First draft

## What type of PR is this? (check all applicable)

- [x] Refactor
- [x] Feature
- [x] Bug Fix
- [?] Optimization


## Have you discussed this change with the InvokeAI team?
- [x] No

     
## Description

- Fixed filter type select using `images` instead of `all` -- Probably
some merge conflict.
- Added loading state for model lists. Handy when the model list takes
longer than a second for any reason to fetch. Better to show this than
an empty screen.
- Refactored the render model list function so we modify the display
component in one area rather than have repeated code.

### Other Issues

- The list can get a bit laggy on initial load when you have a hundreds
of models / loras. This needs to be fixed. Will make another PR for
this.

flake.nix

@@ -34,6 +34,10 @@
             cudaPackages.cudnn
             cudaPackages.cuda_nvrtc
             cudatoolkit
+            pkgconfig
+            libconfig
+            cmake
+            blas
             freeglut
             glib
             gperf
@@ -42,6 +46,12 @@
             libGLU
             linuxPackages.nvidia_x11
             python
+            (opencv4.override {
+              enableGtk3 = true;
+              enableFfmpeg = true;
+              enableCuda = true;
+              enableUnfree = true;
+            })
             stdenv.cc
             stdenv.cc.cc.lib
             xorg.libX11

invokeai/app/invocations/compel.py

@@ -108,14 +108,15 @@ class CompelInvocation(BaseInvocation):
         for trigger in re.findall(r"<[a-zA-Z0-9., _-]+>", self.prompt):
             name = trigger[1:-1]
             try:
-                ti_list.append(
+                ti_list.append((
+                    name,
                     context.services.model_manager.get_model(
                         model_name=name,
                         base_model=self.clip.text_encoder.base_model,
                         model_type=ModelType.TextualInversion,
                         context=context,
                     ).context.model
-                )
+                ))
             except ModelNotFoundException:
                 # print(e)
                 # import traceback
@@ -196,14 +197,15 @@ class SDXLPromptInvocationBase:
         for trigger in re.findall(r"<[a-zA-Z0-9., _-]+>", prompt):
             name = trigger[1:-1]
             try:
-                ti_list.append(
+                ti_list.append((
+                    name,
                     context.services.model_manager.get_model(
                         model_name=name,
                         base_model=clip_field.text_encoder.base_model,
                         model_type=ModelType.TextualInversion,
                         context=context,
                     ).context.model
-                )
+                ))
             except ModelNotFoundException:
                 # print(e)
                 # import traceback
@@ -270,14 +272,15 @@ class SDXLPromptInvocationBase:
         for trigger in re.findall(r"<[a-zA-Z0-9., _-]+>", prompt):
             name = trigger[1:-1]
             try:
-                ti_list.append(
+                ti_list.append((
+                    name,
                     context.services.model_manager.get_model(
                         model_name=name,
                         base_model=clip_field.text_encoder.base_model,
                         model_type=ModelType.TextualInversion,
                         context=context,
                     ).context.model
-                )
+                ))
             except ModelNotFoundException:
                 # print(e)
                 # import traceback

invokeai/app/invocations/onnx.py

@@ -65,7 +65,6 @@ class ONNXPromptInvocation(BaseInvocation):
             **self.clip.text_encoder.dict(),
         )
         with tokenizer_info as orig_tokenizer, text_encoder_info as text_encoder, ExitStack() as stack:
-            # loras = [(stack.enter_context(context.services.model_manager.get_model(**lora.dict(exclude={"weight"}))), lora.weight) for lora in self.clip.loras]
             loras = [
                 (context.services.model_manager.get_model(**lora.dict(exclude={"weight"})).context.model, lora.weight)
                 for lora in self.clip.loras
@@ -75,20 +74,14 @@ class ONNXPromptInvocation(BaseInvocation):
             for trigger in re.findall(r"<[a-zA-Z0-9., _-]+>", self.prompt):
                 name = trigger[1:-1]
                 try:
-                    ti_list.append(
-                        # stack.enter_context(
-                        #    context.services.model_manager.get_model(
-                        #        model_name=name,
-                        #        base_model=self.clip.text_encoder.base_model,
-                        #        model_type=ModelType.TextualInversion,
-                        #    )
-                        # )
+                    ti_list.append((
+                        name,
                         context.services.model_manager.get_model(
                             model_name=name,
                             base_model=self.clip.text_encoder.base_model,
                             model_type=ModelType.TextualInversion,
                         ).context.model
-                    )
+                    ))
                 except Exception:
                     # print(e)
                     # import traceback

invokeai/backend/model_management/lora.py

@@ -562,7 +562,7 @@ class ModelPatcher:
         cls,
         tokenizer: CLIPTokenizer,
         text_encoder: CLIPTextModel,
-        ti_list: List[Any],
+        ti_list: List[Tuple[str, Any]],
     ) -> Tuple[CLIPTokenizer, TextualInversionManager]:
         init_tokens_count = None
         new_tokens_added = None
@@ -572,27 +572,27 @@ class ModelPatcher:
             ti_manager = TextualInversionManager(ti_tokenizer)
             init_tokens_count = text_encoder.resize_token_embeddings(None).num_embeddings
 
-            def _get_trigger(ti, index):
-                trigger = ti.name
+            def _get_trigger(ti_name, index):
+                trigger = ti_name
                 if index > 0:
                     trigger += f"-!pad-{i}"
                 return f"<{trigger}>"
 
             # modify tokenizer
             new_tokens_added = 0
-            for ti in ti_list:
+            for ti_name, ti in ti_list:
                 for i in range(ti.embedding.shape[0]):
-                    new_tokens_added += ti_tokenizer.add_tokens(_get_trigger(ti, i))
+                    new_tokens_added += ti_tokenizer.add_tokens(_get_trigger(ti_name, i))
 
             # modify text_encoder
             text_encoder.resize_token_embeddings(init_tokens_count + new_tokens_added)
             model_embeddings = text_encoder.get_input_embeddings()
 
-            for ti in ti_list:
+            for ti_name, ti in ti_list:
                 ti_tokens = []
                 for i in range(ti.embedding.shape[0]):
                     embedding = ti.embedding[i]
-                    trigger = _get_trigger(ti, i)
+                    trigger = _get_trigger(ti_name, i)
 
                     token_id = ti_tokenizer.convert_tokens_to_ids(trigger)
                     if token_id == ti_tokenizer.unk_token_id:
@@ -637,7 +637,6 @@ class ModelPatcher:
 
 
 class TextualInversionModel:
-    name: str
     embedding: torch.Tensor  # [n, 768]|[n, 1280]
 
     @classmethod
@@ -651,7 +650,6 @@ class TextualInversionModel:
             file_path = Path(file_path)
 
         result = cls()  # TODO:
-        result.name = file_path.stem  # TODO:
 
         if file_path.suffix == ".safetensors":
             state_dict = load_file(file_path.absolute().as_posix(), device="cpu")
@@ -828,7 +826,7 @@ class ONNXModelPatcher:
         cls,
         tokenizer: CLIPTokenizer,
         text_encoder: IAIOnnxRuntimeModel,
-        ti_list: List[Any],
+        ti_list: List[Tuple[str, Any]],
     ) -> Tuple[CLIPTokenizer, TextualInversionManager]:
         from .models.base import IAIOnnxRuntimeModel
 
@@ -841,17 +839,17 @@ class ONNXModelPatcher:
             ti_tokenizer = copy.deepcopy(tokenizer)
             ti_manager = TextualInversionManager(ti_tokenizer)
 
-            def _get_trigger(ti, index):
-                trigger = ti.name
+            def _get_trigger(ti_name, index):
+                trigger = ti_name
                 if index > 0:
                     trigger += f"-!pad-{i}"
                 return f"<{trigger}>"
 
             # modify tokenizer
             new_tokens_added = 0
-            for ti in ti_list:
+            for ti_name, ti in ti_list:
                 for i in range(ti.embedding.shape[0]):
-                    new_tokens_added += ti_tokenizer.add_tokens(_get_trigger(ti, i))
+                    new_tokens_added += ti_tokenizer.add_tokens(_get_trigger(ti_name, i))
 
             # modify text_encoder
             orig_embeddings = text_encoder.tensors["text_model.embeddings.token_embedding.weight"]
@@ -861,11 +859,11 @@ class ONNXModelPatcher:
                 axis=0,
             )
 
-            for ti in ti_list:
+            for ti_name, ti in ti_list:
                 ti_tokens = []
                 for i in range(ti.embedding.shape[0]):
                     embedding = ti.embedding[i].detach().numpy()
-                    trigger = _get_trigger(ti, i)
+                    trigger = _get_trigger(ti_name, i)
 
                     token_id = ti_tokenizer.convert_tokens_to_ids(trigger)
                     if token_id == ti_tokenizer.unk_token_id:

invokeai/backend/model_management/model_cache.py

@@ -210,6 +210,31 @@ class ModelCache(object):
 
         return self.ModelLocker(self, key, cache_entry.model, gpu_load, cache_entry.size)
 
+    def clear_one_model(self) -> bool:
+        reserved = self.max_vram_cache_size * GIG
+        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")
+        smallest_key = None
+        smallest_size = float("inf")
+        for model_key, cache_entry in sorted(self._cached_models.items(), key=lambda x: x[1].size):
+            if not cache_entry.locked and cache_entry.loaded:
+                if cache_entry.size > 0 and cache_entry.size < smallest_size:
+                    smallest_key = model_key
+                    smallest_size = cache_entry.size
+
+        if smallest_key is not None:
+            cache_entry = self._cached_models[smallest_key]
+            self.logger.debug(f"!!!!!!!!!!!Offloading {smallest_key} from {self.execution_device} into {self.storage_device}")
+            with VRAMUsage() as mem:
+                cache_entry.model.to(self.storage_device)
+            self.logger.debug(f"GPU VRAM freed: {(mem.vram_used/GIG):.2f} GB")
+            vram_in_use += mem.vram_used  # note vram_used is negative
+            self.logger.debug(f"{(vram_in_use/GIG):.2f}GB VRAM used for models; max allowed={(reserved/GIG):.2f}GB")
+
+        torch.cuda.empty_cache()
+        gc.collect()
+        return smallest_key is not None
+
     class ModelLocker(object):
         def __init__(self, cache, key, model, gpu_load, size_needed):
             """
@@ -236,17 +261,48 @@ class ModelCache(object):
                 self.cache_entry.lock()
 
                 try:
-                    if self.cache.lazy_offloading:
-                        self.cache._offload_unlocked_models(self.size_needed)
+                    self.cache.logger.debug(f"Moving {self.key} into {self.cache.execution_device}")
+                    while True:
+                        try:
+                            with VRAMUsage() as mem:
+                                self.model.to(self.cache.execution_device)  # move into GPU
+                            self.cache.logger.debug(f"GPU VRAM used for load: {(mem.vram_used/GIG):.2f} GB")
 
-                    if self.model.device != self.cache.execution_device:
-                        self.cache.logger.debug(f"Moving {self.key} into {self.cache.execution_device}")
-                        with VRAMUsage() as mem:
-                            self.model.to(self.cache.execution_device)  # move into GPU
-                        self.cache.logger.debug(f"GPU VRAM used for load: {(mem.vram_used/GIG):.2f} GB")
+                            self.cache.logger.debug(f"Locking {self.key} in {self.cache.execution_device}")
+                            self.cache._print_cuda_stats()
 
-                    self.cache.logger.debug(f"Locking {self.key} in {self.cache.execution_device}")
-                    self.cache._print_cuda_stats()
+                            def my_forward(module, cache, *args, **kwargs):
+                                while True:
+                                    try:
+                                        return module._orig_forward(*args, **kwargs)
+                                    except:
+                                        if not cache.clear_one_model():
+                                            raise
+
+                            import functools
+                            from diffusers.models.unet_2d_blocks import DownBlock2D, CrossAttnDownBlock2D, UpBlock2D, CrossAttnUpBlock2D
+                            from transformers.models.clip.modeling_clip import CLIPEncoderLayer
+                            from diffusers.models.unet_2d_blocks import DownEncoderBlock2D, UpDecoderBlock2D
+
+                            for module_name, module in self.model.named_modules():
+                                if type(module) not in [
+                                    DownBlock2D, CrossAttnDownBlock2D, UpBlock2D, CrossAttnUpBlock2D, # unet blocks
+                                    CLIPEncoderLayer, # CLIPTextTransformer clip
+                                    DownEncoderBlock2D, UpDecoderBlock2D, # vae
+                                ]:
+                                    continue
+                                # better here filter to only specific model modules
+                                module._orig_forward = module.forward
+                                module.forward = functools.partial(my_forward, module, self.cache)
+
+                            self.model._orig_forward = self.model.forward
+                            self.model.forward = functools.partial(my_forward, self.model, self.cache)
+
+                            break
+
+                        except:
+                            if not self.cache.clear_one_model():
+                                raise
 
                 except:
                     self.cache_entry.unlock()
@@ -264,10 +320,19 @@ class ModelCache(object):
             if not hasattr(self.model, "to"):
                 return
 
+            if hasattr(self.model, "_orig_forward"):
+                self.model.forward = self.model._orig_forward
+                delattr(self.model, "_orig_forward")
+
+            for module_name, module in self.model.named_modules():
+                if hasattr(module, "_orig_forward"):
+                    module.forward = module._orig_forward
+                    delattr(module, "_orig_forward")
+
             self.cache_entry.unlock()
-            if not self.cache.lazy_offloading:
-                self.cache._offload_unlocked_models()
-                self.cache._print_cuda_stats()
+            #if not self.cache.lazy_offloading:
+            #    self.cache._offload_unlocked_models()
+            #    self.cache._print_cuda_stats()
 
     # TODO: should it be called untrack_model?
     def uncache_model(self, cache_id: str):

invokeai/frontend/web/src/features/ui/components/tabs/ModelManager/subpanels/ModelManagerPanel/ModelList.tsx

@@ -1,4 +1,4 @@
-import { ButtonGroup, Flex, Text } from '@chakra-ui/react';
+import { ButtonGroup, Flex, Spinner, Text } from '@chakra-ui/react';
 import { EntityState } from '@reduxjs/toolkit';
 import IAIButton from 'common/components/IAIButton';
 import IAIInput from 'common/components/IAIInput';
@@ -6,23 +6,23 @@ import { forEach } from 'lodash-es';
 import type { ChangeEvent, PropsWithChildren } from 'react';
 import { useCallback, useState } from 'react';
 import { useTranslation } from 'react-i18next';
+import { ALL_BASE_MODELS } from 'services/api/constants';
 import {
+  LoRAModelConfigEntity,
   MainModelConfigEntity,
   OnnxModelConfigEntity,
+  useGetLoRAModelsQuery,
   useGetMainModelsQuery,
   useGetOnnxModelsQuery,
-  useGetLoRAModelsQuery,
-  LoRAModelConfigEntity,
 } from 'services/api/endpoints/models';
 import ModelListItem from './ModelListItem';
-import { ALL_BASE_MODELS } from 'services/api/constants';
 
 type ModelListProps = {
   selectedModelId: string | undefined;
   setSelectedModelId: (name: string | undefined) => void;
 };
 
-type ModelFormat = 'images' | 'checkpoint' | 'diffusers' | 'olive' | 'onnx';
+type ModelFormat = 'all' | 'checkpoint' | 'diffusers' | 'olive' | 'onnx';
 
 type ModelType = 'main' | 'lora' | 'onnx';
 
@@ -33,47 +33,63 @@ const ModelList = (props: ModelListProps) => {
   const { t } = useTranslation();
   const [nameFilter, setNameFilter] = useState<string>('');
   const [modelFormatFilter, setModelFormatFilter] =
-    useState<CombinedModelFormat>('images');
+    useState<CombinedModelFormat>('all');
 
-  const { filteredDiffusersModels } = useGetMainModelsQuery(ALL_BASE_MODELS, {
-    selectFromResult: ({ data }) => ({
-      filteredDiffusersModels: modelsFilter(
-        data,
-        'main',
-        'diffusers',
-        nameFilter
-      ),
-    }),
-  });
+  const { filteredDiffusersModels, isLoadingDiffusersModels } =
+    useGetMainModelsQuery(ALL_BASE_MODELS, {
+      selectFromResult: ({ data, isLoading }) => ({
+        filteredDiffusersModels: modelsFilter(
+          data,
+          'main',
+          'diffusers',
+          nameFilter
+        ),
+        isLoadingDiffusersModels: isLoading,
+      }),
+    });
 
-  const { filteredCheckpointModels } = useGetMainModelsQuery(ALL_BASE_MODELS, {
-    selectFromResult: ({ data }) => ({
-      filteredCheckpointModels: modelsFilter(
-        data,
-        'main',
-        'checkpoint',
-        nameFilter
-      ),
-    }),
-  });
+  const { filteredCheckpointModels, isLoadingCheckpointModels } =
+    useGetMainModelsQuery(ALL_BASE_MODELS, {
+      selectFromResult: ({ data, isLoading }) => ({
+        filteredCheckpointModels: modelsFilter(
+          data,
+          'main',
+          'checkpoint',
+          nameFilter
+        ),
+        isLoadingCheckpointModels: isLoading,
+      }),
+    });
 
-  const { filteredLoraModels } = useGetLoRAModelsQuery(undefined, {
-    selectFromResult: ({ data }) => ({
-      filteredLoraModels: modelsFilter(data, 'lora', undefined, nameFilter),
-    }),
-  });
+  const { filteredLoraModels, isLoadingLoraModels } = useGetLoRAModelsQuery(
+    undefined,
+    {
+      selectFromResult: ({ data, isLoading }) => ({
+        filteredLoraModels: modelsFilter(data, 'lora', undefined, nameFilter),
+        isLoadingLoraModels: isLoading,
+      }),
+    }
+  );
 
-  const { filteredOnnxModels } = useGetOnnxModelsQuery(ALL_BASE_MODELS, {
-    selectFromResult: ({ data }) => ({
-      filteredOnnxModels: modelsFilter(data, 'onnx', 'onnx', nameFilter),
-    }),
-  });
+  const { filteredOnnxModels, isLoadingOnnxModels } = useGetOnnxModelsQuery(
+    ALL_BASE_MODELS,
+    {
+      selectFromResult: ({ data, isLoading }) => ({
+        filteredOnnxModels: modelsFilter(data, 'onnx', 'onnx', nameFilter),
+        isLoadingOnnxModels: isLoading,
+      }),
+    }
+  );
 
-  const { filteredOliveModels } = useGetOnnxModelsQuery(ALL_BASE_MODELS, {
-    selectFromResult: ({ data }) => ({
-      filteredOliveModels: modelsFilter(data, 'onnx', 'olive', nameFilter),
-    }),
-  });
+  const { filteredOliveModels, isLoadingOliveModels } = useGetOnnxModelsQuery(
+    ALL_BASE_MODELS,
+    {
+      selectFromResult: ({ data, isLoading }) => ({
+        filteredOliveModels: modelsFilter(data, 'onnx', 'olive', nameFilter),
+        isLoadingOliveModels: isLoading,
+      }),
+    }
+  );
 
   const handleSearchFilter = useCallback((e: ChangeEvent<HTMLInputElement>) => {
     setNameFilter(e.target.value);
@@ -84,8 +100,8 @@ const ModelList = (props: ModelListProps) => {
       <Flex flexDirection="column" gap={4} paddingInlineEnd={4}>
         <ButtonGroup isAttached>
           <IAIButton
-            onClick={() => setModelFormatFilter('images')}
-            isChecked={modelFormatFilter === 'images'}
+            onClick={() => setModelFormatFilter('all')}
+            isChecked={modelFormatFilter === 'all'}
             size="sm"
           >
             {t('modelManager.allModels')}
@@ -139,95 +155,76 @@ const ModelList = (props: ModelListProps) => {
           maxHeight={window.innerHeight - 280}
           overflow="scroll"
         >
-          {['images', 'diffusers'].includes(modelFormatFilter) &&
+          {/* Diffusers List */}
+          {isLoadingDiffusersModels && (
+            <FetchingModelsLoader loadingMessage="Loading Diffusers..." />
+          )}
+          {['all', 'diffusers'].includes(modelFormatFilter) &&
+            !isLoadingDiffusersModels &&
             filteredDiffusersModels.length > 0 && (
-              <StyledModelContainer>
-                <Flex sx={{ gap: 2, flexDir: 'column' }}>
-                  <Text variant="subtext" fontSize="sm">
-                    Diffusers
-                  </Text>
-                  {filteredDiffusersModels.map((model) => (
-                    <ModelListItem
-                      key={model.id}
-                      model={model}
-                      isSelected={selectedModelId === model.id}
-                      setSelectedModelId={setSelectedModelId}
-                    />
-                  ))}
-                </Flex>
-              </StyledModelContainer>
+              <ModelListWrapper
+                title="Diffusers"
+                modelList={filteredDiffusersModels}
+                selected={{ selectedModelId, setSelectedModelId }}
+                key="diffusers"
+              />
             )}
-          {['images', 'checkpoint'].includes(modelFormatFilter) &&
+          {/* Checkpoints List */}
+          {isLoadingCheckpointModels && (
+            <FetchingModelsLoader loadingMessage="Loading Checkpoints..." />
+          )}
+          {['all', 'checkpoint'].includes(modelFormatFilter) &&
+            !isLoadingCheckpointModels &&
             filteredCheckpointModels.length > 0 && (
-              <StyledModelContainer>
-                <Flex sx={{ gap: 2, flexDir: 'column' }}>
-                  <Text variant="subtext" fontSize="sm">
-                    Checkpoints
-                  </Text>
-                  {filteredCheckpointModels.map((model) => (
-                    <ModelListItem
-                      key={model.id}
-                      model={model}
-                      isSelected={selectedModelId === model.id}
-                      setSelectedModelId={setSelectedModelId}
-                    />
-                  ))}
-                </Flex>
-              </StyledModelContainer>
+              <ModelListWrapper
+                title="Checkpoints"
+                modelList={filteredCheckpointModels}
+                selected={{ selectedModelId, setSelectedModelId }}
+                key="checkpoints"
+              />
             )}
-          {['images', 'olive'].includes(modelFormatFilter) &&
-            filteredOliveModels.length > 0 && (
-              <StyledModelContainer>
-                <Flex sx={{ gap: 2, flexDir: 'column' }}>
-                  <Text variant="subtext" fontSize="sm">
-                    Olives
-                  </Text>
-                  {filteredOliveModels.map((model) => (
-                    <ModelListItem
-                      key={model.id}
-                      model={model}
-                      isSelected={selectedModelId === model.id}
-                      setSelectedModelId={setSelectedModelId}
-                    />
-                  ))}
-                </Flex>
-              </StyledModelContainer>
-            )}
-          {['images', 'onnx'].includes(modelFormatFilter) &&
-            filteredOnnxModels.length > 0 && (
-              <StyledModelContainer>
-                <Flex sx={{ gap: 2, flexDir: 'column' }}>
-                  <Text variant="subtext" fontSize="sm">
-                    Onnx
-                  </Text>
-                  {filteredOnnxModels.map((model) => (
-                    <ModelListItem
-                      key={model.id}
-                      model={model}
-                      isSelected={selectedModelId === model.id}
-                      setSelectedModelId={setSelectedModelId}
-                    />
-                  ))}
-                </Flex>
-              </StyledModelContainer>
-            )}
-          {['images', 'lora'].includes(modelFormatFilter) &&
+
+          {/* LoRAs List */}
+          {isLoadingLoraModels && (
+            <FetchingModelsLoader loadingMessage="Loading LoRAs..." />
+          )}
+          {['all', 'lora'].includes(modelFormatFilter) &&
+            !isLoadingLoraModels &&
             filteredLoraModels.length > 0 && (
-              <StyledModelContainer>
-                <Flex sx={{ gap: 2, flexDir: 'column' }}>
-                  <Text variant="subtext" fontSize="sm">
-                    LoRAs
-                  </Text>
-                  {filteredLoraModels.map((model) => (
-                    <ModelListItem
-                      key={model.id}
-                      model={model}
-                      isSelected={selectedModelId === model.id}
-                      setSelectedModelId={setSelectedModelId}
-                    />
-                  ))}
-                </Flex>
-              </StyledModelContainer>
+              <ModelListWrapper
+                title="LoRAs"
+                modelList={filteredLoraModels}
+                selected={{ selectedModelId, setSelectedModelId }}
+                key="loras"
+              />
+            )}
+          {/* Olive List */}
+          {isLoadingOliveModels && (
+            <FetchingModelsLoader loadingMessage="Loading Olives..." />
+          )}
+          {['all', 'olive'].includes(modelFormatFilter) &&
+            !isLoadingOliveModels &&
+            filteredOliveModels.length > 0 && (
+              <ModelListWrapper
+                title="Olives"
+                modelList={filteredOliveModels}
+                selected={{ selectedModelId, setSelectedModelId }}
+                key="olive"
+              />
+            )}
+          {/* Onnx List */}
+          {isLoadingOnnxModels && (
+            <FetchingModelsLoader loadingMessage="Loading ONNX..." />
+          )}
+          {['all', 'onnx'].includes(modelFormatFilter) &&
+            !isLoadingOnnxModels &&
+            filteredOnnxModels.length > 0 && (
+              <ModelListWrapper
+                title="ONNX"
+                modelList={filteredOnnxModels}
+                selected={{ selectedModelId, setSelectedModelId }}
+                key="onnx"
+              />
             )}
         </Flex>
       </Flex>
@@ -287,3 +284,52 @@ const StyledModelContainer = (props: PropsWithChildren) => {
     </Flex>
   );
 };
+
+type ModelListWrapperProps = {
+  title: string;
+  modelList:
+    | MainModelConfigEntity[]
+    | LoRAModelConfigEntity[]
+    | OnnxModelConfigEntity[];
+  selected: ModelListProps;
+};
+
+function ModelListWrapper(props: ModelListWrapperProps) {
+  const { title, modelList, selected } = props;
+  return (
+    <StyledModelContainer>
+      <Flex sx={{ gap: 2, flexDir: 'column' }}>
+        <Text variant="subtext" fontSize="sm">
+          {title}
+        </Text>
+        {modelList.map((model) => (
+          <ModelListItem
+            key={model.id}
+            model={model}
+            isSelected={selected.selectedModelId === model.id}
+            setSelectedModelId={selected.setSelectedModelId}
+          />
+        ))}
+      </Flex>
+    </StyledModelContainer>
+  );
+}
+
+function FetchingModelsLoader({ loadingMessage }: { loadingMessage?: string }) {
+  return (
+    <StyledModelContainer>
+      <Flex
+        justifyContent="center"
+        alignItems="center"
+        flexDirection="column"
+        p={4}
+        gap={8}
+      >
+        <Spinner />
+        <Text variant="subtext">
+          {loadingMessage ? loadingMessage : 'Fetching...'}
+        </Text>
+      </Flex>
+    </StyledModelContainer>
+  );
+}

invokeai/frontend/web/src/services/api/endpoints/models.ts

@@ -181,7 +181,7 @@ export const modelsApi = api.injectEndpoints({
       },
       providesTags: (result, error, arg) => {
         const tags: ApiFullTagDescription[] = [
-          { id: 'OnnxModel', type: LIST_TAG },
+          { type: 'OnnxModel', id: LIST_TAG },
         ];
 
         if (result) {
@@ -266,6 +266,7 @@ export const modelsApi = api.injectEndpoints({
       invalidatesTags: [
         { type: 'MainModel', id: LIST_TAG },
         { type: 'SDXLRefinerModel', id: LIST_TAG },
+        { type: 'OnnxModel', id: LIST_TAG },
       ],
     }),
     importMainModels: build.mutation<
@@ -282,6 +283,7 @@ export const modelsApi = api.injectEndpoints({
       invalidatesTags: [
         { type: 'MainModel', id: LIST_TAG },
         { type: 'SDXLRefinerModel', id: LIST_TAG },
+        { type: 'OnnxModel', id: LIST_TAG },
       ],
     }),
     addMainModels: build.mutation<AddMainModelResponse, AddMainModelArg>({
@@ -295,6 +297,7 @@ export const modelsApi = api.injectEndpoints({
       invalidatesTags: [
         { type: 'MainModel', id: LIST_TAG },
         { type: 'SDXLRefinerModel', id: LIST_TAG },
+        { type: 'OnnxModel', id: LIST_TAG },
       ],
     }),
     deleteMainModels: build.mutation<
@@ -310,6 +313,7 @@ export const modelsApi = api.injectEndpoints({
       invalidatesTags: [
         { type: 'MainModel', id: LIST_TAG },
         { type: 'SDXLRefinerModel', id: LIST_TAG },
+        { type: 'OnnxModel', id: LIST_TAG },
       ],
     }),
     convertMainModels: build.mutation<
@@ -326,6 +330,7 @@ export const modelsApi = api.injectEndpoints({
       invalidatesTags: [
         { type: 'MainModel', id: LIST_TAG },
         { type: 'SDXLRefinerModel', id: LIST_TAG },
+        { type: 'OnnxModel', id: LIST_TAG },
       ],
     }),
     mergeMainModels: build.mutation<MergeMainModelResponse, MergeMainModelArg>({
@@ -339,6 +344,7 @@ export const modelsApi = api.injectEndpoints({
       invalidatesTags: [
         { type: 'MainModel', id: LIST_TAG },
         { type: 'SDXLRefinerModel', id: LIST_TAG },
+        { type: 'OnnxModel', id: LIST_TAG },
       ],
     }),
     syncModels: build.mutation<SyncModelsResponse, void>({
@@ -351,6 +357,7 @@ export const modelsApi = api.injectEndpoints({
       invalidatesTags: [
         { type: 'MainModel', id: LIST_TAG },
         { type: 'SDXLRefinerModel', id: LIST_TAG },
+        { type: 'OnnxModel', id: LIST_TAG },
       ],
     }),
     getLoRAModels: build.query<EntityState<LoRAModelConfigEntity>, void>({