fix(diffusers): forward original config for single files

Assisted-by: Codex:gpt-5

Author: localai-org-maint-bot

backend/backend.proto

@@ -498,6 +498,7 @@ message ModelOptions {
   // applied verbatim to the backend's engine constructor (e.g. vLLM AsyncEngineArgs).
   // Unknown keys produce an error at LoadModel time.
   string EngineArgs = 73;
+  string OriginalConfigFile = 76;
 
   // Proxy carries the cloud-proxy backend's per-model configuration.
   // Empty for non-proxy backends.

backend/python/diffusers/backend.py

@@ -35,6 +35,7 @@
     get_available_pipelines,
     load_diffusers_pipeline,
 )
+from load_options import single_file_load_kwargs
 
 # Import specific items still needed for special cases and safety checker
 from diffusers import DiffusionPipeline, ControlNetModel
@@ -464,6 +465,9 @@ def _load_pipeline(self, request, model_ref, from_single_file, local_only, torch
 
         # Build kwargs for dynamic loading
         load_kwargs = {"torch_dtype": torchType}
+        load_kwargs.update(
+            single_file_load_kwargs(request.OriginalConfigFile, from_single_file)
+        )
 
         # Add variant if not loading from single file
         if not from_single_file and variant:

backend/python/diffusers/load_options.py

@@ -0,0 +1,7 @@
+# SPDX-License-Identifier: MIT
+
+
+def single_file_load_kwargs(original_config_file: str, from_single_file: bool) -> dict:
+    if from_single_file and original_config_file:
+        return {"original_config_file": original_config_file}
+    return {}

backend/python/diffusers/test_load_options.py

@@ -0,0 +1,26 @@
+# SPDX-License-Identifier: MIT
+
+import unittest
+
+from backend.python.diffusers.load_options import single_file_load_kwargs
+
+
+class SingleFileLoadKwargsTest(unittest.TestCase):
+    def test_includes_original_config_for_single_file(self):
+        self.assertEqual(
+            single_file_load_kwargs("configs/v1-inference.yaml", True),
+            {"original_config_file": "configs/v1-inference.yaml"},
+        )
+
+    def test_omits_original_config_for_pretrained_model(self):
+        self.assertEqual(
+            single_file_load_kwargs("configs/v1-inference.yaml", False),
+            {},
+        )
+
+    def test_omits_empty_original_config(self):
+        self.assertEqual(single_file_load_kwargs("", True), {})
+
+
+if __name__ == "__main__":
+    unittest.main()

core/backend/options.go

@@ -425,6 +425,7 @@ func grpcModelOpts(c config.ModelConfig, modelPath string) *pb.ModelOptions {
 		EnableScore:          c.HasUsecases(config.FLAG_SCORE),
 		CLIPSkip:             int32(c.Diffusers.ClipSkip),
 		ControlNet:           c.Diffusers.ControlNet,
+		OriginalConfigFile:   c.Diffusers.OriginalConfigFile,
 		ContextSize:          int32(ctxSize),
 		Seed:                 getSeed(c),
 		NBatch:               int32(b),

core/backend/options_internal_test.go

@@ -44,6 +44,21 @@ var _ = Describe("grpcModelOpts EngineArgs", func() {
 	})
 })
 
+var _ = Describe("grpcModelOpts Diffusers options", func() {
+	It("forwards original_config_file without rewriting it", func() {
+		threads := 1
+		cfg := config.ModelConfig{
+			Threads: &threads,
+			Diffusers: config.Diffusers{
+				OriginalConfigFile: "configs/v1-inference.yaml",
+			},
+		}
+
+		opts := grpcModelOpts(cfg, "/tmp/models")
+		Expect(opts.OriginalConfigFile).To(Equal("configs/v1-inference.yaml"))
+	})
+})
+
 // Guards the DisableReasoning -> enable_thinking metadata conversion that the
 // per-request reasoning_effort feature (issue #10072) relies on: the request
 // merge sets ReasoningConfig.DisableReasoning, and gRPCPredictOpts is where it

core/config/model_config.go

@@ -1104,15 +1104,16 @@ type GRPC struct {
 
 // @Description Diffusers configuration
 type Diffusers struct {
-	CUDA             bool   `yaml:"cuda,omitempty" json:"cuda,omitempty"`
-	PipelineType     string `yaml:"pipeline_type,omitempty" json:"pipeline_type,omitempty"`
-	SchedulerType    string `yaml:"scheduler_type,omitempty" json:"scheduler_type,omitempty"`
-	EnableParameters string `yaml:"enable_parameters,omitempty" json:"enable_parameters,omitempty"` // A list of comma separated parameters to specify
-	IMG2IMG          bool   `yaml:"img2img,omitempty" json:"img2img,omitempty"`                     // Image to Image Diffuser
-	ClipSkip         int    `yaml:"clip_skip,omitempty" json:"clip_skip,omitempty"`                 // Skip every N frames
-	ClipModel        string `yaml:"clip_model,omitempty" json:"clip_model,omitempty"`               // Clip model to use
-	ClipSubFolder    string `yaml:"clip_subfolder,omitempty" json:"clip_subfolder,omitempty"`       // Subfolder to use for clip model
-	ControlNet       string `yaml:"control_net,omitempty" json:"control_net,omitempty"`
+	CUDA               bool   `yaml:"cuda,omitempty" json:"cuda,omitempty"`
+	PipelineType       string `yaml:"pipeline_type,omitempty" json:"pipeline_type,omitempty"`
+	SchedulerType      string `yaml:"scheduler_type,omitempty" json:"scheduler_type,omitempty"`
+	OriginalConfigFile string `yaml:"original_config_file,omitempty" json:"original_config_file,omitempty"`
+	EnableParameters   string `yaml:"enable_parameters,omitempty" json:"enable_parameters,omitempty"` // A list of comma separated parameters to specify
+	IMG2IMG            bool   `yaml:"img2img,omitempty" json:"img2img,omitempty"`                     // Image to Image Diffuser
+	ClipSkip           int    `yaml:"clip_skip,omitempty" json:"clip_skip,omitempty"`                 // Skip every N frames
+	ClipModel          string `yaml:"clip_model,omitempty" json:"clip_model,omitempty"`               // Clip model to use
+	ClipSubFolder      string `yaml:"clip_subfolder,omitempty" json:"clip_subfolder,omitempty"`       // Subfolder to use for clip model
+	ControlNet         string `yaml:"control_net,omitempty" json:"control_net,omitempty"`
 }
 
 // @Description LLMConfig is a struct that holds the configuration that are generic for most of the LLM backends.

docs/content/advanced/model-configuration.md

@@ -715,6 +715,7 @@ For image generation models using the `diffusers` backend:
 | `diffusers.cuda` | bool | Enable CUDA for diffusers |
 | `diffusers.pipeline_type` | string | Pipeline type (e.g., `stable-diffusion`, `stable-diffusion-xl`) |
 | `diffusers.scheduler_type` | string | Scheduler type (e.g., `euler`, `ddpm`) |
+| `diffusers.original_config_file` | string | Local path or URL to the original configuration for loading a single-file checkpoint |
 | `diffusers.enable_parameters` | string | Comma-separated parameters to enable |
 | `diffusers.cfg_scale` | float32 | Classifier-free guidance scale |
 | `diffusers.img2img` | bool | Enable image-to-image transformation |

docs/content/features/image-generation.md

@@ -224,6 +224,18 @@ diffusers:
 cfg_scale: 8
 ```
 
+For an offline single-file checkpoint that needs its original Diffusers configuration, keep both files in the mounted models directory:
+
+```yaml
+name: offline-stable-diffusion
+parameters:
+  model: model.safetensors
+backend: diffusers
+diffusers:
+  pipeline_type: StableDiffusionPipeline
+  original_config_file: /models/v1-inference.yaml
+```
+
 #### Configuration parameters
 
 The following parameters are available in the configuration file: