add a config variable that disable VRAM OOM conditions

invokeai/app/services/config/config_default.py

@@ -103,6 +103,7 @@ class InvokeAIAppConfig(BaseSettings):
         convert_cache: Maximum size of on-disk converted models cache (GB).
         lazy_offload: Keep models in VRAM until their space is needed.
         log_memory_usage: If True, a memory snapshot will be captured before and after every model cache operation, and the result will be logged (at debug level). There is a time cost to capturing the memory snapshots, so it is recommended to only enable this feature if you are actively inspecting the model cache's behaviour.
+        disable_vram_check: If True, disable the check for sufficient VRAM memory prior to loading a model. This may lead to unpredictable behavior, so use for debugging memory problems only.
         device: Preferred execution device. `auto` will choose the device depending on the hardware platform and the installed torch capabilities.<br>Valid values: `auto`, `cpu`, `cuda`, `cuda:1`, `mps`
         precision: Floating point precision. `float16` will consume half the memory of `float32` but produce slightly lower-quality images. The `auto` setting will guess the proper precision based on your video card and operating system.<br>Valid values: `auto`, `float16`, `bfloat16`, `float32`, `autocast`
         sequential_guidance: Whether to calculate guidance in serial instead of in parallel, lowering memory requirements.
@@ -171,6 +172,7 @@ class InvokeAIAppConfig(BaseSettings):
     convert_cache:                float = Field(default=DEFAULT_CONVERT_CACHE, ge=0, description="Maximum size of on-disk converted models cache (GB).")
     lazy_offload:                  bool = Field(default=True,               description="Keep models in VRAM until their space is needed.")
     log_memory_usage:              bool = Field(default=False,              description="If True, a memory snapshot will be captured before and after every model cache operation, and the result will be logged (at debug level). There is a time cost to capturing the memory snapshots, so it is recommended to only enable this feature if you are actively inspecting the model cache's behaviour.")
+    disable_vram_check:            bool = Field(default=False,              description="If True, disable the check for sufficient VRAM memory prior to loading a model. This may lead to unpredictable behavior, so use for debugging memory problems only.")
 
     # DEVICE
     device:                      DEVICE = Field(default="auto",             description="Preferred execution device. `auto` will choose the device depending on the hardware platform and the installed torch capabilities.")

invokeai/app/services/model_manager/model_manager_default.py

@@ -82,6 +82,7 @@ class ModelManagerService(ModelManagerServiceBase):
             max_vram_cache_size=app_config.vram,
             logger=logger,
             execution_device=execution_device,
+            disable_memory_check=app_config.disable_vram_check,
         )
         convert_cache = ModelConvertCache(cache_path=app_config.convert_cache_path, max_size=app_config.convert_cache)
         loader = ModelLoadService(

invokeai/backend/model_manager/load/model_cache/model_cache_default.py

@@ -68,6 +68,7 @@ class ModelCache(ModelCacheBase[AnyModel]):
         sha_chunksize: int = 16777216,
         log_memory_usage: bool = False,
         logger: Optional[Logger] = None,
+        disable_memory_check: bool = False,
     ):
         """
         Initialize the model RAM cache.
@@ -82,6 +83,7 @@ class ModelCache(ModelCacheBase[AnyModel]):
             operation, and the result will be logged (at debug level). There is a time cost to capturing the memory
             snapshots, so it is recommended to disable this feature unless you are actively inspecting the model cache's
             behaviour.
+        :param disable_memory_check: If True disable the check for insufficient VRAM when loading a model.
         """
         # allow lazy offloading only when vram cache enabled
         self._lazy_offloading = lazy_offloading and max_vram_cache_size > 0
@@ -93,6 +95,7 @@ class ModelCache(ModelCacheBase[AnyModel]):
         self._logger = logger or InvokeAILogger.get_logger(self.__class__.__name__)
         self._log_memory_usage = log_memory_usage
         self._stats: Optional[CacheStats] = None
+        self._disable_memory_check = disable_memory_check
 
         self._cached_models: Dict[str, CacheRecord[AnyModel]] = {}
         self._cache_stack: List[str] = []
@@ -270,6 +273,9 @@ class ModelCache(ModelCacheBase[AnyModel]):
             return
 
         # may raise an exception here if insufficient GPU VRAM
+        if self._disable_memory_check:
+            self.logger.warning("VRAM memory check disabled. Unpredictable behavior may result.")
+        else:
             self._check_free_vram(target_device, cache_entry.size)
 
         start_model_to_time = time.time()

tests/test_docs.py

@@ -3,7 +3,7 @@ from scripts.update_config_docstring import generate_config_docstrings
 
 
 def test_app_config_docstrings_are_current():
-    # If this test fails, run `python scripts/generate_config_docstring.py`. See the comments in that script for
+    # If this test fails, run `python scripts/update_config_docstring.py`. See the comments in that script for
     # an explanation of why this is necessary.
     #
     # A make target is provided to run the script: `make update-config-docstring`.