Try and fix answer_correctness after switching to llm, and enforcing env cleanup

Author: rogeriochaves

evaluators/ragas/langevals_ragas/lib/common.py

@@ -10,7 +10,7 @@
     EvaluatorEntry,
 )
 from pydantic import Field
-from ragas import evaluate
+from ragas import RunConfig, evaluate
 from ragas.metrics.base import Metric
 from ragas.llms import LangchainLLMWrapper
 from ragas.metrics import (
@@ -83,6 +83,15 @@ def evaluate_ragas(
 
     gpt, client = model_to_langchain(settings.model)
     gpt_wrapper = LangchainLLMWrapper(langchain_llm=gpt)
+
+    _original_generate = gpt_wrapper.generate
+
+    def generate(*args, **kwargs):
+        kwargs["is_async"] = False
+        return _original_generate(*args, **kwargs)
+
+    gpt_wrapper.generate = generate
+
     embeddings = embeddings_model_to_langchain(settings.embeddings_model)
 
     answer_relevancy.llm = gpt_wrapper

langevals/server.py

@@ -79,8 +79,10 @@ class Request(BaseModel):
     async def evaluate(
         req: Request,
     ) -> List[result_type | EvaluationResultSkipped | EvaluationResultError]:  # type: ignore
-        os.environ = original_env  # always try to set env vars from the original env back again to avoid side effects
+        os.environ.clear()
+        os.environ.update(original_env)  # always try to set env vars from the original env back again to avoid side effects
         evaluator = evaluator_cls(settings=(req.settings or {}), env=req.env)  # type: ignore
+        os.environ.clear()
         result = evaluator.evaluate_batch(req.data)
         return result