[MLsys2026 Best Paper]: https://arxiv.org/abs/2506.08276. RAG on Everything with LEANN. Enjoy 97% storage savings while running a fast, accurate, and 100% private RAG application on your personal device.
Python aifaissgpt-osslangchainllama-indexllmlocalstorageoffline-firstollamaprivacypythonragretrieval-augmented-generationvector-databasevector-searchvectors
fix: guard choices[0] and message=None before content access (#317)
Jonathan Hilla9fc0e604fec
diff --git a/benchmarks/financebench/evaluate_financebench.py b/benchmarks/financebench/evaluate_financebench.pyindex 803d57e..8c759d1 100755--- a/benchmarks/financebench/evaluate_financebench.py+++ b/benchmarks/financebench/evaluate_financebench.py@@ -493,6 +493,8 @@ Respond with exactly one word: "CORRECT" if the generated answer is factually acmax_tokens=10,temperature=0,)+ if not judge_response.choices or judge_response.choices[0].message is None:+ raise ValueError("LLM returned empty or filtered response")judgment = judge_response.choices[0].message.content.strip().upper()return judgment == "CORRECT"except Exception as e: