fix

.github/workflows/fastvideo-kernel-publish.yml

@@ -67,9 +67,9 @@ jobs:
             - torch-version: '2.9.1'
               cuda-version: '12.8.0'
               torch-cuda-short: 'cu128'
-            - torch-version: '2.10.0'
-              cuda-version: '12.8.0'
-              torch-cuda-short: 'cu128'
+            # - torch-version: '2.10.0'
+            #   cuda-version: '12.8.0'
+            #   torch-cuda-short: 'cu128'
 
     steps:
       - name: Free up disk space
@@ -223,3 +223,6 @@ jobs:
         uses: pypa/gh-action-pypi-publish@release/v1
         with:
           packages-dir: fastvideo-kernel/dist/
+          # PyPI does not allow replacing an existing file with the same name.
+          # This makes re-runs idempotent by skipping files already uploaded.
+          skip-existing: true

fastvideo/tests/ssim/test_causal_similarity.py

@@ -19,6 +19,10 @@ if "A40" in device_name:
     device_reference_folder = "A40" + device_reference_folder_suffix
 elif "L40S" in device_name:
     device_reference_folder = "L40S" + device_reference_folder_suffix
+elif "H100" in device_name:
+    device_reference_folder = "H100" + device_reference_folder_suffix
+elif "H200" in device_name:
+    device_reference_folder = "H200" + device_reference_folder_suffix
 else:
     # device_reference_folder = "L40S" + device_reference_folder_suffix
     logger.warning(f"Unsupported device for ssim tests: {device_name}")

fastvideo/tests/ssim/test_inference_similarity.py

@@ -41,6 +41,10 @@ if "A40" in device_name:
     device_reference_folder = "A40" + device_reference_folder_suffix
 elif "L40S" in device_name:
     device_reference_folder = "L40S" + device_reference_folder_suffix
+elif "H100" in device_name:
+    device_reference_folder = "H100" + device_reference_folder_suffix
+elif "H200" in device_name:
+    device_reference_folder = "H200" + device_reference_folder_suffix
 else:
     # device_reference_folder = "L40S" + device_reference_folder_suffix
     logger.warning(f"Unsupported device for ssim tests: {device_name}")

fastvideo/tests/ssim/test_lingbot_similarity.py

@@ -47,6 +47,8 @@ elif "L40S" in device_name:
     device_reference_folder = "L40S" + device_reference_folder_suffix
 elif "H100" in device_name:
     device_reference_folder = "H100" + device_reference_folder_suffix
+elif "H200" in device_name:
+    device_reference_folder = "H200" + device_reference_folder_suffix
 else:
     device_reference_folder = None
     logger.warning("Unsupported device for ssim tests: %s", device_name)

fastvideo/tests/ssim/test_longcat_similarity.py

@@ -35,6 +35,8 @@ elif "L40S" in device_name:
     device_reference_folder = "L40S" + device_reference_folder_suffix
 elif "H100" in device_name:
     device_reference_folder = "H100" + device_reference_folder_suffix
+elif "H200" in device_name:
+    device_reference_folder = "H200" + device_reference_folder_suffix
 else:
     logger.warning(f"Unsupported device for ssim tests: {device_name}")
 

fastvideo/tests/ssim/test_matrixgame_similarity.py

@@ -22,6 +22,10 @@ if "A40" in device_name:
     device_reference_folder = "A40" + device_reference_folder_suffix
 elif "L40S" in device_name:
     device_reference_folder = "L40S" + device_reference_folder_suffix
+elif "H100" in device_name:
+    device_reference_folder = "H100" + device_reference_folder_suffix
+elif "H200" in device_name:
+    device_reference_folder = "H200" + device_reference_folder_suffix
 else:
     # device_reference_folder = "L40S" + device_reference_folder_suffix
     logger.warning(f"Unsupported device for ssim tests: {device_name}")

fastvideo/tests/ssim/test_sd35_similarity.py

@@ -22,6 +22,8 @@ def _device_reference_folder() -> str:
         return "L40S" + suffix
     if "H100" in device_name:
         return "H100" + suffix
+    if "H200" in device_name:
+        return "H200" + suffix
     if "RTX 4090" in device_name or "4090" in device_name:
         return "RTX4090" + suffix
 

fastvideo/tests/training/VSA/test_training_loss_VSA.py

@@ -121,9 +121,9 @@ def test_distributed_training():
     wandb_summary = json.load(open(summary_file))
 
     fields_and_thresholds = {
-        'avg_step_time': 5,
+        'avg_step_time': 15,
         'grad_norm': 0.5,
-        'step_time': 5,
+        'step_time': 15,
         'train_loss': 0.04
     }
 

fastvideo/tests/training/Vanilla/test_training_loss.py

@@ -136,9 +136,9 @@ def test_distributed_training():
     wandb_summary = json.load(open(summary_file))
 
     fields_and_thresholds = {
-        'avg_step_time': 6.0,
+        'avg_step_time': 15.0,
         'grad_norm': 0.3,
-        'step_time': 6.0,
+        'step_time': 15.0,
         'train_loss': 0.0025
     }
 

fastvideo/training/training_pipeline.py

@@ -19,6 +19,7 @@ from einops import rearrange
 from torch.utils.data import DataLoader
 from torchdata.stateful_dataloader import StatefulDataLoader
 from tqdm.auto import tqdm
+from diffusers import FlowMatchEulerDiscreteScheduler
 
 import fastvideo.envs as envs
 try:
@@ -605,6 +606,7 @@ class TrainingPipeline(LoRAPipeline, ABC):
             device="cpu").manual_seed(self.seed + self.global_rank)
         logger.info("Initialized random seeds with seed: %s",
                     self.seed + self.global_rank)
+        self.noise_scheduler = FlowMatchEulerDiscreteScheduler()
 
         if self.training_args.resume_from_checkpoint:
             self._resume_from_checkpoint()