mv to utils

examples/train/configs/genrl_wan2.1_t2v_1.3B_longcat.yaml

@@ -115,8 +115,7 @@ training:
     enable_gradient_checkpointing_type: full
 
 callbacks:
-  grad_clip:
-    max_grad_norm: 0.0  # Disabled; GenRLMethod clips internally.
+  # Gradnorm call back Disabled; GenRLMethod clips internally.
   ema:
     decay: 0.9
     start_iter: 0

fastvideo/train/methods/rl/genrl.py

@@ -31,27 +31,27 @@ from fastvideo.train.methods.base import (
     LogScalar,
     TrainingMethod,
 )
-from fastvideo.train.methods.rl.advantages import (
+from fastvideo.train.methods.rl.utils.advantages import (
     compute_advantages,
 )
-from fastvideo.train.methods.rl.data import (
+from fastvideo.train.methods.rl.utils.data import (
     build_prompt_dataloaders,
 )
-from fastvideo.train.methods.rl.diffusion import (
+from fastvideo.train.methods.rl.utils.diffusion import (
     compute_log_prob,
 )
-from fastvideo.train.methods.rl.embeddings import (
+from fastvideo.train.methods.rl.utils.embeddings import (
     compute_text_embeddings,
 )
-from fastvideo.train.methods.rl.rewards import (
+from fastvideo.train.methods.rl.utils.rewards import (
     move_reward_models,
     multi_score,
     reward_models_on_device,
 )
-from fastvideo.train.methods.rl.sampling import (
+from fastvideo.train.methods.rl.utils.sampling import (
     sample_epoch,
 )
-from fastvideo.train.methods.rl.stat_tracking import (
+from fastvideo.train.methods.rl.utils.stat_tracking import (
     PerPromptStatTracker,
 )
 from fastvideo.train.models.base import ModelBase

fastvideo/train/methods/rl/utils/__init__.py

@@ -0,0 +1,2 @@
+# SPDX-License-Identifier: Apache-2.0
+"""Utility modules for RL training."""

fastvideo/train/methods/rl/utils/advantages.py

@@ -8,7 +8,7 @@ from typing import Any
 import numpy as np
 
 from fastvideo.logger import init_logger
-from fastvideo.train.methods.rl.stat_tracking import (
+from fastvideo.train.methods.rl.utils.stat_tracking import (
     EPSILON,
     PerPromptStatTracker,
 )

fastvideo/train/methods/rl/utils/diffusion.py

@@ -5,7 +5,7 @@ from __future__ import annotations
 
 import torch
 
-from fastvideo.train.methods.rl.sde import (
+from fastvideo.train.methods.rl.utils.sde import (
     sde_step_with_logprob,
 )
 

fastvideo/train/methods/rl/utils/evaluation.py

@@ -9,10 +9,10 @@ from typing import Any
 import torch
 
 from fastvideo.logger import init_logger
-from fastvideo.train.methods.rl.embeddings import (
+from fastvideo.train.methods.rl.utils.embeddings import (
     compute_text_embeddings,
 )
-from fastvideo.train.methods.rl.pipeline import (
+from fastvideo.train.methods.rl.utils.pipeline import (
     wan_denoising_with_logprob,
 )
 

fastvideo/train/methods/rl/utils/pipeline.py

@@ -19,7 +19,7 @@ from fastvideo.logger import init_logger
 
 _pipeline_logger = init_logger(__name__)
 
-from fastvideo.train.methods.rl.sde import (
+from fastvideo.train.methods.rl.utils.sde import (
     sde_step_with_logprob,
 )
 

fastvideo/train/methods/rl/utils/sampling.py

@@ -11,10 +11,10 @@ from typing import Any
 import torch
 
 from fastvideo.logger import init_logger
-from fastvideo.train.methods.rl.embeddings import (
+from fastvideo.train.methods.rl.utils.embeddings import (
     compute_text_embeddings,
 )
-from fastvideo.train.methods.rl.pipeline import (
+from fastvideo.train.methods.rl.utils.pipeline import (
     wan_denoising_with_logprob,
 )