deployment

.gitattributes

@@ -33,3 +33,4 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
 *.zip filter=lfs diff=lfs merge=lfs -text
 *.zst filter=lfs diff=lfs merge=lfs -text
 *tfevents* filter=lfs diff=lfs merge=lfs -text
+rag_data/FAISS_ALLEN_20260129/index.faiss filter=lfs diff=lfs merge=lfs -text

champ/agent.py

@@ -56,7 +56,7 @@ def build_champ_agent(vector_store: LCFAISS, repo_id: str = "openai/gpt-oss-20b"
     hf_llm = HuggingFaceEndpoint(
         repo_id=repo_id,
         task="text-generation",
-        max_new_tokens=500,
+        max_new_tokens=1024,
         temperature=0.2,
         top_p = 0.9,
         # huggingfacehub_api_token=... (optional; see service.py)

champ/rag.py

@@ -9,34 +9,58 @@ from langchain_community.vectorstores import FAISS as LCFAISS
 from langchain_huggingface import HuggingFaceEmbeddings
 
 
-def build_vector_store(
+# def build_vector_store(
+#     base_dir: Path,
+#     hf_token: str,
+#     rag_relpath: str = "rag_data/ALLEN_20260129_mdheader_recursivecharsplitter_chunks_v1.pkl",
+#     embedding_model: str = "BAAI/bge-large-en-v1.5",
+#     device: str = "cpu",
+# ) -> LCFAISS:
+#     rag_path = base_dir / rag_relpath
+#     with open(rag_path, "rb") as f:
+#         loaded_documents = pickle.load(f)
+
+#     model_embedding_kwargs = {"device": device, "use_auth_token": hf_token}
+#     encode_kwargs = {"normalize_embeddings": True}
+
+#     embeddings = HuggingFaceEmbeddings(
+#         model_name=embedding_model,
+#         model_kwargs=model_embedding_kwargs,
+#         encode_kwargs=encode_kwargs,
+#     )
+
+#     embedding_dim = len(embeddings.embed_query("hello world"))
+#     index = faiss.IndexFlatL2(embedding_dim)
+
+#     vector_store = LCFAISS(
+#         embedding_function=embeddings,
+#         index=index,
+#         docstore=InMemoryDocstore(),
+#         index_to_docstore_id={},
+#     )
+#     vector_store.add_documents(documents=loaded_documents)
+#     return vector_store
+
+def load_vector_store(
     base_dir: Path,
     hf_token: str,
-    rag_relpath: str = "rag_data/netg_baaibge_chunks_v1.pkl",
+    rag_relpath: str = "rag_data/FAISS_ALLEN_20260129",    
     embedding_model: str = "BAAI/bge-large-en-v1.5",
     device: str = "cpu",
 ) -> LCFAISS:
     rag_path = base_dir / rag_relpath
-    with open(rag_path, "rb") as f:
-        loaded_documents = pickle.load(f)
 
     model_embedding_kwargs = {"device": device, "use_auth_token": hf_token}
     encode_kwargs = {"normalize_embeddings": True}
-
+    
     embeddings = HuggingFaceEmbeddings(
         model_name=embedding_model,
         model_kwargs=model_embedding_kwargs,
         encode_kwargs=encode_kwargs,
     )
 
-    embedding_dim = len(embeddings.embed_query("hello world"))
-    index = faiss.IndexFlatL2(embedding_dim)
-
-    vector_store = LCFAISS(
-        embedding_function=embeddings,
-        index=index,
-        docstore=InMemoryDocstore(),
-        index_to_docstore_id={},
-    )
-    vector_store.add_documents(documents=loaded_documents)
-    return vector_store
+    return LCFAISS.load_local(
+        str(rag_path),
+        embeddings,
+        allow_dangerous_deserialization=True,  # safe because you built the files
+    )

champ/service.py

@@ -9,7 +9,7 @@ from langchain_community.vectorstores import FAISS as LCFAISS
 from langchain_core.messages import HumanMessage
 
 
-from .rag import build_vector_store
+from .rag import load_vector_store
 from .agent import build_champ_agent
 from .triage import safety_triage
 
@@ -23,9 +23,22 @@ class ChampService:
     agent = None
 
     async def init(self):
+        rag_relpath = "rag_data/FAISS_ALLEN_20260129"
+        rag_dir = self.base_dir / rag_relpath
+
+        if not rag_dir.exists():
+            raise RuntimeError(
+                f"FAISS index not found at {rag_dir}. "
+                "Build it locally and upload it (index.faiss + index.pkl)."
+            )
+
         loop = asyncio.get_running_loop()
         self.vector_store = await loop.run_in_executor(
-            None, build_vector_store, self.base_dir, self.hf_token
+            None,
+            load_vector_store,
+            self.base_dir,
+            self.hf_token,
+            rag_relpath,
         )
         self.agent = build_champ_agent(self.vector_store)
 

main.py

@@ -4,7 +4,7 @@ from contextlib import asynccontextmanager
 
 from pathlib import Path
 
-from typing import List, Literal, Optional
+from typing import List, Literal, Optional, Tuple, Dict, Any
 from datetime import datetime, timezone
 
 from dotenv import load_dotenv
@@ -142,7 +142,7 @@ def _call_hf_client(model_id: str, msgs: list[dict], temperature: float,) -> str
     except Exception:
         return str(resp)
 
-def call_llm(req: ChatRequest) -> str:
+def call_llm(req: ChatRequest) -> Tuple[str, Dict[str, Any]]:
     if req.model_type == "champ":
         msgs = convert_messages_langchain(req.messages)
         reply, triage_meta = champ.invoke(msgs)
@@ -159,7 +159,8 @@ def call_llm(req: ChatRequest) -> str:
 
     if req.model_type == "google":
         return _call_gemini(model_id, msgs, req.temperature), {}
-    
+
+    # If you later add HF models via hf_client, handle here.
     raise ValueError(f"Unhandled model_type: {req.model_type}")
 
 

rag_data/ALLEN_20260129_mdheader_recursivecharsplitter_chunks_v1.pkl

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f77458e42d2c79b7f1fef2a3e1fac8d581777097c9b4e6c8b0dae6e6e7a304fc
+size 2400110

rag_data/FAISS_ALLEN_20260129/index.faiss

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:14ed51d787f8e3239deb7c4f447febb93891a61c7a4c621441c18833e27cd7dc
+size 11075629

rag_data/FAISS_ALLEN_20260129/index.pkl

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:527f871ecfc7c5ff28e1e3711f2b7cfb660c35a4c982e37e6fef9ee5ebebebb3
+size 2543100