feat(models): add support to qwen-image

Signed-off-by: Ettore Di Giacinto <[email protected]>

Author: mudler

backend/python/diffusers/backend.py

@@ -65,6 +65,19 @@ def sc(self, clip_input, images): return images, [False for i in images]
     UniPCMultistepScheduler,
 )
 
+def is_float(s):
+    try:
+        float(s)
+        return True
+    except ValueError:
+        return False
+
+def is_int(s):
+    try:
+        int(s)
+        return True
+    except ValueError:
+        return False
 
 # The scheduler list mapping was taken from here: https://github.com/neggles/animatediff-cli/blob/6f336f5f4b5e38e85d7f06f1744ef42d0a45f2a7/src/animatediff/schedulers.py#L39
 # Credits to https://github.com/neggles
@@ -169,8 +182,24 @@ def LoadModel(self, request, context):
                 if ":" not in opt:
                     continue
                 key, value = opt.split(":")
+                # if value is a number, convert it to the appropriate type
+                if is_float(value):
+                    value = float(value)
+                elif is_int(value):
+                    value = int(value)
                 self.options[key] = value
 
+            # From options, extract if present "torch_dtype" and set it to the appropriate type
+            if "torch_dtype" in self.options:
+                if self.options["torch_dtype"] == "fp16":
+                    torchType = torch.float16
+                elif self.options["torch_dtype"] == "bf16":
+                    torchType = torch.bfloat16
+                elif self.options["torch_dtype"] == "fp32":
+                    torchType = torch.float32
+                # remove it from options
+                del self.options["torch_dtype"]
+
             print(f"Options: {self.options}", file=sys.stderr)
 
             local = False

gallery/index.yaml

@@ -1,4 +1,17 @@
 ---
+- &qwenimage
+  name: "qwen-image"
+  url: "github:mudler/LocalAI/gallery/qwen-image.yaml@master"
+  urls:
+    - https://huggingface.co/Qwen/Qwen-Image
+  icon: https://qianwen-res.oss-cn-beijing.aliyuncs.com/Qwen-Image/qwen_image_logo.png
+  license: apache-2.0
+  tags:
+    - qwen-image
+    - gpu
+    - text-to-image
+  description: |
+    We are thrilled to release Qwen-Image, an image generation foundation model in the Qwen series that achieves significant advances in complex text rendering and precise image editing. Experiments show strong general capabilities in both image generation and editing, with exceptional performance in text rendering, especially for Chinese.
 - &gptoss
   name: "gpt-oss-20b"
   url: "github:mudler/LocalAI/gallery/harmony.yaml@master"

gallery/qwen-image.yaml

@@ -0,0 +1,19 @@
+---
+name: "flux"
+
+config_file: |
+    backend: diffusers
+    cfg_scale: 0
+    diffusers:
+      cuda: true
+      enable_parameters: num_inference_steps
+      pipeline_type: DiffusionPipeline
+    f16: true
+    low_vram: true
+    name: qwen-image
+    parameters:
+      model: Qwen/Qwen-Image
+    step: 50
+    options:
+      - true_cfg_scale:4.0
+      - torch_dtype:bf16