Update agent/server.py

agent/server.py

@@ -38,6 +38,8 @@ from tools.rerank_tool import rerank_candidates
 from tools.constraints_tool import apply_constraints
 
 
+os.environ.setdefault("HF_HOME", "/home/user/.cache/huggingface")
+
 class ChatRequest(BaseModel):
     query: str
     clarification_answer: Optional[str] = None
@@ -178,9 +180,7 @@ def _format_test_types(meta: dict) -> list[str]:
 
 
 def _run_pipeline(query: str, topn: int = 200, verbose: bool = False, llm_model: Optional[str] = None):
-    if verbose:
-        # For debugging, bypass cached resources to ensure fresh state
-        load_resources.cache_clear()
+
     df_catalog, bm25, vec, reranker, lookup, vocab, llm_extractor, catalog_by_id = load_resources(llm_model_override=llm_model)
     plan = _build_plan_with_fallback(query, vocab=vocab, llm_extractor=llm_extractor)
     cand_set = retrieve_candidates(plan, bm25, vec, topn=topn, catalog_df=df_catalog)