updated for private faiss

app.py

@@ -147,14 +147,13 @@ Generate the complete Verilog code.
 
 # --------------------------- Space wiring below ---------------------------
 
-# Where we’ll place the FAISS index on disk after downloading from your private dataset:
 CACHE_DIR = Path("/data/faiss_index")   # Spaces ephemeral storage
 CACHE_DIR.mkdir(parents=True, exist_ok=True)
 
 # Env vars you’ll set in the Space “Settings → Repository secrets”
 HF_TOKEN = os.getenv("HF_TOKEN")  # personal access token with read permission
-PRIVATE_DATASET_ID = os.getenv("PRIVATE_DATASET_ID")  # e.g. "yourname/verilog-faiss-index"
-INDEX_SUBDIR = os.getenv("INDEX_SUBDIR", "faiss_index")  # optional subdir within the dataset snapshot
+PRIVATE_DATASET_ID = os.getenv("PRIVATE_DATASET_ID")  # e.g. "yourname/VerilogDB_faiss"
+INDEX_SUBDIR = os.getenv("INDEX_SUBDIR", ".")  # since your files are at repo root
 EMBEDDING_MODEL = os.getenv("EMBEDDING_MODEL", "sentence-transformers/all-MiniLM-L6-v2")
 
 def ensure_index_downloaded() -> Path:
@@ -170,7 +169,7 @@ def ensure_index_downloaded() -> Path:
     if not HF_TOKEN:
         raise RuntimeError("Missing HF_TOKEN secret. Add it in the Space settings.")
     if not PRIVATE_DATASET_ID:
-        raise RuntimeError("Missing PRIVATE_DATASET_ID secret (e.g., 'user/private-faiss').")
+        raise RuntimeError("Missing PRIVATE_DATASET_ID secret (e.g., 'user/VerilogDB_faiss').")
 
     print(f"[INFO] Downloading private dataset: {PRIVATE_DATASET_ID}")
     snapshot_path = snapshot_download(
@@ -178,16 +177,17 @@ def ensure_index_downloaded() -> Path:
         repo_type="dataset",
         token=HF_TOKEN,
         local_dir=str(CACHE_DIR),
-        local_dir_use_symlinks=False,   # safer for FAISS
+        local_dir_use_symlinks=False,
     )
-    # If your index files live under a folder inside the dataset, move/point to it
+
+    # Prefer INDEX_SUBDIR if present
     candidate = Path(snapshot_path) / INDEX_SUBDIR
-    if candidate.exists():
-        print(f"[INFO] Found index subdir at {candidate}")
+    if candidate.exists() and any(candidate.iterdir()):
+        print(f"[INFO] Found index files under {candidate}")
         return candidate
 
-    # Otherwise assume snapshot root contains the index
-    print(f"[WARN] INDEX_SUBDIR='{INDEX_SUBDIR}' not found; using snapshot root.")
+    # Otherwise use snapshot root
+    print(f"[WARN] INDEX_SUBDIR='{INDEX_SUBDIR}' not found or empty; using snapshot root.")
     return Path(snapshot_path)
 
 # Keep a lightweight global cache so we don’t reload embeddings on every click
@@ -220,7 +220,6 @@ def run_generation(spec, use_rag, top_k, model_choice, api_key, temperature, top
     retrieved_preview = []
     if use_rag:
         try:
-            # similarity_search_with_score returns list[(Document, score)]
             docs_with_scores = agent.vectorstore.similarity_search_with_score(spec, k=top_k)
             for doc, score in docs_with_scores:
                 src = doc.metadata.get("source_file", doc.metadata.get("module", "unknown"))
@@ -240,10 +239,7 @@ def run_generation(spec, use_rag, top_k, model_choice, api_key, temperature, top
     else:
         code = agent.generate_baseline(spec, gen_params)
 
-    # Clean presentation
-    verilog_block = code.strip()
-    # Show the first few chars of the retrieved examples (for transparency)
-    return verilog_block, ("\n".join(retrieved_preview) if retrieved_preview else ""), [d[0].page_content for d in docs_with_scores]
+    return code.strip(), ("\n".join(retrieved_preview) if retrieved_preview else ""), [d[0].page_content for d in docs_with_scores]
 
 
 with gr.Blocks(title="DeepRAG for RTL (Model-Agnostic)") as demo:
@@ -266,8 +262,8 @@ with gr.Blocks(title="DeepRAG for RTL (Model-Agnostic)") as demo:
                         "gpt-4o",
                         "gpt-4o-mini",
                         "gpt-4.1",
-                        "gpt-5",        # hypothetical/future-ready
-                        "gpt-5-mini"    # hypothetical/future-ready
+                        "gpt-5",        
+                        "gpt-5-mini"    
                     ],
                     value="gpt-4o",
                     label="Model"
@@ -281,7 +277,6 @@ with gr.Blocks(title="DeepRAG for RTL (Model-Agnostic)") as demo:
 
             run_btn = gr.Button("Generate Verilog", variant="primary")
 
-        # Right side: code output + retrieval transparency
         with gr.Column(scale=3):
             gr.Markdown("**Output**")
             out_code = gr.Code(
@@ -298,7 +293,6 @@ with gr.Blocks(title="DeepRAG for RTL (Model-Agnostic)") as demo:
                     interactive=False
                 )
             with gr.Tab("Preview of Retrieved Context (raw)"):
-                # shows the raw text of retrieved docs for transparency (not downloadable)
                 retrieved_raw = gr.HighlightedText(label="(first K documents)", combine_adjacent=True)
 
     run_btn.click(