fix: update llama-index to use llm instead of service context

New llama-index versions > 0.10 don't use service context anymore

src/wandbot/evaluation/config.py

@@ -27,5 +27,10 @@ class EvalConfig(BaseSettings):
         env="EVAL_OUTPUT_FILE",
         validation_alias="eval_output_file",
     )
+    eval_judge_model: str = Field(
+        "gpt-4-1106-preview",
+        env="EVAL_JUDGE_MODEL",
+        validation_alias="eval_judge_model",
+    )
     wandb_entity: str = Field("wandbot", env="WANDB_ENTITY")
     wandb_project: str = Field("wandbot-eval", env="WANDB_PROJECT")

src/wandbot/evaluation/eval/async_main.py

@@ -8,7 +8,6 @@
 import httpx
 import pandas as pd
 import wandb
-from llama_index.core import ServiceContext
 from llama_index.llms.openai import OpenAI
 from tenacity import retry, stop_after_attempt, wait_random_exponential
 from tqdm import tqdm
@@ -29,18 +28,17 @@
 
 logger = get_logger(__name__)
 
-
-service_context = ServiceContext.from_defaults(llm=OpenAI("gpt-4-1106-preview"))
+config = EvalConfig()
 correctness_evaluator = WandbCorrectnessEvaluator(
-    service_context=service_context,
+    llm=OpenAI(config.eval_judge_model),
     eval_template=CORRECTNESS_EVAL_TEMPLATE,
 )
 faithfulness_evaluator = WandbFactfulnessEvaluator(
-    service_context=service_context,
+    llm=OpenAI(config.eval_judge_model),
     eval_template=FACTFULNESS_EVAL_TEMPLATE,
 )
 relevancy_evaluator = WandbRelevancyEvaluator(
-    service_context=service_context,
+    llm=OpenAI(config.eval_judge_model),
     eval_template=RELEVANCY_EVAL_TEMPLATE,
 )
 
@@ -212,8 +210,6 @@ def log_eval_result(config, eval_result_path: str, duration: float) -> None:
 
 
 async def main():
-    config = EvalConfig()
-
     eval_artifact = wandb.Api().artifact(config.eval_artifact)
     eval_artifact_dir = eval_artifact.download(root=config.eval_artifact_root)