minor fixes from audit

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

.github/skills/check-models/SKILL.md

@@ -34,9 +34,9 @@ This skill combines four audit surfaces. Run them in this order unless user scop
 - `modules/sd_detect.py`
 - `modules/sd_models.py`
 - `modules/modeldata.py`
-- `data/reference.json`
+- `data/reference-base.json`
 - `data/reference-cloud.json`
-- `data/reference-quant.json`
+- `data/reference-quantized.json`
 - `data/reference-distilled.json`
 - `data/reference-nunchaku.json`
 - `data/reference-community.json`
@@ -83,9 +83,9 @@ Verify references for model families intended to appear in model references.
 Checks:
 
 - Correct category file placement by type:
-  - base -> `data/reference.json`
+  - base -> `data/reference-base.json`
   - cloud -> `data/reference-cloud.json`
-  - quant -> `data/reference-quant.json`
+  - quant -> `data/reference-quantized.json`
   - distilled -> `data/reference-distilled.json`
   - nunchaku -> `data/reference-nunchaku.json`
   - community -> `data/reference-community.json`

modules/img2img.py

@@ -230,6 +230,8 @@ def img2img(id_task: str, state: str, mode: int,
     elif mode == 4: # inpaint upload mask
         if init_img_inpaint is None:
             return [], '', '', 'Error: inpaint image not provided'
+        if init_mask_inpaint is None:
+            return [], '', '', 'Error: inpaint mask not provided'
         image = init_img_inpaint
         mask = init_mask_inpaint
     elif mode == 5: # process batch

modules/ui_caption.py

@@ -445,7 +445,7 @@ def create_ui():
     # Load/Unload model buttons
     vlm_load_btn.click(fn=vqa.load_model, inputs=[vlm_model], outputs=[])
     vlm_unload_btn.click(fn=vqa.unload_model, inputs=[], outputs=[])
-    analyze_load_btn.click(fn=vqa.load_model, inputs=[vlm_model], outputs=[])
+    analyze_load_btn.click(fn=vqa.load_model, inputs=[analyze_model], outputs=[])
     analyze_unload_btn.click(fn=vqa.unload_model, inputs=[], outputs=[])
 
     def tagger_load_wrapper(model_name):

pipelines/f_lite/pipeline.py

@@ -36,7 +36,7 @@ class FLitePipelineOutput(BaseOutput):
             num_channels)`. PIL images or numpy array present the denoised images of the diffusion pipeline.
     """
 
-    images: Union[List[Image.Image], np.ndarray]
+    images: Union[List[Image.Image], np.ndarray, torch.Tensor]
 
 
 class FLitePipeline(DiffusionPipeline):
@@ -147,6 +147,7 @@ class FLitePipeline(DiffusionPipeline):
 
     def to(self, torch_device=None, torch_dtype=None, silence_dtype_warnings=False):
         """Move pipeline components to specified device and dtype."""
+        _ = silence_dtype_warnings
         if hasattr(self, "vae"):
             self.vae.to(device=torch_device, dtype=torch_dtype)
         if hasattr(self, "text_encoder"):
@@ -169,6 +170,8 @@ class FLitePipeline(DiffusionPipeline):
         dtype: Optional[torch.dtype] = None,
         alpha: Optional[float] = None,
         apg_config: Optional[APGConfig] = None,
+        output_type: str = "pil",
+        return_dict: bool = True,
         **kwargs,
     ):
         """Generate images from text prompt."""
@@ -293,10 +296,22 @@ class FLitePipeline(DiffusionPipeline):
                 raise
 
         # 8. Post-process images
-        from modules.image import convert
         images = (decoded_images / 2 + 0.5).clamp(0, 1)
-        pil_images = [convert.to_pil(img) for img in images]
+        if output_type == "latent":
+            output = latents
+        elif output_type == "pt":
+            output = images
+        elif output_type == "np":
+            output = images.permute(0, 2, 3, 1).to(device="cpu", dtype=torch.float32).numpy()
+        elif output_type == "pil":
+            from modules.image import convert
+            output = [convert.to_pil(img) for img in images]
+        else:
+            raise ValueError(f"Unsupported output_type: {output_type}")
+
+        if not return_dict:
+            return (output,)
 
         return FLitePipelineOutput(
-            images=pil_images,
+            images=output,
         )

pipelines/meissonic/pipeline.py

@@ -18,11 +18,10 @@ from transformers import CLIPTextModelWithProjection, CLIPTokenizer
 
 from diffusers.image_processor import VaeImageProcessor
 from diffusers.models import VQModel
-
-from .scheduler import Scheduler
 from diffusers.utils import replace_example_docstring
 from diffusers.pipelines.pipeline_utils import DiffusionPipeline, ImagePipelineOutput
 
+from .scheduler import Scheduler
 from .transformer import Transformer2DModel
 
 
@@ -34,7 +33,7 @@ EXAMPLE_DOC_STRING = """
 """
 
 
-def _prepare_latent_image_ids(batch_size, height, width, device, dtype):
+def _prepare_latent_image_ids(batch_size, height, width, device, dtype): # pylint: disable=unused-argument
     latent_image_ids = torch.zeros(height // 2, width // 2, 3)
     latent_image_ids[..., 1] = latent_image_ids[..., 1] + torch.arange(height // 2)[:, None]
     latent_image_ids[..., 2] = latent_image_ids[..., 2] + torch.arange(width // 2)[None, :]

pipelines/model_instaflow.py

@@ -18,5 +18,7 @@ def load_instaflow(checkpoint_info, diffusers_load_config=None):
 
     pipeline = diffusers.utils.get_class_from_dynamic_module('instaflow_one_step', module_file='pipeline.py')
     generic.set_pipeline('InstaFlow', pipeline)
-    sd_model = pipeline.from_pretrained(checkpoint_info.path, cache_dir=shared.opts.diffusers_dir, **diffusers_load_config)
+    load_config = {**diffusers_load_config, **load_args}
+    sd_model = pipeline.from_pretrained(checkpoint_info.path, cache_dir=shared.opts.diffusers_dir, **load_config)
+    devices.torch_gc(force=True, reason='load')
     return sd_model

pipelines/model_segmoe.py

@@ -17,6 +17,8 @@ def load_segmoe(checkpoint_info, diffusers_load_config=None):
     if repo_id is None or repo_id.lower() == 'none':
         return None
 
-    sd_model = SegMoEPipeline(checkpoint_info.path, cache_dir=shared.opts.diffusers_dir, **diffusers_load_config)
+    load_config = {**diffusers_load_config, **load_args}
+    sd_model = SegMoEPipeline(checkpoint_info.path, cache_dir=shared.opts.diffusers_dir, **load_config)
     sd_model = sd_model.pipe # segmoe pipe does its stuff in __init__ and __call__ is the original pipeline
+    devices.torch_gc(force=True, reason='load')
     return sd_model