Defer LTX model loading to GPU worker

app.py

@@ -242,26 +242,7 @@ pipeline = DistilledPipeline(
     quantization=QuantizationPolicy.fp8_cast(),
 )
 
-print("[startup] Preloading LTX models")
-ledger = pipeline.model_ledger
-_transformer = ledger.transformer()
-_video_encoder = ledger.video_encoder()
-_video_decoder = ledger.video_decoder()
-_audio_decoder = ledger.audio_decoder()
-_vocoder = ledger.vocoder()
-_spatial_upsampler = ledger.spatial_upsampler()
-_text_encoder = ledger.text_encoder()
-_embeddings_processor = ledger.gemma_embeddings_processor()
-
-ledger.transformer = lambda: _transformer
-ledger.video_encoder = lambda: _video_encoder
-ledger.video_decoder = lambda: _video_decoder
-ledger.audio_decoder = lambda: _audio_decoder
-ledger.vocoder = lambda: _vocoder
-ledger.spatial_upsampler = lambda: _spatial_upsampler
-ledger.text_encoder = lambda: _text_encoder
-ledger.gemma_embeddings_processor = lambda: _embeddings_processor
-print("[startup] LTX pipeline ready")
+print("[startup] LTX assets ready; model loading is deferred to the GPU worker")
 
 
 def log_memory(tag):