reduce mandatory requirements

Signed-off-by: Vladimir Mandic <mandic00@live.com>

modules/seedvr/src/common/config.py

@@ -1,7 +1,12 @@
 import importlib
 from typing import Any, Callable, List, Union
-from omegaconf import DictConfig, ListConfig, OmegaConf
 
+try:
+    from installer import install
+    install('omegaconf')
+    from omegaconf import DictConfig, ListConfig, OmegaConf
+except Exception as e:
+    raise ImportError(f"Failed to import omegaconf. Error: {e}") from e
 
 try:
     OmegaConf.register_new_resolver("eval", eval)
@@ -10,7 +15,7 @@ except Exception as e:
         raise
 
 
-def load_config(path: str, argv: List[str] = None) -> Union[DictConfig, ListConfig]:
+def load_config(path: str, argv: List[str] = None):
     """
     Load a configuration. Will resolve inheritance.
     """
@@ -25,7 +30,7 @@ def load_config(path: str, argv: List[str] = None) -> Union[DictConfig, ListConf
 
 def resolve_recursive(
     config: Any,
-    resolver: Callable[[Union[DictConfig, ListConfig]], Union[DictConfig, ListConfig]],
+    resolver: Callable[[Any], Any],
 ) -> Any:
     config = resolver(config)
     if isinstance(config, DictConfig):
@@ -41,7 +46,7 @@ def resolve_recursive(
     return config
 
 
-def resolve_inheritance(config: Union[DictConfig, ListConfig]) -> Any:
+def resolve_inheritance(config: Any) -> Any:
     """
     Recursively resolve inheritance if the config contains:
     __inherit__: path/to/parent.yaml or a ListConfig of such paths.
@@ -104,7 +109,7 @@ def import_item(path: Union[str, List[str]], name: str) -> Any:
         raise ValueError(f"Path must be string or list of strings, got: {type(path)}")
 
 
-def create_object(config: DictConfig) -> Any:
+def create_object(config: Any) -> Any:
     """
     Create an object from config.
     The config is expected to contains the following:

modules/seedvr/src/common/diffusion/config.py

@@ -17,7 +17,6 @@ Utility functions for creating schedules and samplers from config.
 """
 
 import torch
-from omegaconf import DictConfig
 
 from .samplers.base import Sampler
 from .samplers.euler import EulerSampler
@@ -28,7 +27,7 @@ from .timesteps.sampling.trailing import UniformTrailingSamplingTimesteps
 
 
 def create_schedule_from_config(
-    config: DictConfig,
+    config,
 ) -> Schedule:
     """
     Create a schedule from configuration.
@@ -40,7 +39,7 @@ def create_schedule_from_config(
 
 
 def create_sampler_from_config(
-    config: DictConfig,
+    config,
     schedule: Schedule,
     timesteps: SamplingTimesteps,
 ) -> Sampler:
@@ -57,7 +56,7 @@ def create_sampler_from_config(
 
 
 def create_sampling_timesteps_from_config(
-    config: DictConfig,
+    config,
     schedule: Schedule,
     device: torch.device,
 ) -> SamplingTimesteps:

modules/seedvr/src/core/infer.py

@@ -1,7 +1,6 @@
 from typing import List, Optional, Tuple, Union
 import torch
 from einops import rearrange
-from omegaconf import DictConfig, ListConfig
 from ..common.diffusion import classifier_free_guidance_dispatcher, create_sampler_from_config, create_sampling_timesteps_from_config, create_schedule_from_config
 from ..models.dit_v2 import na
 
@@ -40,7 +39,9 @@ def optimized_channels_to_second(tensor):
 
 
 class VideoDiffusionInfer():
-    def __init__(self, config: DictConfig, device: str, dtype: torch.dtype):
+    def __init__(self, config, device: str, dtype: torch.dtype):
+        from installer import install
+        install('omegaconf')
         self.config = config
         self.device = device
         self.dtype = dtype
@@ -48,6 +49,7 @@ class VideoDiffusionInfer():
         self.dit = None
         self.sampler = None
         self.schedule = None
+
     def get_condition(self, latent: torch.Tensor, latent_blur: torch.Tensor, task: str) -> torch.Tensor:
         t, h, w, c = latent.shape
         cond = torch.zeros([t, h, w, c + 1], device=latent.device, dtype=latent.dtype)
@@ -93,6 +95,7 @@ class VideoDiffusionInfer():
 
     @torch.no_grad()
     def vae_encode(self, samples: List[torch.Tensor]) -> List[torch.Tensor]:
+        from omegaconf import ListConfig
         use_sample = self.config.vae.get("use_sample", True)
         latents = []
         if len(samples) > 0:
@@ -138,6 +141,7 @@ class VideoDiffusionInfer():
     @torch.no_grad()
     def vae_decode(self, latents: List[torch.Tensor], target_dtype: torch.dtype = None) -> List[torch.Tensor]:
         """🚀 VAE decode optimisé - décodage direct sans chunking, compatible avec autocast externe"""
+        from omegaconf import ListConfig
         samples = []
         if len(latents) > 0:
             device = self.device

modules/seedvr/src/core/model_manager.py

@@ -1,6 +1,5 @@
 import os
 import torch
-from omegaconf import OmegaConf
 from safetensors.torch import load_file as load_safetensors_file
 from huggingface_hub import hf_hub_download
 from ..optimization.memory_manager import preinitialize_rope_cache
@@ -9,6 +8,10 @@ from ..core.infer import VideoDiffusionInfer
 
 
 def configure_runner(model_name, cache_dir, device:str='cpu', dtype:torch.dtype=None):
+    from installer import install
+    install('omegaconf')
+    from omegaconf import OmegaConf
+
     repo_id = "vladmandic/SeedVR2"
     script_directory = os.path.dirname(os.path.dirname(os.path.dirname(os.path.abspath(__file__))))
     config_path = os.path.join(script_directory, './config_7b.yaml') if "7b" in model_name else os.path.join(script_directory, './config_3b.yaml')