主干部分fp16,最敏感psnr=25.21,可以考虑对主干部分太敏感的部分回退fp32

scripts/evaluation/world_model_interaction.py

@@ -571,6 +571,11 @@ def run_inference(args: argparse.Namespace, gpu_num: int, gpu_no: int) -> None:
         torch.save(model, prepared_path)
         print(f">>> Prepared model saved ({os.path.getsize(prepared_path) / 1024**3:.1f} GB).")
 
+    # ---- BF16: only convert the diffusion backbone, keep VAE/CLIP/embedder in FP32 ----
+    model.model.to(torch.bfloat16)
+    model.model.diffusion_model.dtype = torch.bfloat16
+    print(">>> Diffusion backbone (model.model) converted to BF16.")
+
     # Build normalizer (always needed, independent of model loading path)
     logging.info("***** Configing Data *****")
     data = instantiate_from_config(config.data)