Merge branch 'main' of github.com:hao-ai-lab/FastVideo

fastvideo/fastvideo_args.py

@@ -622,7 +622,6 @@ class TrainingArgs(FastVideoArgs):
     max_grad_norm: float = 0.0
     enable_gradient_checkpointing_type: str | None = None
     selective_checkpointing: float = 0.0
-    allow_tf32: bool = False
     mixed_precision: str = ""
     train_sp_batch_size: int = 0
     fsdp_sharding_startegy: str = ""
@@ -883,9 +882,6 @@ class TrainingArgs(FastVideoArgs):
         parser.add_argument("--selective-checkpointing",
                             type=float,
                             help="Selective checkpointing threshold")
-        parser.add_argument("--allow-tf32",
-                            action=StoreBoolean,
-                            help="Whether to allow TF32")
         parser.add_argument("--mixed-precision",
                             type=str,
                             help="Mixed precision training type")

fastvideo/tests/nightly/test_e2e_i2v_overfit_single_sample.py

@@ -116,7 +116,6 @@ def run_training():
         "--validation_sampling_steps", "50",
         "--log_validation",
         "--checkpoints_total_limit", "3",
-        "--allow_tf32",
         "--ema_start_step", "0",
         "--training_cfg_rate", "0.1",
         "--output_dir", LOCAL_OUTPUT_DIR,

fastvideo/tests/nightly/test_e2e_overfit_single_sample.py

@@ -116,7 +116,6 @@ def run_training():
         "--validation_sampling_steps", "50",
         "--log_validation",
         "--checkpoints_total_limit", "3",
-        "--allow_tf32",
         "--ema_start_step", "0",
         "--training_cfg_rate", "0.0",
         "--output_dir", LOCAL_OUTPUT_DIR,

fastvideo/tests/training/VSA/test_training_loss_VSA.py

@@ -51,7 +51,6 @@ def run_worker():
         "--validation_sampling_steps", "50",
         "--log_validation",
         "--checkpoints_total_limit", "3",
-        "--allow_tf32",
         "--ema_start_step", "0",
         "--training_cfg_rate", "0.0",
         "--output_dir", "data/wan_finetune_test_VSA",

fastvideo/tests/training/Vanilla/test_training_loss.py

@@ -55,7 +55,6 @@ def run_worker():
         "--validation_sampling_steps", "8",
         "--log_validation",
         "--checkpoints_total_limit", "3",
-        "--allow_tf32",
         "--ema_start_step", "0",
         "--training_cfg_rate", "0.0",
         "--output_dir", "data/wan_finetune_test",

fastvideo/tests/training/distill/test_distill_dmd.py

@@ -54,7 +54,6 @@ def run_worker():
         "--validation_sampling_steps", "3",
         "--log_validation",
         "--checkpoints_total_limit", "3",
-        "--allow_tf32",
         "--ema_start_step", "0",
         "--training_cfg_rate", "0.0",
         "--output_dir", "data/wan_finetune_test",

fastvideo/tests/training/lora/test_lora_training.py

@@ -60,7 +60,6 @@ def test_lora_training():
         "--validation_sampling_steps", "50",
         "--log_validation",
         "--checkpoints_total_limit", "3",
-        "--allow_tf32",
         "--ema_start_step", "0",
         "--training_cfg_rate", "0.1",
         "--output_dir", "/workspace",