ENH: update model "DeepSeek-OCR" JSON (#4444)

Co-authored-by: github-actions[bot] <41898282+github-actions[bot]@users.noreply.github.com>

Author: OliverBryant

doc/source/gen_docs.py

@@ -241,6 +241,13 @@ def _can_use_transformers_legacy(model, model_spec):
     abilities = set(model.get("model_ability", []))
     return "chat" in abilities or "generate" in abilities
 
+def _extract_primary_model_src(model):
+    if model.get("model_specs"):
+        for spec in model["model_specs"]:
+            if isinstance(spec, dict) and "model_src" in spec:
+                return spec["model_src"]
+    return model.get("model_src")
+
 def main():
     template_dir = '../templates' 
     env = Environment(loader=FileSystemLoader(template_dir))
@@ -456,9 +463,10 @@ def main():
 
         for model in sorted_models:
             # Process model_src for template compatibility
-            if 'model_src' in model:
-                if 'huggingface' in model['model_src']:
-                    hf_src = model['model_src']['huggingface']
+            model_src = _extract_primary_model_src(model)
+            if model_src:
+                if 'huggingface' in model_src:
+                    hf_src = model_src['huggingface']
                     model['model_id'] = hf_src['model_id']
                     # Handle GGUF related fields
                     if 'gguf_model_id' in hf_src:
@@ -470,8 +478,8 @@ def main():
                         model['lightning_model_id'] = hf_src['lightning_model_id']
                     if 'lightning_versions' in hf_src:
                         model['lightning_versions'] = ", ".join(hf_src['lightning_versions'])
-                elif 'modelscope' in model['model_src']:
-                    model['model_id'] = model['model_src']['modelscope']['model_id']
+                elif 'modelscope' in model_src:
+                    model['model_id'] = model_src['modelscope']['model_id']
 
             available_controlnet = [cn["model_name"] for cn in model.get("controlnet", [])]
             if not available_controlnet:
@@ -502,11 +510,12 @@ def main():
 
         for model in sorted_models:
             # Process model_src for template compatibility
-            if 'model_src' in model:
-                if 'huggingface' in model['model_src']:
-                    model['model_id'] = model['model_src']['huggingface']['model_id']
-                elif 'modelscope' in model['model_src']:
-                    model['model_id'] = model['model_src']['modelscope']['model_id']
+            model_src = _extract_primary_model_src(model)
+            if model_src:
+                if 'huggingface' in model_src:
+                    model['model_id'] = model_src['huggingface']['model_id']
+                elif 'modelscope' in model_src:
+                    model['model_id'] = model_src['modelscope']['model_id']
 
             rendered = env.get_template('audio.rst.jinja').render(model)
             output_file_path = os.path.join(output_dir, f"{model['model_name'].lower()}.rst")
@@ -527,11 +536,12 @@ def main():
 
         for model in sorted_models:
             # Process model_src for template compatibility
-            if 'model_src' in model:
-                if 'huggingface' in model['model_src']:
-                    model['model_id'] = model['model_src']['huggingface']['model_id']
-                elif 'modelscope' in model['model_src']:
-                    model['model_id'] = model['model_src']['modelscope']['model_id']
+            model_src = _extract_primary_model_src(model)
+            if model_src:
+                if 'huggingface' in model_src:
+                    model['model_id'] = model_src['huggingface']['model_id']
+                elif 'modelscope' in model_src:
+                    model['model_id'] = model_src['modelscope']['model_id']
 
             model["model_ability"] = ', '.join(model.get("model_ability"))
             rendered = env.get_template('video.rst.jinja').render(model)

xinference/model/image/model_spec.json

@@ -998,18 +998,48 @@
         "#system_numpy#"
       ]
     },
-    "model_src": {
-      "huggingface": {
-        "model_id": "deepseek-ai/DeepSeek-OCR",
-        "model_revision": "main"
+    "model_specs": [
+      {
+        "model_format": "pytorch",
+        "model_src": {
+          "huggingface": {
+            "model_id": "deepseek-ai/DeepSeek-OCR",
+            "model_revision": "main"
+          },
+          "modelscope": {
+            "model_id": "deepseek-ai/DeepSeek-OCR",
+            "model_revision": "master"
+          }
+        }
       },
-      "modelscope": {
-        "model_id": "deepseek-ai/DeepSeek-OCR",
-        "model_revision": "master"
+      {
+        "model_format": "mlx",
+        "model_src": {
+          "huggingface": {
+            "model_id": "mlx-community/DeepSeek-OCR-{quantization}",
+            "model_revision": "main",
+            "quantizations": [
+              "8bit",
+              "6bit",
+              "5bit",
+              "4bit"
+            ]
+          },
+          "modelscope": {
+            "model_id": "mlx-community/DeepSeek-OCR-{quantization}",
+            "model_revision": "master",
+            "quantizations": [
+              "8bit",
+              "6bit",
+              "5bit",
+              "4bit"
+            ]
+          }
+        }
       }
-    },
-    "updated_at": 1766387624,
-    "featured": false
+    ],
+    "featured": false,
+    "updated_at": 1767691154
   },
   {
     "version": 2,