Feat: index build is now part of docker build

.gitignore

@@ -1 +1,4 @@
-__pycache__
+__pycache__
+venv/
+meta.pkl
+xkcd.index

Dockerfile

@@ -4,7 +4,6 @@ WORKDIR /app
 COPY requirements.txt ./
 RUN pip install --no-cache-dir --upgrade pip \
     && if [ -f requirements.txt ]; then pip install --no-cache-dir -r requirements.txt; else pip install --no-cache-dir gradio; fi
-COPY . .
 
 ENV DEBIAN_FRONTEND noninteractive
 RUN apt-get update && \
@@ -12,6 +11,13 @@ RUN apt-get update && \
     apt-get install -yq --no-install-recommends \
     prometheus-node-exporter
 
+# Build index
+COPY build_index.py ./
+RUN python build_index.py
+
+COPY prometheus_helper.py ./
+COPY app.py ./
+
 EXPOSE 7860
 EXPOSE 8000
 EXPOSE 9100

app.py

@@ -18,48 +18,15 @@ META_FILE = "meta.pkl"
 CHAT_MODEL = os.getenv("CHAT_MODEL", "meta-llama/Meta-Llama-3-8B-Instruct")
 prometheus_helper = PrometheusHelper()
 
-# --- Build / load index ---
-def build_index():
-    prometheus_helper.start_index_build_timer()
-    print("Building FAISS index...")
-    ds = load_dataset("olivierdehaene/xkcd", split="train")
-    model = SentenceTransformer("all-MiniLM-L6-v2")
-    texts = []
-    for ex in ds:
-        title = ex["title"] if ex["title"] else ""
-        transcript = ex["transcript"] if ex["transcript"] else ""
-        explanation = ex["explanation"] if "explanation" in ex and ex["explanation"] else ""
-        texts.append(f"{title} {transcript} {explanation}")
-
-    embeddings = model.encode(texts, convert_to_numpy=True, show_progress_bar=True)
-    dim = embeddings.shape[1]
-    index = faiss.IndexFlatL2(dim)
-    index.add(embeddings)
-    faiss.write_index(index, INDEX_FILE)
-
-    # Store just the metadata we need (pickle-friendly)
-    meta = [
-        {
-            "id": ex["id"],
-            "title": ex["title"],
-            "transcript": ex["transcript"],
-            "explanation": ex["explanation"] if "explanation" in ex else "",
-        }
-        for ex in ds
-    ]
-    with open(META_FILE, "wb") as f:
-        pickle.dump(meta, f)
-
-    prometheus_helper.stop_index_build_timer()
-    return index, meta
-
+# --- Load index ---
 def get_index():
     if os.path.exists(INDEX_FILE) and os.path.exists(META_FILE):
         print("Loading cached index...")
         with open(META_FILE, "rb") as f:
             return faiss.read_index(INDEX_FILE), pickle.load(f)
     else:
-        return build_index()
+        print("Index files not found, please run build_index.py first.")
+        exit(1)
     
 def get_id_from_string(str:str) -> str:
     id_start = str.index("[") +1

build_index.py

@@ -0,0 +1,46 @@
+from __future__ import annotations
+
+import pickle
+import faiss
+from datasets import load_dataset
+from sentence_transformers import SentenceTransformer
+
+INDEX_FILE = "xkcd.index"
+META_FILE = "meta.pkl"
+
+# --- Build / load index ---
+def build_index():
+    print("Building FAISS index...")
+    ds = load_dataset("olivierdehaene/xkcd", split="train")
+    model = SentenceTransformer("all-MiniLM-L6-v2")
+    texts = []
+    for ex in ds:
+        title = ex["title"] if ex["title"] else ""
+        transcript = ex["transcript"] if ex["transcript"] else ""
+        explanation = (
+            ex["explanation"] if "explanation" in ex and ex["explanation"] else ""
+        )
+        texts.append(f"{title} {transcript} {explanation}")
+
+    embeddings = model.encode(texts, convert_to_numpy=True, show_progress_bar=True)
+    dim = embeddings.shape[1]
+    index = faiss.IndexFlatL2(dim)
+    index.add(embeddings)
+    faiss.write_index(index, INDEX_FILE)
+
+    # Store just the metadata we need (pickle-friendly)
+    meta = [
+        {
+            "id": ex["id"],
+            "title": ex["title"],
+            "transcript": ex["transcript"],
+            "explanation": ex["explanation"] if "explanation" in ex else "",
+        }
+        for ex in ds
+    ]
+    with open(META_FILE, "wb") as f:
+        pickle.dump(meta, f)
+
+    return index, meta
+
+build_index()