fix

README.md

@@ -35,12 +35,13 @@ remain on CPU Basic: when a user starts training, AutoTrain SpaceRunner creates
 temporary L40S training Space under that user's account and publishes the finished
 LoRA to their model repository.
 
-Captioning defaults to `lienthealien/mage-vl-demo`, which uses the Apache-2.0
-`microsoft/Mage-VL` model. If that Space is private, add
-`MAGE_VL_TOKEN` as a secret in this Space's settings. It only needs permission to
-call the Mage-VL Space. `MAGE_VL_SPACE_ID` and `MAGE_VL_API_NAME` can be set as
-Space variables if you use a different duplicate or endpoint. Uploaded images are
-sent to that configured Space for captioning, so use a Space you trust.
+Captioning defaults to the public `microsoft/mage-vl-demo`, which uses the
+Apache-2.0 `microsoft/Mage-VL` model. No repository token is required for reading
+its API. `MAGE_VL_SPACE_ID` and `MAGE_VL_API_NAME` can be set as Space variables
+if you use a different duplicate or endpoint. `MAGE_VL_BASE_URL` can override the
+derived `https://OWNER-SPACE.hf.space` URL. If that override is private, add a token
+with access to it as the `MAGE_VL_TOKEN` Space secret. Uploaded images are sent to
+the configured captioning Space, so use a Space you trust.
 
 The user must:
 

app.py

@@ -1,10 +1,11 @@
 import gradio as gr
 import subprocess
 import os
+import requests
 is_spaces = True if os.environ.get('SPACE_ID') else False
 if is_spaces:
     import spaces
-from gradio_client import Client, handle_file, utils as gradio_client_utils
+from gradio_client import utils as gradio_client_utils
 from huggingface_hub import snapshot_download, HfApi
 import uuid
 import shutil
@@ -28,8 +29,12 @@ TRAINING_SCRIPT = Path("train_dreambooth_lora_sdxl_advanced.py")
 training_script_url = f"https://raw.githubusercontent.com/huggingface/diffusers/{DIFFUSERS_COMMIT}/examples/advanced_diffusion_training/{TRAINING_SCRIPT.name}"
 orchestrator_script_url = "https://huggingface.co/datasets/multimodalart/lora-ease-helper/raw/main/script.py"
 
-MAGE_VL_SPACE_ID = os.environ.get("MAGE_VL_SPACE_ID", "lienthealien/mage-vl-demo")
+MAGE_VL_SPACE_ID = os.environ.get("MAGE_VL_SPACE_ID", "microsoft/mage-vl-demo")
 MAGE_VL_API_NAME = os.environ.get("MAGE_VL_API_NAME", "/ask_image")
+MAGE_VL_BASE_URL = os.environ.get(
+    "MAGE_VL_BASE_URL",
+    f"https://{MAGE_VL_SPACE_ID.replace('/', '-').lower()}.hf.space",
+).rstrip("/")
 mage_vl_client = None
 caption_cache = {}
 
@@ -76,12 +81,75 @@ def get_face_prior_dataset():
     return dataset_path
 
 
+class MageVLClient:
+    """Minimal Gradio HTTP client, independent of the local Gradio version."""
+
+    def __init__(self, base_url, api_name, token=None):
+        self.base_url = base_url
+        self.api_name = api_name.strip("/")
+        self.session = requests.Session()
+        if token:
+            self.session.headers.update({"Authorization": f"Bearer {token}"})
+
+        response = self.session.get(f"{self.base_url}/gradio_api/info", timeout=30)
+        response.raise_for_status()
+        endpoints = response.json().get("named_endpoints", {})
+        if f"/{self.api_name}" not in endpoints:
+            raise RuntimeError(
+                f"Mage-VL endpoint '/{self.api_name}' was not found at {self.base_url}."
+            )
+
+    def predict(self, image, instruction, max_new_tokens):
+        with open(image, "rb") as image_file:
+            upload = self.session.post(
+                f"{self.base_url}/gradio_api/upload",
+                files={"files": (os.path.basename(image), image_file)},
+                timeout=60,
+            )
+        upload.raise_for_status()
+        uploaded_path = upload.json()[0]
+
+        payload = {
+            "data": [
+                {
+                    "path": uploaded_path,
+                    "orig_name": os.path.basename(image),
+                    "meta": {"_type": "gradio.FileData"},
+                },
+                instruction,
+                int(max_new_tokens),
+            ]
+        }
+        start = self.session.post(
+            f"{self.base_url}/gradio_api/call/{self.api_name}",
+            json=payload,
+            timeout=30,
+        )
+        start.raise_for_status()
+        event_id = start.json()["event_id"]
+
+        result = self.session.get(
+            f"{self.base_url}/gradio_api/call/{self.api_name}/{event_id}",
+            timeout=180,
+        )
+        result.raise_for_status()
+        event = None
+        for line in result.text.splitlines():
+            if line.startswith("event:"):
+                event = line.partition(":")[2].strip()
+            elif line.startswith("data:") and event == "complete":
+                return json.loads(line.partition(":")[2].strip())
+            elif line.startswith("data:") and event == "error":
+                raise RuntimeError(line.partition(":")[2].strip())
+        raise RuntimeError(f"Mage-VL returned no completed result: {result.text[:500]}")
+
+
 def get_captioner():
     """Connect to the dedicated Mage-VL Space without loading a VLM here."""
     global mage_vl_client
     if mage_vl_client is None:
-        token = os.environ.get("MAGE_VL_TOKEN") or os.environ.get("HF_TOKEN")
-        mage_vl_client = Client(MAGE_VL_SPACE_ID, hf_token=token, verbose=False)
+        token = os.environ.get("MAGE_VL_TOKEN")
+        mage_vl_client = MageVLClient(MAGE_VL_BASE_URL, MAGE_VL_API_NAME, token)
     return mage_vl_client
 
 
@@ -91,12 +159,7 @@ def caption_with_mage(client, image, instruction):
     if cache_key in caption_cache:
         return caption_cache[cache_key]
 
-    result = client.predict(
-        handle_file(image),
-        instruction,
-        160,
-        api_name=MAGE_VL_API_NAME,
-    )
+    result = client.predict(image, instruction, 160)
     caption = result[0] if isinstance(result, (list, tuple)) else result
     caption = str(caption).strip().rstrip(" .,")
     caption_cache[cache_key] = caption