Scalability improvement: HF login, dynamic test updates, improved warnings for downloading issues

README.md

@@ -8,6 +8,10 @@ sdk_version: 5.20.0
 app_file: app.py
 pinned: false
 license: apache-2.0
+hf_oauth: true
+hf_oauth_scopes:
+  - read-repos
+  - gated-repos
 ---
 
 Check out the configuration reference at https://huggingface.co/docs/hub/spaces-config-reference

app.py

@@ -5,7 +5,7 @@ import gradio_client.utils as _gc_utils
 import json
 import time
 import traceback
-from validation import validate_json, validate_croissant, validate_records, validate_rai, generate_validation_report
+from validation import validate_json, validate_croissant, validate_records, validate_rai, generate_validation_report, set_active_token, clear_active_token
 
 # Patch gradio_client.utils to handle boolean JSON schemas.
 # gradio_client 1.7.2 crashes when a component schema has `additionalProperties: true/false`.
@@ -41,15 +41,15 @@ def process_file(file):
     if not croissant_valid:
         return results, None
 
-    # Check 3: Records validation (with timeout-safe and error-specific logic)
+    # Check 3: Responsible AI metadata
+    rai_valid, rai_message = validate_rai(json_data)
+    results.append(("Responsible AI Metadata", rai_valid, rai_message, "pass" if rai_valid else "error"))
+
+    # Check 4: Records validation (with timeout-safe and error-specific logic)
     records_valid, records_message, records_status = validate_records(json_data)
     records_message = records_message.replace("\n✓\n", "\n")
     results.append(("Records Generation Test", records_valid, records_message, records_status))
 
-    # Check 4: Responsible AI metadata
-    rai_valid, rai_message = validate_rai(json_data)
-    results.append(("Responsible AI Metadata", rai_valid, rai_message, "pass" if rai_valid else "error"))
-
     # Generate final report
     report = generate_validation_report(filename, json_data, results)
 
@@ -68,10 +68,43 @@ def create_ui():
         The validator will check:
         1. If the file is valid JSON
         2. If it passes Croissant schema validation
-        3. If records can be generated within a reasonable time
-        4. If all required Responsible AI metadata fields are present
+        3. If all required Responsible AI metadata fields are present
+        4. If records can be generated within a reasonable time
         """)
-        
+
+        gr.HTML("""
+        <style>
+        #login-card {
+            display: flex !important;
+            flex-direction: column !important;
+            align-items: center !important;
+            gap: 10px !important;
+            padding: 16px 24px !important;
+            border: 1px solid rgba(128,128,128,0.35) !important;
+            border-radius: 10px !important;
+            background-color: rgba(0,0,0,0.03) !important;
+            margin-bottom: 16px !important;
+        }
+        #login-card > * {
+            background: none !important;
+            border: none !important;
+            box-shadow: none !important;
+        }
+        #login-card .login-button {
+            background-color: white !important;
+            color: black !important;
+            border: 1px solid #ccc !important;
+        }
+        </style>
+        """)
+        if os.environ.get("SPACE_ID"):
+            with gr.Column(elem_id="login-card"):
+                gr.HTML("<p style='text-align:center; margin:0;'>"
+                        "If your dataset is hosted on Hugging Face, log in first. "
+                        "This avoids rate limiting issues and allows you to fetch gated datasets. "
+                        "If your dataset is not on Hugging Face, you can still use the validator without logging in.</p>")
+                gr.LoginButton(elem_classes=["login-button"])
+
         # Track the active tab for conditional UI updates
         active_tab = gr.State("upload")  # Default to upload tab
         
@@ -384,75 +417,86 @@ def create_ui():
                 None   # Clear report file
             ]
                 
-        def fetch_from_url(url):
+        def fetch_from_url(url, oauth_token: gr.OAuthToken | None = None):
             if not url:
-                return [
+                yield [
                     """<div class="progress-status">Please enter a URL</div>""",
                     gr.update(value=""),
                     gr.update(visible=False),
                     None,
                     None
                 ]
+                return
             try:
                 # Fetch JSON from URL
                 response = requests.get(url, timeout=10)
                 response.raise_for_status()
                 json_data = response.json()
-                
-                # Process validation
-                progress_html = """<div class="progress-status">✅ JSON fetched successfully from URL</div>"""
-                
-                # Validate the fetched JSON
+
                 results = []
                 results.append(("JSON Format Validation", True, "The URL returned valid JSON."))
-                
+
                 croissant_valid, croissant_message, croissant_status = validate_croissant(json_data)
                 results.append(("Croissant Schema Validation", croissant_valid, croissant_message, croissant_status))
-                
+
                 if not croissant_valid:
-                    return [
+                    yield [
                         """<div class="progress-status">✅ JSON fetched successfully from URL</div>""",
                         build_results_html(results),
                         gr.update(visible=False),
                         None,
                         None
                     ]
-                
-                records_valid, records_message, records_status = validate_records(json_data)
-                results.append(("Records Generation Test (Optional)", records_valid, records_message, records_status))
+                    return
 
-                # Check 4: Responsible AI metadata
+                # Responsible AI metadata (fast)
                 rai_valid, rai_message = validate_rai(json_data)
                 results.append(("Responsible AI Metadata", rai_valid, rai_message, "pass" if rai_valid else "error"))
 
-                # Generate report
+                # Show partial results while records test runs
+                records_spinner = """<div class="validation-progress"><div class="loading-spinner"></div><span>Running records generation test...</span></div>"""
+                yield [
+                    """<div class="progress-status">✅ JSON fetched successfully from URL</div>""",
+                    gr.update(value=build_results_html(results)["value"] + records_spinner, visible=True),
+                    gr.update(visible=False),
+                    None,
+                    None
+                ]
+
+                # Records validation (slow)
+                set_active_token(oauth_token.token if oauth_token else None)
+                try:
+                    records_valid, records_message, records_status = validate_records(json_data)
+                finally:
+                    clear_active_token()
+                results.append(("Records Generation Test (Optional)", records_valid, records_message, records_status))
+
                 report = generate_validation_report(url.split("/")[-1], json_data, results)
                 report_filename = f"report_croissant-validation_{json_data.get('name', 'unnamed')}.md"
-
                 if report:
                     with open(report_filename, "w") as f:
                         f.write(report)
 
-                return [
+                yield [
                     """<div class="progress-status">✅ JSON fetched successfully from URL</div>""",
                     build_results_html(results),
                     gr.update(visible=True),
                     report,
                     report_filename
                 ]
-                            
+
             except requests.exceptions.RequestException as e:
                 error_message = f"Error fetching URL: {str(e)}"
-                return [
+                yield [
                     f"""<div class="progress-status">{error_message}</div>""",
                     gr.update(value=""),
                     gr.update(visible=False),
                     None,
                     None
-                ]                
+                ]
             except json.JSONDecodeError as e:
                 error_message = f"URL did not return valid JSON: {str(e)}"
-                return [
+                yield [
                     f"""<div class="progress-status">{error_message}</div>""",
                     gr.update(value=""),
                     gr.update(visible=False),
@@ -461,7 +505,7 @@ def create_ui():
                 ]
             except Exception as e:
                 error_message = f"Unexpected error: {str(e)}"
-                return [
+                yield [
                     f"""<div class="progress-status">{error_message}</div>""",
                     gr.update(value=""),
                     gr.update(visible=False),
@@ -487,7 +531,14 @@ def create_ui():
                     status_class = "status-warning"
                     status_icon = "?"
                     if "Records" in test_name:
-                        message_with_emoji = "⚠️ Could not automatically generate records. This is oftentimes not an issue (e.g. datasets could be too large or too complex), and it's not required to pass this test to submit to NeurIPS.\n\n" + message
+                        if "429" in message or "Too Many Requests" in message or "rate limit" in message.lower():
+                            message_with_emoji = (
+                                "⚠️ Rate limit hit while trying to generate records. "
+                                "Log in with your Hugging Face account (button above) to use your own token and avoid this.\n\n"
+                                + message
+                            )
+                        else:
+                            message_with_emoji = "⚠️ Could not automatically generate records. This is oftentimes not an issue (e.g. datasets could be too large or too complex), and it's not required to pass this test to submit to NeurIPS.\n\n" + message
                     else:
                         message_with_emoji = "⚠️ " + message
                 else:  # error
@@ -524,7 +575,7 @@ def create_ui():
             html += '</div>'
             return gr.update(value=html, visible=True)
 
-        def on_validate(file):
+        def on_validate(file, oauth_token: gr.OAuthToken | None = None):
             if file is None:
                 yield [
                     gr.update(value=""),  # validation_results
@@ -535,7 +586,7 @@ def create_ui():
                 ]
                 return
 
-            # Show progress spinner
+            # Show initial spinner
             progress_html = """
             <div class="validation-progress">
                 <div class="loading-spinner"></div>
@@ -543,37 +594,77 @@ def create_ui():
             </div>
             """
             yield [
-                gr.update(value=""),  # validation_results
-                gr.update(visible=True, value=progress_html),  # validation_progress
-                gr.update(visible=False),  # report_group
-                None,  # report_text
-                None   # report_md
+                gr.update(value=""),
+                gr.update(visible=True, value=progress_html),
+                gr.update(visible=False),
+                None,
+                None
             ]
 
-            # Process the file and get results
-            results, report = process_file(file)
+            results = []
+            filename = file.name.split("/")[-1]
 
-            # Extract dataset name from the JSON for the report filename
-            try:
-                with open(file.name, 'r') as f:
-                    json_data = json.load(f)
-                dataset_name = json_data.get('name', 'unnamed')
-            except:
-                dataset_name = 'unnamed'
+            # Check 1: JSON
+            json_valid, json_message, json_data = validate_json(file.name)
+            json_message = json_message.replace("\n✓\n", "\n")
+            results.append(("JSON Format Validation", json_valid, json_message, "pass" if json_valid else "error"))
+            if not json_valid:
+                yield [build_results_html(results), gr.update(visible=False), gr.update(visible=False), None, None]
+                return
 
-            # Save report to file with new naming convention
+            # Check 2: Croissant schema
+            croissant_valid, croissant_message, croissant_status = validate_croissant(json_data)
+            croissant_message = croissant_message.replace("\n✓\n", "\n")
+            results.append(("Croissant Schema Validation", croissant_valid, croissant_message, croissant_status))
+            if not croissant_valid:
+                yield [build_results_html(results), gr.update(visible=False), gr.update(visible=False), None, None]
+                return
+
+            # Check 3: Responsible AI metadata (fast)
+            rai_valid, rai_message = validate_rai(json_data)
+            results.append(("Responsible AI Metadata", rai_valid, rai_message, "pass" if rai_valid else "error"))
+
+            # Show partial results + spinner for records test
+            records_spinner = """
+            <div class="validation-progress">
+                <div class="loading-spinner"></div>
+                <div>
+                    <div>Running records generation test...</div>
+                    <div>This test tries to load a sample of the data and may be slow for large datasets.</div>
+                </div>
+            </div>
+            """
+            yield [
+                build_results_html(results),
+                gr.update(visible=True, value=records_spinner),
+                gr.update(visible=False),
+                None,
+                None
+            ]
+
+            # Check 4: Records (slow)
+            set_active_token(oauth_token.token if oauth_token else None)
+            try:
+                records_valid, records_message, records_status = validate_records(json_data)
+            finally:
+                clear_active_token()
+            records_message = records_message.replace("\n✓\n", "\n")
+            results.append(("Records Generation Test", records_valid, records_message, records_status))
+
+            # Generate report
+            report = generate_validation_report(filename, json_data, results)
+            dataset_name = json_data.get('name', 'unnamed') if isinstance(json_data, dict) else 'unnamed'
             report_filename = f"report_croissant-validation_{dataset_name}.md"
             if report:
                 with open(report_filename, "w") as f:
                     f.write(report)
 
-            # Return final state
             yield [
-                build_results_html(results),  # validation_results
-                gr.update(visible=False),  # validation_progress
-                gr.update(visible=True) if report else gr.update(visible=False),  # report_group
-                report if report else None,  # report_text
-                report_filename if report else None  # report_md
+                build_results_html(results),
+                gr.update(visible=False),
+                gr.update(visible=True) if report else gr.update(visible=False),
+                report if report else None,
+                report_filename if report else None
             ]
 
         # Connect UI events to functions with updated outputs
@@ -619,7 +710,7 @@ def create_ui():
         # Footer
         gr.HTML("""
         <div style="text-align: center; margin-top: 20px;">
-            <p>Learn more about  🥐<a href="https://github.com/mlcommons/croissant" target="_blank">Croissant</a>.</p>
+            <p>Learn more about  🥐 <a href="https://github.com/mlcommons/croissant" target="_blank">Croissant</a>.</p>
         </div>
         """)
 
@@ -627,7 +718,8 @@ def create_ui():
         <div class="progress-status" style="text-align: left; color: #d35400;">
             ⚠️ It is possible that this validator is currently being used by a lot of people at the same time, which may trigger rate limiting by the platform hosting your data. 
             The app will then try again and may get into a very long loop. If it takes too long to run, we recommend using any of the following options:
-            <ul style="text-align:left; margin: 0 auto; display:inline-block;">
+            <ul style="text-align:left; margin: 12px auto 0; display:inline-block;">
+                <li>🤗 If your dataset is on Hugging Face, please log in first to avoid rate limiting issues.</li>
                 <li>🔁 Click the button with the three dots (⋯) above and select "Duplicate this Space" to run this app in your own Hugging Face space.</li>
                 <li>💻 Click the button with the three dots (⋯) above and select "Run Locally" and then "Clone (git)" to get instructions to run the checker locally. You can also use docker option (you don't need the tokens).</li>
                 <li>🥐 Run the Croissant validation code yourself (<a href="https://github.com/mlcommons/croissant" target="_blank">GitHub</a>), e.g. with <a href="https://github.com/mlcommons/croissant/tree/7a632f34438e9c8e3812c6a0049898560259c6d4/python/mlcroissant/mlcroissant/scripts" target="_blank">these scripts</a> (validate and load).</li>

requirements.txt

@@ -1,6 +1,6 @@
 mlcroissant>=1.0.17
 pydantic>=2.10.6
-gradio>=3.50.2
+gradio[oauth]>=3.50.2
 func_timeout
 requests
 huggingface-hub>=0.30.1

validation.py

@@ -1,9 +1,45 @@
 import mlcroissant._src.operation_graph.operations.download as dl_mod
+import requests as _requests_mod
 import requests
 import os
 
-HF_TOKEN = os.environ.get("HF_TOKEN")
-print("[DEBUG] HF_TOKEN is", "set" if HF_TOKEN else "missing")
+_SERVER_HF_TOKEN = os.environ.get("HF_TOKEN")
+print("[DEBUG] HF_TOKEN is", "set" if _SERVER_HF_TOKEN else "missing")
+
+# _active_token holds the HF token to use for the current validation request.
+# It defaults to the server-level HF_TOKEN but can be overridden per-request
+# via set_active_token() so that logged-in users' own tokens are used instead.
+_active_token: dict = {"token": _SERVER_HF_TOKEN}
+
+
+def set_active_token(token: str | None) -> None:
+    """Set the HF token to use for the current validation request."""
+    _active_token["token"] = token if token else _SERVER_HF_TOKEN
+
+
+def clear_active_token() -> None:
+    """Reset the HF token back to the server-level default."""
+    _active_token["token"] = _SERVER_HF_TOKEN
+
+
+# Patch requests.Session.send to fail immediately on HTTP 429 instead of
+# letting mlcroissant / fsspec / huggingface_hub retry silently for minutes.
+_orig_session_send = _requests_mod.Session.send
+
+def _rate_limit_aware_send(self, request, **kwargs):
+    response = _orig_session_send(self, request, **kwargs)
+    if response.status_code == 429:
+        retry_after = response.headers.get("Retry-After", "unknown")
+        raise _requests_mod.exceptions.HTTPError(
+            f"HTTP 429 Too Many Requests for {request.url}. "
+            f"Retry-After: {retry_after}s. "
+            "You are being rate limited. Log in with your Hugging Face account to avoid this.",
+            response=response,
+        )
+    return response
+
+_requests_mod.Session.send = _rate_limit_aware_send
+
 
 # Only send HF credentials when downloading from huggingface.co.
 # The default get_basic_auth_from_env() applies auth to ALL URLs, which
@@ -12,9 +48,10 @@ _orig_download_from_http = dl_mod.Download._download_from_http
 
 def _hf_aware_download(self, filepath):
     url = self.node.content_url or ""
-    if HF_TOKEN and "huggingface.co" in url:
+    token = _active_token["token"]
+    if token and "huggingface.co" in url:
         os.environ["CROISSANT_BASIC_AUTH_USERNAME"] = "hf_user"
-        os.environ["CROISSANT_BASIC_AUTH_PASSWORD"] = HF_TOKEN
+        os.environ["CROISSANT_BASIC_AUTH_PASSWORD"] = token
     else:
         os.environ.pop("CROISSANT_BASIC_AUTH_USERNAME", None)
         os.environ.pop("CROISSANT_BASIC_AUTH_PASSWORD", None)
@@ -134,7 +171,6 @@ RAI_FIELDS = [
     "rai:dataUseCases",
     "rai:dataSocialImpact",
     "rai:hasSyntheticData",
-    "prov:wasGeneratedBy",
 ]
 
 RAI_GUIDELINES_URL = "https://neurips.cc/Conferences/2026/EvaluationsDatasetsHosting"