Update app.py

app.py

@@ -12,16 +12,12 @@ from huggingface_hub import InferenceClient
 # -----------------------------
 # Config
 # -----------------------------
-HF_TOKEN = os.getenv("HUGGINGFACEHUB_API_TOKEN") or os.getenv("HF_TOKEN")
+HF_TOKEN = (os.getenv("HUGGINGFACEHUB_API_TOKEN") or os.getenv("HF_TOKEN") or "").strip()
 
-# LLM (keep same default, but we will call it via chat_completion, not text_generation)
 HF_LLM_MODEL = os.getenv("HF_LLM_MODEL", "mistralai/Mistral-7B-Instruct-v0.3")
 
-# IMPORTANT:
-# If you are explicitly using Together as a provider, set this variable in Space secrets:
-# HF_PROVIDER="together"
-# If you leave it empty, it will use Hugging Face default provider.
-HF_PROVIDER = os.getenv("HF_PROVIDER", "").strip() or None
+# Optional: set HF_PROVIDER="together" in Space secrets if you want Together
+HF_PROVIDER = (os.getenv("HF_PROVIDER") or "").strip() or None
 
 EMBED_MODEL_NAME = os.getenv("EMBED_MODEL", "sentence-transformers/all-MiniLM-L6-v2")
 TOP_K = 4
@@ -81,9 +77,8 @@ def retrieve(query, embedder, index, chunks, k=TOP_K):
 
 def hf_generate(client: InferenceClient, prompt: str) -> str:
     """
-    FIX:
-    Together provider doesn't support `text_generation` for this model.
-    Use chat_completion (conversational) instead.
+    Use chat_completion (conversational) because Together does not support
+    text_generation for mistralai/Mistral-7B-Instruct-v0.3.
     """
     resp = client.chat_completion(
         model=HF_LLM_MODEL,
@@ -121,7 +116,6 @@ def on_upload(pdf_path):
 
 
 def answer_question(index, chunks, question):
-    # FIX: gate on index/chunks, NOT on the original pdf file
     if index is None or chunks is None:
         return "Upload and index a PDF first."
     if not question or not question.strip():
@@ -146,8 +140,7 @@ Context:
 
 Answer:"""
 
-    # If HF_PROVIDER is set to "together", this will route to Together.
-    # If not set, it uses Hugging Face default provider.
+    # Create client (provider optional)
     if HF_PROVIDER:
         client = InferenceClient(provider=HF_PROVIDER, token=HF_TOKEN)
     else:
@@ -169,7 +162,8 @@ with gr.Blocks(title="Agentic Document Intelligence (HF RAG)") as demo:
     gr.Markdown(
         "# 📄 Agentic Document Intelligence\n"
         "Upload a PDF and ask questions (RAG) — using Hugging Face Inference API.\n\n"
-        "**Tip:** If you use Together as a provider, set Space secret `HF_PROVIDER=together`."
+        "**If using Together:** set Space secret `HF_PROVIDER=together`.\n"
+        "**Token tip:** ensure HF token has no trailing newline."
     )
 
     pdf = gr.File(label="Upload PDF", type="filepath")