xinference 1.11.0__py3-none-any.whl → 1.12.0__py3-none-any.whl

xinference/__init__.py

@@ -12,6 +12,14 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 
+import os
+
+# Configure MPS memory management to avoid "invalid low watermark ratio" error in PyTorch 3.13+
+if os.environ.get("PYTORCH_MPS_HIGH_WATERMARK_RATIO") is None:
+    os.environ["PYTORCH_MPS_HIGH_WATERMARK_RATIO"] = "1.0"
+if os.environ.get("PYTORCH_MPS_LOW_WATERMARK_RATIO") is None:
+    os.environ["PYTORCH_MPS_LOW_WATERMARK_RATIO"] = "0.2"
+
 from . import _version
 
 __version__ = _version.get_versions()["version"]

xinference/_version.py

@@ -8,11 +8,11 @@ import json
 
 version_json = '''
 {
- "date": "2025-10-19T20:53:12+0800",
+ "date": "2025-11-02T20:34:18+0800",
  "dirty": false,
  "error": null,
- "full-revisionid": "baaa40b463e4948762b078f5995d67775df53704",
- "version": "1.11.0"
+ "full-revisionid": "117ba29e07a77a7534496f18b9ced1d567c0673f",
+ "version": "1.12.0"
 }
 '''  # END VERSION_JSON
 

xinference/api/oauth2/utils.py

@@ -14,10 +14,8 @@
 from datetime import datetime, timedelta
 from typing import Union
 
+import bcrypt
 from jose import jwt
-from passlib.context import CryptContext
-
-pwd_context = CryptContext(schemes=["bcrypt"], deprecated="auto")
 
 
 def create_access_token(
@@ -37,8 +35,31 @@ def create_access_token(
 
 
 def verify_password(plain_password, hashed_password):
-    return pwd_context.verify(plain_password, hashed_password)
+    if isinstance(plain_password, str):
+        plain_password = plain_password.encode("utf-8")
+    if isinstance(hashed_password, str):
+        hashed_password = hashed_password.encode("utf-8")
+
+    if len(plain_password) > 72:
+        import hashlib
+
+        password_hash = hashlib.sha256(plain_password).digest()
+        plain_password = password_hash[:72]
+
+    return bcrypt.checkpw(plain_password, hashed_password)
 
 
 def get_password_hash(password):
-    return pwd_context.hash(password)
+    if isinstance(password, str):
+        password = password.encode("utf-8")
+
+    if len(password) > 72:
+        import hashlib
+
+        password_hash = hashlib.sha256(password).digest()
+        password = password_hash[:72]
+
+    salt = bcrypt.gensalt()
+    hashed = bcrypt.hashpw(password, salt)
+
+    return hashed.decode("utf-8")

xinference/core/model.py

@@ -206,10 +206,6 @@ class ModelActor(xo.StatelessActor, CancelMixin):
     ):
         super().__init__()
 
-        from ..model.llm.llama_cpp.core import XllamaCppModel
-        from ..model.llm.lmdeploy.core import LMDeployModel
-        from ..model.llm.sglang.core import SGLANGModel
-        from ..model.llm.transformers.core import PytorchModel
         from ..model.llm.vllm.core import VLLMModel
 
         self._supervisor_address = supervisor_address
@@ -223,12 +219,7 @@ class ModelActor(xo.StatelessActor, CancelMixin):
         self._pending_requests: asyncio.Queue = asyncio.Queue()
         self._handle_pending_requests_task = None
         self._lock = (
-            None
-            if isinstance(
-                self._model,
-                (PytorchModel, VLLMModel, SGLANGModel, LMDeployModel, XllamaCppModel),
-            )
-            else asyncio.locks.Lock()
+            None if getattr(self._model, "allow_batch", False) else asyncio.locks.Lock()
         )
         self._worker_ref = None
         self._progress_tracker_ref = None

xinference/device_utils.py

@@ -108,7 +108,17 @@ def empty_cache():
     if torch.cuda.is_available():
         torch.cuda.empty_cache()
     if torch.backends.mps.is_available():
-        torch.mps.empty_cache()
+        try:
+            torch.mps.empty_cache()
+        except RuntimeError as e:
+            # Handle known MPS memory management issues in PyTorch 3.13+
+            if "invalid low watermark ratio" in str(e):
+                # This is a known issue with PyTorch 3.13+ on macOS.
+                # We can safely ignore this error as it doesn't affect functionality.
+                pass
+            else:
+                # Re-raise other RuntimeErrors
+                raise
     if is_xpu_available():
         torch.xpu.empty_cache()
     if is_npu_available():

xinference/model/embedding/model_spec.json

@@ -1190,5 +1190,75 @@
       ],
       "no_build_isolation": true
     }
+  },
+  {
+    "version": 2,
+    "model_name": "gme-Qwen2-VL-2B-Instruct",
+    "dimensions": 1536,
+    "max_tokens": 32768,
+    "language": [
+      "en",
+      "zh"
+    ],
+    "virtualenv": {
+      "packages": [
+        "sentence_transformers",
+        "transformers==4.51.3"
+      ]
+    },
+    "model_specs": [
+      {
+        "model_format": "pytorch",
+        "model_src": {
+          "huggingface": {
+            "model_id": "Alibaba-NLP/gme-Qwen2-VL-2B-Instruct",
+            "quantizations": [
+              "none"
+            ]
+          },
+          "modelscope": {
+            "model_id": "iic/gme-Qwen2-VL-2B-Instruct",
+            "quantizations": [
+              "none"
+            ]
+          }
+        }
+      }
+    ]
+  },
+  {
+    "version": 2,
+    "model_name": "gme-Qwen2-VL-7B-Instruct",
+    "dimensions": 3584,
+    "max_tokens": 32768,
+    "language": [
+      "en",
+      "zh"
+    ],
+    "virtualenv": {
+      "packages": [
+        "sentence_transformers",
+        "transformers==4.51.3"
+      ]
+    },
+    "model_specs": [
+      {
+        "model_format": "pytorch",
+        "model_src": {
+          "huggingface": {
+            "model_id": "Alibaba-NLP/gme-Qwen2-VL-7B-Instruct",
+            "quantizations": [
+              "none"
+            ]
+          },
+          "modelscope": {
+            "model_id": "iic/gme-Qwen2-VL-7B-Instruct",
+            "quantizations": [
+              "none"
+            ]
+          }
+        }
+      }
+    ]
   }
 ]

xinference/model/image/core.py

@@ -21,6 +21,7 @@ from typing import Dict, List, Literal, Optional, Union
 from ...types import PeftModelConfig
 from ..core import CacheableModelSpec, VirtualEnvSettings
 from ..utils import ModelInstanceInfoMixin
+from .ocr.deepseek_ocr import DeepSeekOCRModel
 from .ocr.got_ocr2 import GotOCR2Model
 from .stable_diffusion.core import DiffusionModel
 from .stable_diffusion.mlx import MLXDiffusionModel
@@ -159,19 +160,29 @@ def create_ocr_model_instance(
     model_spec: ImageModelFamilyV2,
     model_path: Optional[str] = None,
     **kwargs,
-) -> GotOCR2Model:
+) -> Union[DeepSeekOCRModel, GotOCR2Model]:
     from .cache_manager import ImageCacheManager
 
     if not model_path:
         cache_manager = ImageCacheManager(model_spec)
         model_path = cache_manager.cache()
-    model = GotOCR2Model(
-        model_uid,
-        model_path,
-        model_spec=model_spec,
-        **kwargs,
-    )
-    return model
+
+    # Choose OCR model based on model_name
+    if model_spec.model_name == "DeepSeek-OCR":
+        return DeepSeekOCRModel(
+            model_uid,
+            model_path,
+            model_spec=model_spec,
+            **kwargs,
+        )
+    else:
+        # Default to GOT-OCR2 for other OCR models
+        return GotOCR2Model(
+            model_uid,
+            model_path,
+            model_spec=model_spec,
+            **kwargs,
+        )
 
 
 def create_image_model_instance(
@@ -187,14 +198,13 @@ def create_image_model_instance(
     lightning_version: Optional[str] = None,
     lightning_model_path: Optional[str] = None,
     **kwargs,
-) -> Union[DiffusionModel, MLXDiffusionModel, GotOCR2Model]:
+) -> Union[DiffusionModel, MLXDiffusionModel, GotOCR2Model, DeepSeekOCRModel]:
     from .cache_manager import ImageCacheManager
 
     model_spec = match_diffusion(model_name, download_hub)
     if model_spec.model_ability and "ocr" in model_spec.model_ability:
         return create_ocr_model_instance(
             model_uid=model_uid,
-            model_name=model_name,
             model_spec=model_spec,
             model_path=model_path,
             **kwargs,

xinference/model/image/model_spec.json

@@ -734,7 +734,7 @@
     ],
     "virtualenv": {
       "packages": [
-        "transformers==4.37.2",
+        "transformers==4.47.1",
         "httpx==0.24.0",
         "deepspeed==0.12.3",
         "peft==0.4.0",
@@ -870,7 +870,15 @@
           "Q6_K",
           "Q8_0"
         ],
-        "gguf_model_file_name_template": "Qwen-Image-Edit-2509-{quantization}.gguf"
+        "gguf_model_file_name_template": "Qwen-Image-Edit-2509-{quantization}.gguf",
+        "lightning_model_id": "lightx2v/Qwen-Image-Lightning",
+        "lightning_versions": [
+          "4steps-V1.0-bf16",
+          "4steps-V1.0-fp32",
+          "8steps-V1.0-bf16",
+          "8steps-V1.0-fp32"
+        ],
+        "lightning_model_file_name_template": "Qwen-Image-Edit-2509/Qwen-Image-Edit-2509-Lightning-{lightning_version}.safetensors"
       },
       "modelscope": {
         "model_id": "Qwen/Qwen-Image-Edit-2509",
@@ -891,7 +899,15 @@
           "Q6_K",
           "Q8_0"
         ],
-        "gguf_model_file_name_template": "Qwen-Image-Edit-2509-{quantization}.gguf"
+        "gguf_model_file_name_template": "Qwen-Image-Edit-2509-{quantization}.gguf",
+        "lightning_model_id": "lightx2v/Qwen-Image-Lightning",
+        "lightning_versions": [
+          "4steps-V1.0-bf16",
+          "4steps-V1.0-fp32",
+          "8steps-V1.0-bf16",
+          "8steps-V1.0-fp32"
+        ],
+        "lightning_model_file_name_template": "Qwen-Image-Edit-2509/Qwen-Image-Edit-2509-Lightning-{lightning_version}.safetensors"
       }
     },
     "default_model_config": {
@@ -911,5 +927,41 @@
       ],
       "no_build_isolation": true
     }
+  },
+  {
+    "version": 2,
+    "model_name": "DeepSeek-OCR",
+    "model_family": "ocr",
+    "model_ability": [
+      "ocr"
+    ],
+    "virtualenv": {
+      "packages": [
+        "torch==2.6.0",
+        "torchvision==0.21.0",
+        "torchaudio==2.6.0",
+        "transformers==4.46.3",
+        "tokenizers==0.20.3",
+        "PyMuPDF",
+        "img2pdf",
+        "einops",
+        "easydict",
+        "addict",
+        "Pillow",
+        "numpy",
+        "flash-attn==2.7.3"
+      ],
+      "no_build_isolation": true
+    },
+    "model_src": {
+      "huggingface": {
+        "model_id": "deepseek-ai/DeepSeek-OCR",
+        "model_revision": "main"
+      },
+      "modelscope": {
+        "model_id": "deepseek-ai/DeepSeek-OCR",
+        "model_revision": "master"
+      }
+    }
   }
 ]

xinference/model/image/ocr/__init__.py

@@ -11,3 +11,8 @@
 # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
 # See the License for the specific language governing permissions and
 # limitations under the License.
+
+from .deepseek_ocr import DeepSeekOCRModel
+from .got_ocr2 import GotOCR2Model
+
+__all__ = ["DeepSeekOCRModel", "GotOCR2Model"]