Fetch GAIA attachments directly from the dataset, bypass broken proxy

Scoring server's /files/{task_id} always 404s (confirmed bug, matching open unmerged PR on agents-course/Unit4_scoring). Fetch attachments directly from gaia-benchmark/GAIA dataset at 2023/validation/<file_name> via hf_hub_download() with the user's OAuth token instead. Verified working for png/mp3/py.

app.py

@@ -7,6 +7,7 @@ import requests
 import inspect
 import pandas as pd
 import spaces
+from huggingface_hub import hf_hub_download
 
 from smolagents import (
     ActionStep,
@@ -22,6 +23,17 @@ from smolagents import (
 # --- Constants ---
 DEFAULT_API_URL = "https://agents-course-unit4-scoring.hf.space"
 
+# The scoring server's own /files/{task_id} endpoint is broken (confirmed via
+# a direct request and a matching open PR on agents-course/Unit4_scoring:
+# it stores GAIA attachment paths as HF-Hub-repo-relative paths but checks
+# them as local filesystem paths, so every lookup 404s: "No file path
+# associated with task_id ..."). Attachments live directly in the gated
+# gaia-benchmark/GAIA dataset instead; fetch them from there using the
+# logged-in user's own OAuth token (requires that account to have requested
+# access to the dataset, and the `gated-repos` OAuth scope in README.md).
+GAIA_DATASET_REPO = "gaia-benchmark/GAIA"
+GAIA_DATASET_SUBDIR = "2023/validation"
+
 
 @spaces.GPU
 def _zerogpu_startup_check():
@@ -175,7 +187,7 @@ class BasicAgent:
         print(f"Agent returning answer: {answer}")
         return answer
 
-def run_and_submit_all( profile: gr.OAuthProfile | None):
+def run_and_submit_all( profile: gr.OAuthProfile | None, oauth_token: gr.OAuthToken | None):
     """
     Fetches all questions, runs the BasicAgent on them, submits all answers,
     and displays the results.
@@ -241,14 +253,25 @@ def run_and_submit_all( profile: gr.OAuthProfile | None):
             file_path = None
             if file_name:
                 try:
-                    file_response = requests.get(f"{api_url}/files/{task_id}", timeout=30)
-                    file_response.raise_for_status()
-                    file_path = os.path.join(tmp_dir, file_name)
-                    with open(file_path, "wb") as f:
-                        f.write(file_response.content)
-                except requests.exceptions.RequestException as e:
-                    print(f"Could not download attached file for task {task_id}: {e}")
-                    file_path = None
+                    file_path = hf_hub_download(
+                        repo_id=GAIA_DATASET_REPO,
+                        repo_type="dataset",
+                        filename=f"{GAIA_DATASET_SUBDIR}/{file_name}",
+                        token=oauth_token.token if oauth_token else None,
+                    )
+                except Exception as e:
+                    print(f"Could not download attached file for task {task_id} from {GAIA_DATASET_REPO}: {e}")
+                    # Fall back to the scoring server's own endpoint, in case
+                    # it's since been fixed (see GAIA_DATASET_REPO comment above).
+                    try:
+                        file_response = requests.get(f"{api_url}/files/{task_id}", timeout=30)
+                        file_response.raise_for_status()
+                        file_path = os.path.join(tmp_dir, file_name)
+                        with open(file_path, "wb") as f:
+                            f.write(file_response.content)
+                    except requests.exceptions.RequestException as e2:
+                        print(f"Fallback download also failed for task {task_id}: {e2}")
+                        file_path = None
 
             try:
                 submitted_answer = agent(question_text, file_path=file_path)