Fix HF Space

README.md

@@ -31,7 +31,7 @@ Users enter keys in the **API keys** panel in the app (password fields, session-
 | Variable | Where to get it |
 |----------|-----------------|
 | `ANTHROPIC_API_KEY` | [console.anthropic.com/settings/api-keys](https://console.anthropic.com/settings/api-keys) |
-| `TABPFN_API_TOKEN` | [ux.priorlabs.ai](https://ux.priorlabs.ai) → API Keys |
+| `TABPFN_API_TOKEN` | [ux.priorlabs.ai](https://ux.priorlabs.ai?utm_source=hf-demo&utm_medium=tuana) → API Keys |
 
 Never commit keys to git. HF Space secrets are optional (auto-connect without showing keys in the UI).
 
@@ -147,6 +147,7 @@ Cold start can take 1–3 minutes because of TabPFN cloud fit + dependency insta
 | App times out on startup | Retry build; consider **CPU upgrade** in Space hardware settings |
 | Works locally but not on HF | Ensure `README.md` YAML has `sdk: gradio` and `app_file: app.py` |
 | `python:3.1: not found` | Quote the version in YAML: `python_version: "3.10"` (bare `3.10` parses as `3.1`) |
+| `unhashable type: 'dict'` / localhost error | `starlette>=0.40` breaks Gradio 4.44 — use pinned `requirements.txt` |
 
 ### 7. Updating the Space
 

app.py

@@ -271,7 +271,7 @@ HIDDEN = gr.update(visible=False)
 SHOW = gr.update(visible=True)
 LOADING_MD = (
     "⏳ **Please wait…** Running TabPFN prediction and computing "
-    "Shapley attributions ([shapiq](https://docs.priorlabs.ai/capabilities/interpretability))."
+    "Shapley attributions ([shapiq](https://docs.priorlabs.ai/capabilities/interpretability?utm_source=hf-demo&utm_medium=tuana))."
 )
 FEATURES_HEADING = "#### Extracted Features"
 PROBS_HEADING = "#### Class Probabilities"
@@ -611,9 +611,9 @@ The CSV is downloaded from UCI on startup — no upload needed.
     with gr.Accordion("TabPFN API key (Prior Labs)", open=False):
         gr.Markdown(
             """
-Predictions run on **TabPFN-3** via the [Prior Labs](https://priorlabs.ai) cloud API:
+Predictions run on **TabPFN-3** via the [Prior Labs](https://priorlabs.ai?utm_source=hf-demo&utm_medium=tuana) cloud API:
 
-1. Create an account at **[ux.priorlabs.ai](https://ux.priorlabs.ai)**
+1. Create an account at **[ux.priorlabs.ai](https://ux.priorlabs.ai?utm_source=hf-demo&utm_medium=tuana)**
 2. Open **API Keys** in the dashboard
 3. Create a key and paste it below (with your Anthropic key)
 
@@ -627,7 +627,7 @@ Free-tier usage limits apply on Prior Labs; see their docs for current quotas.
     with gr.Accordion("🔑 API keys (required to predict)", open=False):
         gr.Markdown(
             "You need your own **Anthropic** key ([get one here](https://console.anthropic.com/settings/api-keys)) "
-            "and **TabPFN** token ([Prior Labs](https://ux.priorlabs.ai) → API Keys). "
+            "and **TabPFN** token ([Prior Labs](https://ux.priorlabs.ai?utm_source=hf-demo&utm_medium=tuana) → API Keys). "
             + KEYS_PROMPT
         )
         with gr.Row():
@@ -750,7 +750,6 @@ Free-tier usage limits apply on Prior Labs; see their docs for current quotas.
     )
 
 if __name__ == "__main__":
-    demo.queue().launch(
-        server_name="0.0.0.0",
-        server_port=int(os.environ.get("PORT", 7860)),
-    )
+    # HF Spaces sets GRADIO_SERVER_NAME / PORT; avoid manual launch args that
+    # break health checks. Starlette must stay <0.40 for Gradio 4.44 (see requirements).
+    demo.queue().launch()

requirements.txt

@@ -1,4 +1,6 @@
 gradio==4.44.1
+starlette>=0.35.0,<0.40.0
+fastapi>=0.100.0,<0.116.0
 huggingface_hub>=0.19.3,<1.0
 anthropic>=0.25.0
 tabpfn-client>=0.3.0