xinference 0.13.4__py3-none-any.whl → 0.14.0.post1__py3-none-any.whl

xinference/_version.py

@@ -8,11 +8,11 @@ import json
 
 version_json = '''
 {
- "date": "2024-08-02T16:08:07+0800",
+ "date": "2024-08-05T11:58:50+0800",
  "dirty": false,
  "error": null,
- "full-revisionid": "dd85cfe015c9cd2d8110c79213640aa0e21f3a6a",
- "version": "0.13.4"
+ "full-revisionid": "111299317120411f407b015b2b7dbf8402aa35c8",
+ "version": "0.14.0.post1"
 }
 '''  # END VERSION_JSON
 

xinference/api/restful_api.py

@@ -46,7 +46,7 @@ from fastapi.staticfiles import StaticFiles
 from PIL import Image
 from sse_starlette.sse import EventSourceResponse
 from starlette.responses import JSONResponse as StarletteJSONResponse
-from starlette.responses import RedirectResponse
+from starlette.responses import PlainTextResponse, RedirectResponse
 from uvicorn import Config, Server
 from xoscar.utils import get_next_port
 
@@ -235,6 +235,13 @@ class RESTfulAPI:
             allow_headers=["*"],
         )
 
+        @self._app.exception_handler(500)
+        async def internal_exception_handler(request: Request, exc: Exception):
+            logger.exception("Handling request %s failed: %s", request.url, exc)
+            return PlainTextResponse(
+                status_code=500, content=f"Internal Server Error: {exc}"
+            )
+
         # internal interface
         self._router.add_api_route("/status", self.get_status, methods=["GET"])
         # conflict with /v1/models/{model_uid} below, so register this first

xinference/core/worker.py

@@ -226,6 +226,7 @@ class WorkerActor(xo.StatelessActor):
         )
         from ..model.flexible import (
             FlexibleModelSpec,
+            generate_flexible_model_description,
             get_flexible_model_descriptions,
             register_flexible_model,
             unregister_flexible_model,
@@ -287,6 +288,7 @@ class WorkerActor(xo.StatelessActor):
                 FlexibleModelSpec,
                 register_flexible_model,
                 unregister_flexible_model,
+                generate_flexible_model_description,
             ),
         }
 

xinference/model/embedding/core.py

@@ -151,8 +151,8 @@ class EmbeddingModel:
 
         patch_trust_remote_code()
         if (
-            "gte-Qwen2" in self._model_spec.model_id
-            or "gte-Qwen2" in self._model_spec.model_name
+            "gte" in self._model_spec.model_name.lower()
+            and "qwen2" in self._model_spec.model_name.lower()
         ):
             self._model = XSentenceTransformer(
                 self._model_path,
@@ -260,8 +260,8 @@ class EmbeddingModel:
                 device = model._target_device
 
             if (
-                "gte-Qwen2" not in self._model_spec.model_id
-                and "gte-Qwen2" not in self._model_spec.model_name
+                "gte" in self._model_spec.model_name.lower()
+                and "qwen2" in self._model_spec.model_name.lower()
             ):
                 model.to(device)
 
@@ -342,8 +342,8 @@ class EmbeddingModel:
             return all_embeddings, all_token_nums
 
         if (
-            "gte-Qwen2" in self._model_spec.model_id
-            or "gte-Qwen2" in self._model_spec.model_name
+            "gte" in self._model_spec.model_name.lower()
+            and "qwen2" in self._model_spec.model_name.lower()
         ):
             all_embeddings, all_token_nums = encode(
                 self._model,

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

@@ -1,14 +1,14 @@
 {
   "files": {
     "main.css": "./static/css/main.4bafd904.css",
-    "main.js": "./static/js/main.af906659.js",
+    "main.js": "./static/js/main.ef2a203a.js",
     "static/media/icon.webp": "./static/media/icon.4603d52c63041e5dfbfd.webp",
     "index.html": "./index.html",
     "main.4bafd904.css.map": "./static/css/main.4bafd904.css.map",
-    "main.af906659.js.map": "./static/js/main.af906659.js.map"
+    "main.ef2a203a.js.map": "./static/js/main.ef2a203a.js.map"
   },
   "entrypoints": [
     "static/css/main.4bafd904.css",
-    "static/js/main.af906659.js"
+    "static/js/main.ef2a203a.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.af906659.js"></script><link href="./static/css/main.4bafd904.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.ef2a203a.js"></script><link href="./static/css/main.4bafd904.css" rel="stylesheet"></head><body><noscript>You need to enable JavaScript to run this app.</noscript><div id="root"></div></body></html>