xinference 1.7.1__py3-none-any.whl → 1.8.0__py3-none-any.whl

xinference/model/video/core.py

@@ -12,21 +12,17 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 import logging
-import os
 from collections import defaultdict
-from typing import Any, Dict, List, Literal, Optional, Tuple
+from typing import Any, Dict, List, Literal, Optional
 
-from ...constants import XINFERENCE_CACHE_DIR
-from ..core import CacheableModelSpec, ModelDescription, VirtualEnvSettings
-from ..utils import valid_model_revision
+from ..core import CacheableModelSpec, VirtualEnvSettings
+from ..utils import ModelInstanceInfoMixin
 from .diffusers import DiffusersVideoModel
 
 logger = logging.getLogger(__name__)
 
-MODEL_NAME_TO_REVISION: Dict[str, List[str]] = defaultdict(list)
 VIDEO_MODEL_DESCRIPTIONS: Dict[str, List[Dict]] = defaultdict(list)
-BUILTIN_VIDEO_MODELS: Dict[str, "VideoModelFamilyV1"] = {}
-MODELSCOPE_VIDEO_MODELS: Dict[str, "VideoModelFamilyV1"] = {}
+BUILTIN_VIDEO_MODELS: Dict[str, List["VideoModelFamilyV2"]] = {}
 
 
 def get_video_model_descriptions():
@@ -35,7 +31,8 @@ def get_video_model_descriptions():
     return copy.deepcopy(VIDEO_MODEL_DESCRIPTIONS)
 
 
-class VideoModelFamilyV1(CacheableModelSpec):
+class VideoModelFamilyV2(CacheableModelSpec, ModelInstanceInfoMixin):
+    version: Literal[2]
     model_family: str
     model_name: str
     model_id: str
@@ -46,57 +43,37 @@ class VideoModelFamilyV1(CacheableModelSpec):
     default_generate_config: Optional[Dict[str, Any]]
     virtualenv: Optional[VirtualEnvSettings]
 
+    class Config:
+        extra = "allow"
 
-class VideoModelDescription(ModelDescription):
-    def __init__(
-        self,
-        address: Optional[str],
-        devices: Optional[List[str]],
-        model_spec: VideoModelFamilyV1,
-        model_path: Optional[str] = None,
-    ):
-        super().__init__(address, devices, model_path=model_path)
-        self._model_spec = model_spec
-
-    @property
-    def spec(self):
-        return self._model_spec
-
-    def to_dict(self):
+    def to_description(self):
         return {
             "model_type": "video",
-            "address": self.address,
-            "accelerators": self.devices,
-            "model_name": self._model_spec.model_name,
-            "model_family": self._model_spec.model_family,
-            "model_revision": self._model_spec.model_revision,
-            "model_ability": self._model_spec.model_ability,
+            "address": getattr(self, "address", None),
+            "accelerators": getattr(self, "accelerators", None),
+            "model_name": self.model_name,
+            "model_family": self.model_family,
+            "model_revision": self.model_revision,
+            "model_ability": self.model_ability,
         }
 
     def to_version_info(self):
-        if self._model_path is None:
-            is_cached = get_cache_status(self._model_spec)
-            file_location = get_cache_dir(self._model_spec)
-        else:
-            is_cached = True
-            file_location = self._model_path
+        from ..cache_manager import CacheManager
 
-        return [
-            {
-                "model_version": self._model_spec.model_name,
-                "model_file_location": file_location,
-                "cache_status": is_cached,
-            }
-        ]
+        cache_manager = CacheManager(self)
+
+        return {
+            "model_version": self.model_name,
+            "model_file_location": cache_manager.get_cache_dir(),
+            "cache_status": cache_manager.get_cache_status(),
+        }
 
 
 def generate_video_description(
-    video_model: VideoModelFamilyV1,
+    video_model: VideoModelFamilyV2,
 ) -> Dict[str, List[Dict]]:
     res = defaultdict(list)
-    res[video_model.model_name].extend(
-        VideoModelDescription(None, None, video_model).to_version_info()
-    )
+    res[video_model.model_name].append(video_model.to_version_info())
     return res
 
 
@@ -105,22 +82,19 @@ def match_diffusion(
     download_hub: Optional[
         Literal["huggingface", "modelscope", "openmind_hub", "csghub"]
     ] = None,
-) -> VideoModelFamilyV1:
+) -> VideoModelFamilyV2:
     from ..utils import download_from_modelscope
-    from . import BUILTIN_VIDEO_MODELS, MODELSCOPE_VIDEO_MODELS
-
-    if download_hub == "modelscope" and model_name in MODELSCOPE_VIDEO_MODELS:
-        logger.debug(f"Video model {model_name} found in ModelScope.")
-        return MODELSCOPE_VIDEO_MODELS[model_name]
-    elif download_hub == "huggingface" and model_name in BUILTIN_VIDEO_MODELS:
-        logger.debug(f"Video model {model_name} found in Huggingface.")
-        return BUILTIN_VIDEO_MODELS[model_name]
-    elif download_from_modelscope() and model_name in MODELSCOPE_VIDEO_MODELS:
-        logger.debug(f"Video model {model_name} found in ModelScope.")
-        return MODELSCOPE_VIDEO_MODELS[model_name]
-    elif model_name in BUILTIN_VIDEO_MODELS:
-        logger.debug(f"Video model {model_name} found in Huggingface.")
-        return BUILTIN_VIDEO_MODELS[model_name]
+    from . import BUILTIN_VIDEO_MODELS
+
+    if model_name in BUILTIN_VIDEO_MODELS:
+        model_families = BUILTIN_VIDEO_MODELS[model_name]
+        if download_hub == "modelscope" or download_from_modelscope():
+            return (
+                [x for x in model_families if x.model_hub == "modelscope"]
+                + [x for x in model_families if x.model_hub == "huggingface"]
+            )[0]
+        else:
+            return [x for x in model_families if x.model_hub == "huggingface"][0]
     else:
         raise ValueError(
             f"Video model {model_name} not found, available"
@@ -128,40 +102,7 @@ def match_diffusion(
         )
 
 
-def cache(model_spec: VideoModelFamilyV1):
-    from ..utils import cache
-
-    return cache(model_spec, VideoModelDescription)
-
-
-def get_cache_dir(model_spec: VideoModelFamilyV1):
-    return os.path.realpath(os.path.join(XINFERENCE_CACHE_DIR, model_spec.model_name))
-
-
-def get_cache_status(
-    model_spec: VideoModelFamilyV1,
-) -> bool:
-    cache_dir = get_cache_dir(model_spec)
-    meta_path = os.path.join(cache_dir, "__valid_download")
-
-    model_name = model_spec.model_name
-    if model_name in BUILTIN_VIDEO_MODELS and model_name in MODELSCOPE_VIDEO_MODELS:
-        hf_spec = BUILTIN_VIDEO_MODELS[model_name]
-        ms_spec = MODELSCOPE_VIDEO_MODELS[model_name]
-
-        return any(
-            [
-                valid_model_revision(meta_path, hf_spec.model_revision),
-                valid_model_revision(meta_path, ms_spec.model_revision),
-            ]
-        )
-    else:  # Usually for UT
-        return valid_model_revision(meta_path, model_spec.model_revision)
-
-
 def create_video_model_instance(
-    subpool_addr: str,
-    devices: List[str],
     model_uid: str,
     model_name: str,
     download_hub: Optional[
@@ -169,10 +110,14 @@ def create_video_model_instance(
     ] = None,
     model_path: Optional[str] = None,
     **kwargs,
-) -> Tuple[DiffusersVideoModel, VideoModelDescription]:
+) -> DiffusersVideoModel:
+    from ..cache_manager import CacheManager
+
     model_spec = match_diffusion(model_name, download_hub)
+
     if not model_path:
-        model_path = cache(model_spec)
+        cache_manager = CacheManager(model_spec)
+        model_path = cache_manager.cache()
     assert model_path is not None
 
     model = DiffusersVideoModel(
@@ -181,7 +126,4 @@ def create_video_model_instance(
         model_spec,
         **kwargs,
     )
-    model_description = VideoModelDescription(
-        subpool_addr, devices, model_spec, model_path=model_path
-    )
-    return model, model_description
+    return model

xinference/model/video/diffusers.py

@@ -30,8 +30,8 @@ from ...device_utils import gpu_count, move_model_to_available_device
 from ...types import Video, VideoList
 
 if TYPE_CHECKING:
-    from ....core.progress_tracker import Progressor
-    from .core import VideoModelFamilyV1
+    from ...core.progress_tracker import Progressor
+    from .core import VideoModelFamilyV2
 
 
 logger = logging.getLogger(__name__)
@@ -60,9 +60,10 @@ class DiffusersVideoModel:
         self,
         model_uid: str,
         model_path: str,
-        model_spec: "VideoModelFamilyV1",
+        model_spec: "VideoModelFamilyV2",
         **kwargs,
     ):
+        self.model_family = model_spec
         self._model_uid = model_uid
         self._model_path = model_path
         self._model_spec = model_spec

xinference/model/video/model_spec.json

@@ -1,9 +1,8 @@
 [
   {
+    "version": 2,
     "model_name": "CogVideoX-2b",
     "model_family": "CogVideoX",
-    "model_id": "THUDM/CogVideoX-2b",
-    "model_revision": "4bbfb1de622b80bc1b77b6e9aced75f816be0e38",
     "model_ability": [
       "text2video"
     ],
@@ -13,13 +12,22 @@
     },
     "default_generate_config": {
       "guidance_scale": 6
+    },
+    "model_src": {
+      "huggingface": {
+        "model_id": "THUDM/CogVideoX-2b",
+        "model_revision": "4bbfb1de622b80bc1b77b6e9aced75f816be0e38"
+      },
+      "modelscope": {
+        "model_id": "ZhipuAI/CogVideoX-2b",
+        "model_revision": "master"
+      }
     }
   },
   {
+    "version": 2,
     "model_name": "CogVideoX-5b",
     "model_family": "CogVideoX",
-    "model_id": "THUDM/CogVideoX-5b",
-    "model_revision": "8d6ea3f817438460b25595a120f109b88d5fdfad",
     "model_ability": [
       "text2video"
     ],
@@ -29,13 +37,22 @@
     },
     "default_generate_config": {
       "guidance_scale": 7
+    },
+    "model_src": {
+      "huggingface": {
+        "model_id": "THUDM/CogVideoX-5b",
+        "model_revision": "8d6ea3f817438460b25595a120f109b88d5fdfad"
+      },
+      "modelscope": {
+        "model_id": "ZhipuAI/CogVideoX-5b",
+        "model_revision": "master"
+      }
     }
   },
   {
+    "version": 2,
     "model_name": "HunyuanVideo",
     "model_family": "HunyuanVideo",
-    "model_id": "hunyuanvideo-community/HunyuanVideo",
-    "model_revision": "e8c2aaa66fe3742a32c11a6766aecbf07c56e773",
     "model_ability": [
       "text2video"
     ],
@@ -43,22 +60,29 @@
       "transformer_torch_dtype": "bfloat16",
       "torch_dtype": "float16"
     },
-    "default_generate_config": {
+    "default_generate_config": {},
+    "model_src": {
+      "huggingface": {
+        "model_id": "hunyuanvideo-community/HunyuanVideo",
+        "model_revision": "e8c2aaa66fe3742a32c11a6766aecbf07c56e773"
+      },
+      "modelscope": {
+        "model_id": "Xorbits/HunyuanVideo",
+        "model_revision": "master"
+      }
     }
   },
   {
+    "version": 2,
     "model_name": "Wan2.1-1.3B",
     "model_family": "Wan",
-    "model_id": "Wan-AI/Wan2.1-T2V-1.3B-Diffusers",
-    "model_revision": "0fad780a534b6463e45facd96134c9f345acfa5b",
     "model_ability": [
       "text2video"
     ],
     "default_model_config": {
       "torch_dtype": "bfloat16"
     },
-    "default_generate_config": {
-    },
+    "default_generate_config": {},
     "virtualenv": {
       "packages": [
         "diffusers>=0.33.0",
@@ -67,21 +91,29 @@
         "imageio",
         "#system_numpy#"
       ]
+    },
+    "model_src": {
+      "huggingface": {
+        "model_id": "Wan-AI/Wan2.1-T2V-1.3B-Diffusers",
+        "model_revision": "0fad780a534b6463e45facd96134c9f345acfa5b"
+      },
+      "modelscope": {
+        "model_id": "Wan-AI/Wan2.1-T2V-1.3B-Diffusers",
+        "model_revision": "master"
+      }
     }
   },
   {
+    "version": 2,
     "model_name": "Wan2.1-14B",
     "model_family": "Wan",
-    "model_id": "Wan-AI/Wan2.1-T2V-14B-Diffusers",
-    "model_revision": "38ec498cb3208fb688890f8cc7e94ede2cbd7f68",
     "model_ability": [
       "text2video"
     ],
     "default_model_config": {
       "torch_dtype": "bfloat16"
     },
-    "default_generate_config": {
-    },
+    "default_generate_config": {},
     "virtualenv": {
       "packages": [
         "diffusers>=0.33.0",
@@ -90,13 +122,22 @@
         "imageio",
         "#system_numpy#"
       ]
+    },
+    "model_src": {
+      "huggingface": {
+        "model_id": "Wan-AI/Wan2.1-T2V-14B-Diffusers",
+        "model_revision": "38ec498cb3208fb688890f8cc7e94ede2cbd7f68"
+      },
+      "modelscope": {
+        "model_id": "Wan-AI/Wan2.1-T2V-14B-Diffusers",
+        "model_revision": "master"
+      }
     }
   },
   {
+    "version": 2,
     "model_name": "Wan2.1-i2v-14B-480p",
     "model_family": "Wan",
-    "model_id": "Wan-AI/Wan2.1-I2V-14B-480P-Diffusers",
-    "model_revision": "b184e23a8a16b20f108f727c902e769e873ffc73",
     "model_ability": [
       "image2video"
     ],
@@ -117,13 +158,22 @@
         "imageio",
         "#system_numpy#"
       ]
+    },
+    "model_src": {
+      "huggingface": {
+        "model_id": "Wan-AI/Wan2.1-I2V-14B-480P-Diffusers",
+        "model_revision": "b184e23a8a16b20f108f727c902e769e873ffc73"
+      },
+      "modelscope": {
+        "model_id": "Wan-AI/Wan2.1-I2V-14B-480P-Diffusers",
+        "model_revision": "master"
+      }
     }
   },
   {
+    "version": 2,
     "model_name": "Wan2.1-i2v-14B-720p",
     "model_family": "Wan",
-    "model_id": "Wan-AI/Wan2.1-I2V-14B-720P-Diffusers",
-    "model_revision": "eb849f76dfa246545b65774a9e25943ee69b3fa3",
     "model_ability": [
       "image2video"
     ],
@@ -144,13 +194,22 @@
         "imageio",
         "#system_numpy#"
       ]
+    },
+    "model_src": {
+      "huggingface": {
+        "model_id": "Wan-AI/Wan2.1-I2V-14B-720P-Diffusers",
+        "model_revision": "eb849f76dfa246545b65774a9e25943ee69b3fa3"
+      },
+      "modelscope": {
+        "model_id": "Wan-AI/Wan2.1-I2V-14B-720P-Diffusers",
+        "model_revision": "master"
+      }
     }
   },
   {
+    "version": 2,
     "model_name": "Wan2.1-flf2v-14B-720p",
     "model_family": "Wan",
-    "model_id": "Wan-AI/Wan2.1-FLF2V-14B-720P-diffusers",
-    "model_revision": "17c30769b1e0b5dcaa1799b117bf20a9c31f59d7",
     "model_ability": [
       "firstlastframe2video"
     ],
@@ -171,6 +230,16 @@
         "imageio",
         "#system_numpy#"
       ]
+    },
+    "model_src": {
+      "huggingface": {
+        "model_id": "Wan-AI/Wan2.1-FLF2V-14B-720P-diffusers",
+        "model_revision": "17c30769b1e0b5dcaa1799b117bf20a9c31f59d7"
+      },
+      "modelscope": {
+        "model_id": "Wan-AI/Wan2.1-FLF2V-14B-720P-diffusers",
+        "model_revision": "master"
+      }
     }
   }
 ]

xinference/types.py

@@ -436,9 +436,11 @@ class PeftModelConfig:
 
     def to_dict(self):
         return {
-            "lora_list": [lora.to_dict() for lora in self.peft_model]
-            if self.peft_model
-            else None,
+            "lora_list": (
+                [lora.to_dict() for lora in self.peft_model]
+                if self.peft_model
+                else None
+            ),
             "image_lora_load_kwargs": self.image_lora_load_kwargs,
             "image_lora_fuse_kwargs": self.image_lora_fuse_kwargs,
         }

xinference/web/ui/build/asset-manifest.json

@@ -1,14 +1,14 @@
 {
   "files": {
     "main.css": "./static/css/main.013f296b.css",
-    "main.js": "./static/js/main.9b12b7f9.js",
+    "main.js": "./static/js/main.7d24df53.js",
     "static/media/icon.webp": "./static/media/icon.4603d52c63041e5dfbfd.webp",
     "index.html": "./index.html",
     "main.013f296b.css.map": "./static/css/main.013f296b.css.map",
-    "main.9b12b7f9.js.map": "./static/js/main.9b12b7f9.js.map"
+    "main.7d24df53.js.map": "./static/js/main.7d24df53.js.map"
   },
   "entrypoints": [
     "static/css/main.013f296b.css",
-    "static/js/main.9b12b7f9.js"
+    "static/js/main.7d24df53.js"
   ]
 }

xinference/web/ui/build/index.html

@@ -1 +1 @@
-<!doctype html><html lang="en"><head><meta charset="utf-8"/><link rel="icon" href="./favicon.svg"/><meta name="viewport" content="width=device-width,initial-scale=1"/><meta name="theme-color" content="#000000"/><meta name="description" content="Web site created using create-react-app"/><link rel="apple-touch-icon" href="./logo192.png"/><link rel="manifest" href="./manifest.json"/><title>Xinference</title><script defer="defer" src="./static/js/main.9b12b7f9.js"></script><link href="./static/css/main.013f296b.css" rel="stylesheet"></head><body><noscript>You need to enable JavaScript to run this app.</noscript><div id="root"></div></body></html>
+<!doctype html><html lang="en"><head><meta charset="utf-8"/><link rel="icon" href="./favicon.svg"/><meta name="viewport" content="width=device-width,initial-scale=1"/><meta name="theme-color" content="#000000"/><meta name="description" content="Web site created using create-react-app"/><link rel="apple-touch-icon" href="./logo192.png"/><link rel="manifest" href="./manifest.json"/><title>Xinference</title><script defer="defer" src="./static/js/main.7d24df53.js"></script><link href="./static/css/main.013f296b.css" rel="stylesheet"></head><body><noscript>You need to enable JavaScript to run this app.</noscript><div id="root"></div></body></html>