improve upload validation

backend/lighteval_task/lighteval_task.py

@@ -218,10 +218,11 @@ def process_judge_response_yourbench(response):
 class JudgeLLMYourBench(JudgeLLM):
     def __init__(self):
         super().__init__(
-            judge_model_name="gpt-4o-2024-08-06",
+            judge_model_name="Qwen/QwQ-32B",
             template=get_judge_prompt,
             process_judge_response=process_judge_response_yourbench,
-            judge_backend="openai",
+            judge_backend="inference-providers",
+            hf_provider="novita",
             short_judge_name="yourbench_judge",
         )
 

backend/pyproject.toml

@@ -23,6 +23,8 @@ dependencies = [
     "fastapi>=0.110.0",
     "uvicorn>=0.29.0",
     "pydantic>=2.6.0",
+    "PyPDF2>=3.0.0",
+    "beautifulsoup4>=4.12.0",
 ]
 
 [build-system]

backend/routes/upload.py

@@ -1,7 +1,9 @@
-from fastapi import APIRouter, UploadFile, File
+from fastapi import APIRouter, UploadFile, File, HTTPException
 import os
 import shutil
 import uuid
+from bs4 import BeautifulSoup
+from PyPDF2 import PdfReader
 
 router = APIRouter(tags=["files"])
 
@@ -12,6 +14,43 @@ session_files = {}
 UPLOAD_ROOT = "uploaded_files"
 os.makedirs(UPLOAD_ROOT, exist_ok=True)
 
+def validate_pdf(file_path: str) -> bool:
+    """Validate if file is a valid PDF."""
+    try:
+        reader = PdfReader(file_path)
+        # Vérifier que le PDF a au moins une page
+        return len(reader.pages) > 0
+    except:
+        return False
+
+def validate_markdown(file_path: str) -> bool:
+    """Validate if file is a valid Markdown file."""
+    try:
+        with open(file_path, 'r', encoding='utf-8') as f:
+            content = f.read()
+            # Simple check: file should contain some content and at least one markdown element
+            return len(content) > 0 and any(marker in content for marker in ['#', '-', '*', '`', '[', '>'])
+    except:
+        return False
+
+def validate_html(file_path: str) -> bool:
+    """Validate if file is a valid HTML file."""
+    try:
+        with open(file_path, 'r', encoding='utf-8') as f:
+            BeautifulSoup(f.read(), 'html.parser')
+            return True
+    except:
+        return False
+
+def validate_txt(file_path: str) -> bool:
+    """Validate if file is a valid text file."""
+    try:
+        with open(file_path, 'r', encoding='utf-8') as f:
+            content = f.read()
+            return len(content.strip()) > 0
+    except:
+        return False
+
 # Initialize session files dictionary with pre-calculated documents
 precalculated_docs = ["the-bitter-lesson", "hurricane-faq", "pokemon-guide"]
 
@@ -48,7 +87,7 @@ async def upload_file(file: UploadFile = File(...)):
     """
     # Vérifier si le fichier est un PDF, TXT, HTML ou MD
     if not file.filename.endswith(('.pdf', '.txt', '.html', '.md')):
-        return {"error": "Only PDF, TXT, HTML and MD files are accepted"}
+        raise HTTPException(status_code=400, detail="Only PDF, TXT, HTML and MD files are accepted")
     
     # Get the file extension
     file_extension = os.path.splitext(file.filename)[1].lower()
@@ -71,11 +110,23 @@ async def upload_file(file: UploadFile = File(...)):
     with open(file_path, "wb") as buffer:
         shutil.copyfileobj(file.file, buffer)
     
+    # Valider le fichier selon son type
+    is_valid = False
+    if file_extension == '.pdf':
+        is_valid = validate_pdf(file_path)
+    elif file_extension == '.md':
+        is_valid = validate_markdown(file_path)
+    elif file_extension == '.html':
+        is_valid = validate_html(file_path)
+    elif file_extension == '.txt':
+        is_valid = validate_txt(file_path)
+    
+    if not is_valid:
+        # Supprimer le fichier invalide
+        os.remove(file_path)
+        raise HTTPException(status_code=400, detail=f"Invalid {file_extension[1:].upper()} file")
+    
     # Store file path for later use
     session_files[session_id] = file_path
     
-    # Débogage pour vérifier l'état des session_files
-    print(f"DEBUG UPLOAD: File uploaded with session_id: {session_id}")
-    print(f"DEBUG UPLOAD: Current session_files: {session_files}")
-    
     return {"filename": standardized_filename, "status": "uploaded", "session_id": session_id} 

frontend/src/components/BenchmarkCreateForm.jsx

@@ -143,7 +143,7 @@ function BenchmarkCreateForm({ onStartGeneration }) {
       if (response.ok) {
         setUploadStatus({
           success: true,
-          message: `File ${result.filename} uploaded successfully`,
+          message: "File uploaded successfully",
         });
         setOpenSnackbar(true);
         setSessionId(result.session_id);
@@ -151,7 +151,7 @@ function BenchmarkCreateForm({ onStartGeneration }) {
       } else {
         setUploadStatus({
           success: false,
-          message: result.error || "Upload failed",
+          message: result.detail || "Upload failed",
         });
         setOpenSnackbar(true);
       }