sigma_t = t fix, spatiotemporal guidance block fix

Author: prishajain1

src/maxdiffusion/pipelines/ltx2/ltx2_pipeline.py

@@ -139,6 +139,7 @@ def create_model(rngs: nnx.Rngs, ltx2_config: dict):
   ltx2_config["remat_policy"] = config.remat_policy
   ltx2_config["names_which_can_be_saved"] = config.names_which_can_be_saved
   ltx2_config["names_which_can_be_offloaded"] = config.names_which_can_be_offloaded
+  ltx2_config["spatio_temporal_guidance_blocks"] = tuple(getattr(config, "spatio_temporal_guidance_blocks", ()))
 
   # 2. eval_shape
   p_model_factory = partial(create_model, ltx2_config=ltx2_config)
@@ -1526,8 +1527,8 @@ def __call__(
               audio_num_frames,
               frame_rate,
               global_batch_size=batch_size,
-              sigma=sigma_t,
-              audio_sigma=sigma_t,
+              sigma=t,
+              audio_sigma=t,
               use_cross_timestep=use_cross_timestep,
               is_cfg_stg_mode=do_cfg and do_stg,
           )
@@ -1984,8 +1985,8 @@ def scan_body(carry, inputs):
           audio_num_frames=audio_num_frames,
           fps=fps,
           global_batch_size=batch_size,
-          sigma=sigma_t,
-          audio_sigma=sigma_t,
+          sigma=t,
+          audio_sigma=t,
           use_cross_timestep=use_cross_timestep,
           is_cfg_stg_mode=do_cfg and do_stg,
       )