spaces 0.52.0.dev0__tar.gz → 0.52.0.dev2__tar.gz

{spaces-0.52.0.dev0 → spaces-0.52.0.dev2}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: spaces
-Version: 0.52.0.dev0
+Version: 0.52.0.dev2
 Summary: Utilities for Hugging Face Spaces
 Author: Charles Bensimon
 Author-email: Charles Bensimon <charles@huggingface.co>

{spaces-0.52.0.dev0 → spaces-0.52.0.dev2}/pyproject.toml

@@ -1,6 +1,6 @@
 [project]
 name = "spaces"
-version = "0.52.0.dev0"
+version = "0.52.0.dev2"
 description = "Utilities for Hugging Face Spaces"
 readme = "README.md"
 requires-python = ">=3.10,<3.15"

{spaces-0.52.0.dev0 → spaces-0.52.0.dev2}/spaces/__init__.py

@@ -19,8 +19,8 @@ if (gr := sys.modules.get('gradio')) is not None: # pragma: no cover
 from .zero.decorator import GPU
 
 
-def gradio_auto_wrap(*args, **kwargs): # pragma: no cover
-    warnings.warn("gradio_auto_wrap has no effect and is deprecated")
+def gradio_auto_wrap(task): # pragma: no cover
+    return task
 
 def disable_gradio_auto_wrap(*args, **kwargs): # pragma: no cover
     warnings.warn("gradio_auto_wrap has no effect and is deprecated")

{spaces-0.52.0.dev0 → spaces-0.52.0.dev2}/spaces/zero/client.py

@@ -18,7 +18,7 @@ from .api import GPUSize
 from .api import AuthLevel
 from .api import QuotaInfos
 from .api import ScheduleResponse
-from .context import ctx_var
+from .context import get_context
 from .context import ZeroGPUContext
 
 
@@ -104,7 +104,7 @@ def schedule(
     _first_attempt: bool = True,
 ) -> ScheduleResponse:
 
-    ctx = ctx_var.get()
+    ctx = get_context()
     token, payload = _get_token_and_payload(ctx.headers)
     if token is not None and (token_error := payload.get('error')):
         message = f"Falling back to IP-based quotas ({token_error})"
@@ -146,6 +146,7 @@ def schedule(
             )
             if _is_api_call(ctx):
                 if auth is None or auth == 'regular':
+                    message += ". "
                     message += (
                         f"Subscribe to Hugging Face PRO to allow GPU tasks up to {PRO_QUOTA_MINUTES} min - "
                         f"{SUBSCRIBE_TO_PRO_URL}"
@@ -159,31 +160,33 @@ def schedule(
         else:
             title = "ZeroGPU quota exceeded"
             if res.wait == timedelta(0):
-                message = "You have exceeded your ZeroGPU runs limit. "
+                message = "You have exceeded your ZeroGPU runs limit"
             else:
                 gpu = "Pro ZeroGPU" if auth == 'pro' else ("free ZeroGPU" if auth == 'regular' else "ZeroGPU")
                 message = (
                     f"You have exceeded your {gpu} quota "
                     f"({requested}s requested vs. {res.left}s left). "
-                    f"Try again in {res.wait}. "
+                    f"Try again in {res.wait}"
                 )
             if _is_api_call(ctx):
-                if auth is None:
-                    message += (
-                        f"Authenticate with a Hugging Face token for more quota - "
-                        f"{TOKENS_SETTINGS_URL}"
-                    )
-                elif auth == 'regular':
-                    message += (
-                        f"Subscribe to Hugging Face PRO to get {PRO_QUOTA_MINUTES} min of ZeroGPU quota a day - "
-                        f"{SUBSCRIBE_TO_PRO_URL}"
-                    )
-                elif auth == 'pro':
-                    message += (
-                        f"Add credits to keep using ZeroGPU. "
-                        f"$1 for 10 minutes of ZeroGPU inference - "
-                        f"{ADD_CREDITS_URL}"
-                    )
+                message += ". "
+                match auth:
+                    case None:
+                        message += (
+                            f"Authenticate with a Hugging Face token for more quota - "
+                            f"{TOKENS_SETTINGS_URL}"
+                        )
+                    case 'regular':
+                        message += (
+                            f"Subscribe to Hugging Face PRO to get {PRO_QUOTA_MINUTES} min of ZeroGPU quota a day - "
+                            f"{SUBSCRIBE_TO_PRO_URL}"
+                        )
+                    case 'pro':
+                        message += (
+                            f"Add credits to keep using ZeroGPU. "
+                            f"$1 for 10 minutes of ZeroGPU inference - "
+                            f"{ADD_CREDITS_URL}"
+                        )
         raise ctx.exception(
             title=title,
             message=message,
@@ -284,7 +287,7 @@ def release(
     allow_404: bool = False,
 ) -> None:
 
-    ctx = ctx_var.get()
+    ctx = get_context()
     res = api_client().release(
         allow_token=allow_token,
         fail=fail,
@@ -324,7 +327,7 @@ def _get_token_and_payload(headers: dict[str, str]) -> tuple[str | None, dict[st
 
 
 def duration_warning():
-    ctx_var.get().warning(
+    get_context().warning(
         title="ZeroGPU duration",
         message="GPU task is exceeding its requested duration and might be aborted",
     )

{spaces-0.52.0.dev0 → spaces-0.52.0.dev2}/spaces/zero/context.py

@@ -120,3 +120,13 @@ class ZeroGPUContext:
 
 
 ctx_var: ContextVar[ZeroGPUContext] = ContextVar('zerogpu_context', default=ZeroGPUContext())
+
+
+_context_resolver: Callable[[], ZeroGPUContext | None] | None = None
+
+def get_context() -> ZeroGPUContext:
+    if _context_resolver is not None:
+        ctx = _context_resolver()
+        if ctx is not None:
+            return ctx
+    return ctx_var.get()

{spaces-0.52.0.dev0 → spaces-0.52.0.dev2}/spaces/zero/gradio.py

@@ -5,7 +5,7 @@ import os
 from functools import wraps
 from typing import Any
 from typing import Literal
-from typing import Type
+from typing import cast
 
 import gradio as gr
 from fastapi import FastAPI
@@ -18,8 +18,14 @@ from gradio.routes import App
 
 from ..utils import create_thread_daemon
 from ..utils import SimpleQueue
+from . import context
+from .context import ZeroGPUContext
+from .context import ZeroGPUErrorResponse
+from .context import ZeroGPULog
 from .lifecycle import zerogpu_mode_enabled
 from .lifecycle import zerogpu_startup
+from .middleware import ASGIScope
+from .middleware import get_server_port
 from .middleware import ZeroGPUMiddleware
 
 
@@ -84,24 +90,42 @@ def patch():
             case 'warning':
                 return gr.Warning(message, **kwargs)
 
+    def exception_mapper(err: ZeroGPUErrorResponse, exc: Exception | None):
+        if isinstance(exc, gr.Error):
+            setattr(exc, 'print_exception', False)
+            return exc
+        detail = err['detail']
+        return error(
+            title=detail['title'],
+            message=detail['message'],
+        )
+
+    def log_emitter(log: ZeroGPULog):
+        return info(
+            title=log['title'],
+            message=log['message'],
+            level=log['level'],
+        )
+
+    def resolve_context():
+        if zerogpu_mode_enabled() and (request := LocalContext.request.get(None)) is not None:
+            scope = cast(ASGIScope | None, request.request.scope if request.request is not None else None)
+            return ZeroGPUContext(
+                exception_mapper=exception_mapper,
+                log_emitter=log_emitter,
+                worker_contextualizer=WorkerContextualizer,
+                disconnect_detector=disconnect_detector,
+                server_port=get_server_port(scope) if scope is not None else None,
+                headers={key.lower(): value for key, value in dict(getattr(request, 'headers', {})).items()},
+            )
+
     def setup_middleware(app: FastAPI, *, skip_zerogpu_startup: bool = False):
         mounted = any(middleware.cls is ZeroGPUMiddleware for middleware in app.user_middleware)
         if not mounted:
             app.add_middleware(
                 ZeroGPUMiddleware,
-                exception_mapper=lambda err, exc: (
-                    setattr(exc, 'print_exception', False) or exc
-                    if isinstance(exc, gr.Error)
-                    else error(
-                        title=err['detail']['title'],
-                        message=err['detail']['message'],
-                    )
-                ),
-                log_emitter=lambda log: info(
-                    title=log['title'],
-                    message=log['message'],
-                    level=log['level'],
-                ),
+                exception_mapper=exception_mapper,
+                log_emitter=log_emitter,
                 worker_contextualizer=WorkerContextualizer,
                 disconnect_detector=disconnect_detector,
                 skip_zerogpu_startup=skip_zerogpu_startup,
@@ -129,6 +153,7 @@ def patch():
     gr.mount_gradio_app = mount_gradio_app
     routes.mount_gradio_app = mount_gradio_app
     http_server.start_server = start_server
+    context._context_resolver = resolve_context
     _patch_queue()
 
 

{spaces-0.52.0.dev0 → spaces-0.52.0.dev2}/spaces/zero/middleware.py

@@ -87,7 +87,7 @@ class ZeroGPUMiddleware:
             worker_contextualizer=self.worker_contextualizer,
             disconnect_detector=self.disconnect_detector,
             connection_state=connection_state,
-            server_port=_server_port(asgi_scope),
+            server_port=get_server_port(asgi_scope),
             headers=_decode_headers(asgi_scope.get('headers', [])),
         ))
 
@@ -97,7 +97,7 @@ class ZeroGPUMiddleware:
             ctx_var.reset(token)
 
 
-def _server_port(scope: ASGIScope):
+def get_server_port(scope: ASGIScope):
     match scope.get('server'):
         case (_, int() as port):
             return port

{spaces-0.52.0.dev0 → spaces-0.52.0.dev2}/spaces/zero/wrappers.py

@@ -34,7 +34,7 @@ from .api import AllowToken
 from .api import GPUSize
 from .api import NvidiaIndex
 from .api import NvidiaUUID
-from .context import ctx_var
+from .context import get_context
 from .tqdm import remove_tqdm_multiprocessing_lock
 from .tqdm import tqdm
 from .utils import apply_cleanups
@@ -170,7 +170,7 @@ def regular_function_wrapper(
         if forked:
             return task(*args, **kwargs)
 
-        ctx = ctx_var.get()
+        ctx = get_context()
         duration_ = static_duration(duration, *args, **kwargs)
         duration_ = process_duration(duration_)
         schedule_response = client.schedule(task_id=task_id, duration=duration_, gpu_size=gpu_size)
@@ -297,7 +297,7 @@ def generator_function_wrapper(
             yield from task(*args, **kwargs)
             return
 
-        ctx = ctx_var.get()
+        ctx = get_context()
         duration_ = static_duration(duration, *args, **kwargs)
         duration_ = process_duration(duration_)
         schedule_response = client.schedule(task_id=task_id, duration=duration_, gpu_size=gpu_size)