_strength, general_strength, motion_strength, dreamlay_strength, mself_strength, dramatic_strength, fluid_strength, liquid_strength, demopose_strength, voice_strength, realism_strength, transition_strength, physics_strength, reasoning_strength, twostep_strength, mcfurry_strength, dm_strength, praxis_strength, threed_strength, concept_strength, bulge_strength) -> tuple[str, str]:
@@ -435,7 +435,7 @@ def prepare_lora_cache(
         return PENDING_LORA_STATUS
     try:
         progress(0.35, desc="Building fused CPU transformer")
-        tmp_ledger = pipeline.model_ledger.__class__(dtype=ledger.dtype, device=torch.device("cpu"), checkpoint_path=str(checkpoint_path), spatial_upsampler_path=str(spatial_upsampler_path), gemma_root_path=str(gemma_root), loras=tuple(loras_for_builder), quantization=None)
+        tmp_ledger = pipeline.model_ledger.__class__(dtype=ledger.dtype, device=torch.device("cpu"), checkpoint_path=str(checkpoint_path), spatial_upsampler_path=str(spatial_upsampler_path), gemma_root_path=str(gemma_root), loras=tuple(loras_for_builder), quantization=QuantizationPolicy.fp8_cast())
         new_transformer_cpu = tmp_ledger.transformer()
         progress(0.70, desc="Extracting fused state_dict")
         state = {k: v.detach().cpu().contiguous() for k, v in new_transformer_cpu.state_dict().items()}