remove the LoWRA LORA, which was deleted from civitai

invokeai/app/services/config/config_default.py

@@ -45,7 +45,6 @@ InvokeAI:
     ram: 13.5
     vram: 0.25
     lazy_offload: true
-    log_memory_usage: false
   Device:
     device: auto
     precision: auto
@@ -262,7 +261,6 @@ class InvokeAIAppConfig(InvokeAISettings):
     ram                 : float = Field(default=7.5, gt=0, description="Maximum memory amount used by model cache for rapid switching (floating point number, GB)", json_schema_extra=Categories.ModelCache, )
     vram                : float = Field(default=0.25, ge=0, description="Amount of VRAM reserved for model storage (floating point number, GB)", json_schema_extra=Categories.ModelCache, )
     lazy_offload        : bool = Field(default=True, description="Keep models in VRAM until their space is needed", json_schema_extra=Categories.ModelCache, )
-    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.", json_schema_extra=Categories.ModelCache)
 
     # DEVICE
     device              : Literal["auto", "cpu", "cuda", "cuda:1", "mps"] = Field(default="auto", description="Generation device", json_schema_extra=Categories.Device)

invokeai/backend/model_management/memory_snapshot.py

@@ -64,7 +64,7 @@ class MemorySnapshot:
         return cls(process_ram, vram, malloc_info)
 
 
-def get_pretty_snapshot_diff(snapshot_1: Optional[MemorySnapshot], snapshot_2: Optional[MemorySnapshot]) -> str:
+def get_pretty_snapshot_diff(snapshot_1: MemorySnapshot, snapshot_2: MemorySnapshot) -> str:
     """Get a pretty string describing the difference between two `MemorySnapshot`s."""
 
     def get_msg_line(prefix: str, val1: int, val2: int):
@@ -73,9 +73,6 @@ def get_pretty_snapshot_diff(snapshot_1: Optional[MemorySnapshot], snapshot_2: O
 
     msg = ""
 
-    if snapshot_1 is None or snapshot_2 is None:
-        return msg
-
     msg += get_msg_line("Process RAM", snapshot_1.process_ram, snapshot_2.process_ram)
 
     if snapshot_1.malloc_info is not None and snapshot_2.malloc_info is not None:

invokeai/backend/model_management/model_cache.py

@@ -117,7 +117,6 @@ class ModelCache(object):
         lazy_offloading: bool = True,
         sha_chunksize: int = 16777216,
         logger: types.ModuleType = logger,
-        log_memory_usage: bool = False,
     ):
         """
         :param max_cache_size: Maximum size of the RAM cache [6.0 GB]
@@ -127,10 +126,6 @@ class ModelCache(object):
         :param lazy_offloading: Keep model in VRAM until another model needs to be loaded
         :param sequential_offload: Conserve VRAM by loading and unloading each stage of the pipeline sequentially
         :param sha_chunksize: Chunksize to use when calculating sha256 model hash
-        :param 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 disable this feature unless you are actively inspecting the model cache's
-            behaviour.
         """
         self.model_infos: Dict[str, ModelBase] = dict()
         # allow lazy offloading only when vram cache enabled
@@ -142,7 +137,6 @@ class ModelCache(object):
         self.storage_device: torch.device = storage_device
         self.sha_chunksize = sha_chunksize
         self.logger = logger
-        self._log_memory_usage = log_memory_usage
 
         # used for stats collection
         self.stats = None
@@ -150,11 +144,6 @@ class ModelCache(object):
         self._cached_models = dict()
         self._cache_stack = list()
 
-    def _capture_memory_snapshot(self) -> Optional[MemorySnapshot]:
-        if self._log_memory_usage:
-            return MemorySnapshot.capture()
-        return None
-
     def get_key(
         self,
         model_path: str,
@@ -234,10 +223,10 @@ class ModelCache(object):
 
             # Load the model from disk and capture a memory snapshot before/after.
             start_load_time = time.time()
-            snapshot_before = self._capture_memory_snapshot()
+            snapshot_before = MemorySnapshot.capture()
             with skip_torch_weight_init():
                 model = model_info.get_model(child_type=submodel, torch_dtype=self.precision)
-            snapshot_after = self._capture_memory_snapshot()
+            snapshot_after = MemorySnapshot.capture()
             end_load_time = time.time()
 
             self_reported_model_size_after_load = model_info.get_size(submodel)
@@ -286,9 +275,9 @@ class ModelCache(object):
             return
 
         start_model_to_time = time.time()
-        snapshot_before = self._capture_memory_snapshot()
+        snapshot_before = MemorySnapshot.capture()
         cache_entry.model.to(target_device)
-        snapshot_after = self._capture_memory_snapshot()
+        snapshot_after = MemorySnapshot.capture()
         end_model_to_time = time.time()
         self.logger.debug(
             f"Moved model '{key}' from {source_device} to"
@@ -297,12 +286,7 @@ class ModelCache(object):
             f"{get_pretty_snapshot_diff(snapshot_before, snapshot_after)}"
         )
 
-        if (
-            snapshot_before is not None
-            and snapshot_after is not None
-            and snapshot_before.vram is not None
-            and snapshot_after.vram is not None
-        ):
+        if snapshot_before.vram is not None and snapshot_after.vram is not None:
             vram_change = abs(snapshot_before.vram - snapshot_after.vram)
 
             # If the estimated model size does not match the change in VRAM, log a warning.
@@ -438,17 +422,12 @@ class ModelCache(object):
         self.logger.debug(f"Before unloading: cached_models={len(self._cached_models)}")
 
         pos = 0
-        models_cleared = 0
         while current_size + bytes_needed > maximum_size and pos < len(self._cache_stack):
             model_key = self._cache_stack[pos]
             cache_entry = self._cached_models[model_key]
 
             refs = sys.getrefcount(cache_entry.model)
 
-            # HACK: This is a workaround for a memory-management issue that we haven't tracked down yet. We are directly
-            # going against the advice in the Python docs by using `gc.get_referrers(...)` in this way:
-            # https://docs.python.org/3/library/gc.html#gc.get_referrers
-
             # manualy clear local variable references of just finished function calls
             # for some reason python don't want to collect it even by gc.collect() immidiately
             if refs > 2:
@@ -474,16 +453,15 @@ class ModelCache(object):
                 f" refs: {refs}"
             )
 
-            # Expected refs:
+            # 2 refs:
             # 1 from cache_entry
             # 1 from getrefcount function
             # 1 from onnx runtime object
-            if not cache_entry.locked and refs <= (3 if "onnx" in model_key else 2):
+            if not cache_entry.locked and refs <= 3 if "onnx" in model_key else 2:
                 self.logger.debug(
                     f"Unloading model {model_key} to free {(model_size/GIG):.2f} GB (-{(cache_entry.size/GIG):.2f} GB)"
                 )
                 current_size -= cache_entry.size
-                models_cleared += 1
                 if self.stats:
                     self.stats.cleared += 1
                 del self._cache_stack[pos]
@@ -493,20 +471,7 @@ class ModelCache(object):
             else:
                 pos += 1
 
-        if models_cleared > 0:
-            # There would likely be some 'garbage' to be collected regardless of whether a model was cleared or not, but
-            # there is a significant time cost to calling `gc.collect()`, so we want to use it sparingly. (The time cost
-            # is high even if no garbage gets collected.)
-            #
-            # Calling gc.collect(...) when a model is cleared seems like a good middle-ground:
-            # - If models had to be cleared, it's a signal that we are close to our memory limit.
-            # - If models were cleared, there's a good chance that there's a significant amount of garbage to be
-            #   collected.
-            #
-            # Keep in mind that gc is only responsible for handling reference cycles. Most objects should be cleaned up
-            # immediately when their reference count hits 0.
-            gc.collect()
-
+        gc.collect()
         torch.cuda.empty_cache()
         if choose_torch_device() == torch.device("mps"):
             mps.empty_cache()
@@ -526,6 +491,7 @@ class ModelCache(object):
                 vram_in_use = torch.cuda.memory_allocated()
                 self.logger.debug(f"{(vram_in_use/GIG):.2f}GB VRAM used for models; max allowed={(reserved/GIG):.2f}GB")
 
+        gc.collect()
         torch.cuda.empty_cache()
         if choose_torch_device() == torch.device("mps"):
             mps.empty_cache()

invokeai/backend/model_management/model_load_optimizations.py

@@ -17,7 +17,7 @@ def skip_torch_weight_init():
     completely unnecessary if the intent is to load checkpoint weights from disk for the layer. This context manager
     monkey-patches common torch layers to skip the weight initialization step.
     """
-    torch_modules = [torch.nn.Linear, torch.nn.modules.conv._ConvNd, torch.nn.Embedding]
+    torch_modules = [torch.nn.Linear, torch.nn.modules.conv._ConvNd]
     saved_functions = [m.reset_parameters for m in torch_modules]
 
     try:

invokeai/backend/model_management/model_manager.py

@@ -351,7 +351,6 @@ class ModelManager(object):
             precision=precision,
             sequential_offload=sequential_offload,
             logger=logger,
-            log_memory_usage=self.app_config.log_memory_usage,
         )
 
         self._read_models(config)

invokeai/configs/INITIAL_MODELS.yaml

@@ -117,9 +117,6 @@ sd-1/embedding/EasyNegative:
    recommended: True
 sd-1/embedding/ahx-beta-453407d:
    repo_id: sd-concepts-library/ahx-beta-453407d
-sd-1/lora/LowRA:
-   path: https://civitai.com/api/download/models/63006
-   recommended: True
 sd-1/lora/Ink scenery:
    path: https://civitai.com/api/download/models/83390
 sd-1/ip_adapter/ip_adapter_sd15:

invokeai/frontend/merge/merge_diffusers.py

@@ -90,7 +90,6 @@ def _parse_args() -> Namespace:
 # ------------------------- GUI HERE -------------------------
 class mergeModelsForm(npyscreen.FormMultiPageAction):
     interpolations = ["weighted_sum", "sigmoid", "inv_sigmoid"]
-    bases = ["sd-1", "sd-2", "sdxl"]
 
     def __init__(self, parentApp, name):
         self.parentApp = parentApp
@@ -277,7 +276,7 @@ class mergeModelsForm(npyscreen.FormMultiPageAction):
 
         args = dict(
             model_names=models,
-            base_model=BaseModelType(self.bases[self.base_select.value[0]]),
+            base_model=tuple(BaseModelType)[self.base_select.value[0]],
             alpha=self.alpha.value,
             interp=interp,
             force=self.force.value,
@@ -320,7 +319,8 @@ class mergeModelsForm(npyscreen.FormMultiPageAction):
         return sorted(model_names)
 
     def _populate_models(self, value=None):
-        base_model = BaseModelType(self.bases[value[0]])
+        bases = ["sd-1", "sd-2", "sdxl"]
+        base_model = BaseModelType(bases[value[0]])
         self.model_names = self.get_model_names(base_model)
 
         models_plus_none = self.model_names.copy()

invokeai/frontend/web/public/locales/en.json

@@ -722,9 +722,7 @@
         "noMatchingModels": "No matching Models",
         "noModelsAvailable": "No models available",
         "selectLoRA": "Select a LoRA",
-        "selectModel": "Select a Model",
-        "noLoRAsInstalled": "No LoRAs installed",
-        "noRefinerModelsInstalled": "No SDXL Refiner models installed"
+        "selectModel": "Select a Model"
     },
     "nodes": {
         "addNode": "Add Node",

invokeai/frontend/web/src/features/lora/components/ParamLoraSelect.tsx

@@ -10,7 +10,6 @@ import { loraAdded } from 'features/lora/store/loraSlice';
 import { MODEL_TYPE_MAP } from 'features/parameters/types/constants';
 import { forEach } from 'lodash-es';
 import { memo, useCallback, useMemo } from 'react';
-import { useTranslation } from 'react-i18next';
 import { useGetLoRAModelsQuery } from 'services/api/endpoints/models';
 
 const selector = createSelector(
@@ -25,7 +24,7 @@ const ParamLoRASelect = () => {
   const dispatch = useAppDispatch();
   const { loras } = useAppSelector(selector);
   const { data: loraModels } = useGetLoRAModelsQuery();
-  const { t } = useTranslation();
+
   const currentMainModel = useAppSelector(
     (state: RootState) => state.generation.model
   );
@@ -80,7 +79,7 @@ const ParamLoRASelect = () => {
     return (
       <Flex sx={{ justifyContent: 'center', p: 2 }}>
         <Text sx={{ fontSize: 'sm', color: 'base.500', _dark: 'base.700' }}>
-          {t('models.noLoRAsInstalled')}
+          No LoRAs Loaded
         </Text>
       </Flex>
     );

invokeai/frontend/web/src/features/sdxl/components/ParamSDXLRefinerCollapse.tsx

@@ -1,4 +1,4 @@
-import { Flex, Text } from '@chakra-ui/react';
+import { Flex } from '@chakra-ui/react';
 import { createSelector } from '@reduxjs/toolkit';
 import { stateSelector } from 'app/store/store';
 import { useAppSelector } from 'app/store/storeHooks';
@@ -14,7 +14,6 @@ import ParamSDXLRefinerStart from './SDXLRefiner/ParamSDXLRefinerStart';
 import ParamSDXLRefinerSteps from './SDXLRefiner/ParamSDXLRefinerSteps';
 import ParamUseSDXLRefiner from './SDXLRefiner/ParamUseSDXLRefiner';
 import { useTranslation } from 'react-i18next';
-import { useIsRefinerAvailable } from 'services/api/hooks/useIsRefinerAvailable';
 
 const selector = createSelector(
   stateSelector,
@@ -32,19 +31,6 @@ const selector = createSelector(
 const ParamSDXLRefinerCollapse = () => {
   const { activeLabel, shouldUseSliders } = useAppSelector(selector);
   const { t } = useTranslation();
-  const isRefinerAvailable = useIsRefinerAvailable();
-
-  if (!isRefinerAvailable) {
-    return (
-      <IAICollapse label={t('sdxl.refiner')} activeLabel={activeLabel}>
-        <Flex sx={{ justifyContent: 'center', p: 2 }}>
-          <Text sx={{ fontSize: 'sm', color: 'base.500', _dark: 'base.700' }}>
-            {t('models.noRefinerModelsInstalled')}
-          </Text>
-        </Flex>
-      </IAICollapse>
-    );
-  }
 
   return (
     <IAICollapse label={t('sdxl.refiner')} activeLabel={activeLabel}>

tests/backend/model_management/test_memory_snapshot.py

@@ -13,11 +13,10 @@ def test_memory_snapshot_capture():
 
 
 snapshots = [
-    MemorySnapshot(process_ram=1, vram=2, malloc_info=Struct_mallinfo2()),
-    MemorySnapshot(process_ram=1, vram=2, malloc_info=None),
-    MemorySnapshot(process_ram=1, vram=None, malloc_info=Struct_mallinfo2()),
-    MemorySnapshot(process_ram=1, vram=None, malloc_info=None),
-    None,
+    MemorySnapshot(process_ram=1.0, vram=2.0, malloc_info=Struct_mallinfo2()),
+    MemorySnapshot(process_ram=1.0, vram=2.0, malloc_info=None),
+    MemorySnapshot(process_ram=1.0, vram=None, malloc_info=Struct_mallinfo2()),
+    MemorySnapshot(process_ram=1.0, vram=None, malloc_info=None),
 ]
 
 
@@ -27,12 +26,10 @@ def test_get_pretty_snapshot_diff(snapshot_1, snapshot_2):
     """Test that get_pretty_snapshot_diff() works with various combinations of missing MemorySnapshot fields."""
     msg = get_pretty_snapshot_diff(snapshot_1, snapshot_2)
 
-    expected_lines = 0
-    if snapshot_1 is not None and snapshot_2 is not None:
+    expected_lines = 1
+    if snapshot_1.vram is not None and snapshot_2.vram is not None:
         expected_lines += 1
-        if snapshot_1.vram is not None and snapshot_2.vram is not None:
-            expected_lines += 1
-        if snapshot_1.malloc_info is not None and snapshot_2.malloc_info is not None:
-            expected_lines += 5
+    if snapshot_1.malloc_info is not None and snapshot_2.malloc_info is not None:
+        expected_lines += 5
 
     assert len(msg.splitlines()) == expected_lines

tests/backend/model_management/test_model_load_optimization.py

@@ -11,7 +11,6 @@ from invokeai.backend.model_management.model_load_optimizations import _no_op, s
         (torch.nn.Conv1d, {"in_channels": 10, "out_channels": 20, "kernel_size": 3}),
         (torch.nn.Conv2d, {"in_channels": 10, "out_channels": 20, "kernel_size": 3}),
         (torch.nn.Conv3d, {"in_channels": 10, "out_channels": 20, "kernel_size": 3}),
-        (torch.nn.Embedding, {"num_embeddings": 10, "embedding_dim": 10}),
     ],
 )
 def test_skip_torch_weight_init_linear(torch_module, layer_args):
@@ -37,14 +36,12 @@ def test_skip_torch_weight_init_linear(torch_module, layer_args):
     # Check that reset_parameters is skipped while `skip_torch_weight_init()` is active.
     assert reset_params_fn_during == _no_op
     assert not torch.allclose(layer_before.weight, layer_during.weight)
-    if hasattr(layer_before, "bias"):
-        assert not torch.allclose(layer_before.bias, layer_during.bias)
+    assert not torch.allclose(layer_before.bias, layer_during.bias)
 
     # Check that the original behavior is restored after `skip_torch_weight_init()` ends.
     assert reset_params_fn_before is reset_params_fn_after
     assert torch.allclose(layer_before.weight, layer_after.weight)
-    if hasattr(layer_before, "bias"):
-        assert torch.allclose(layer_before.bias, layer_after.bias)
+    assert torch.allclose(layer_before.bias, layer_after.bias)
 
 
 def test_skip_torch_weight_init_restores_base_class_behavior():