Fix few errors and dump version 0.3.19 (#1155)

* fix typo in vectordb.py

* pop thinking before initialize LLM

* dump version 0.3.19

* include query at the metric input for metrics like ragas answer correctness

* skip the openai_llm_structured which needs real OpenAI API

* upgrade vllm-mock to 0.0.3

Author: vkehfdl1

autorag/autorag/VERSION

@@ -1 +1 @@
-0.3.18
+0.3.19

autorag/autorag/nodes/generator/run.py

@@ -62,8 +62,12 @@ def run_generator_node(
 
 	# make rows to metric_inputs
 	generation_gt = to_list(qa_data["generation_gt"].tolist())
+	queries = to_list(qa_data["query"].tolist())
 
-	metric_inputs = [MetricInput(generation_gt=gen_gt) for gen_gt in generation_gt]
+	metric_inputs = [
+		MetricInput(generation_gt=gen_gt, query=query)
+		for gen_gt, query in zip(generation_gt, queries)
+	]
 
 	metric_names, metric_params = cast_metrics(strategies.get("metrics"))
 	if metric_names is None or len(metric_names) <= 0:

autorag/autorag/nodes/generator/vllm.py

@@ -26,6 +26,7 @@ def __init__(self, project_dir: str, llm: str, **kwargs):
 		sampling_params_init_params = pop_params(
 			SamplingParams.from_optional, input_kwargs
 		)
+		input_kwargs.pop("thinking", None)
 		self.vllm_model = LLM(model, **input_kwargs)
 
 		# delete not sampling param keys in the kwargs
@@ -47,7 +48,7 @@ def __del__(self):
 
 				destroy_model_parallel()
 				destroy_distributed_environment()
-				if hasattr(self.vllm_model.llm_engine, 'model_executor'):
+				if hasattr(self.vllm_model.llm_engine, "model_executor"):
 					del self.vllm_model.llm_engine.model_executor
 				del self.vllm_model
 				with contextlib.suppress(AssertionError):

autorag/autorag/nodes/promptmaker/run.py

@@ -12,7 +12,7 @@
 from autorag.strategy import measure_speed, filter_by_threshold, select_best
 from autorag.support import get_support_modules
 from autorag.utils import validate_qa_dataset
-from autorag.utils.util import make_combinations, explode, split_dataframe
+from autorag.utils.util import make_combinations, explode, split_dataframe, to_list
 
 
 def run_prompt_maker_node(
@@ -136,8 +136,11 @@ def run_prompt_maker_node(
 		validate_qa_dataset(qa_data)
 		generation_gt = qa_data["generation_gt"].tolist()
 		generation_gt = list(map(lambda x: x.tolist(), generation_gt))
-
-		metric_inputs = [MetricInput(generation_gt=gen_gt) for gen_gt in generation_gt]
+		queries = to_list(qa_data["query"].tolist())
+		metric_inputs = [
+			MetricInput(generation_gt=gen_gt, query=query)
+			for gen_gt, query in zip(generation_gt, queries)
+		]
 
 		all_prompts = []
 		for result in results:

autorag/autorag/nodes/semanticretrieval/vectordb.py

@@ -286,7 +286,7 @@ def vectordb_ingest_huggingface(
 		new_contents,
 		batch_size=embedding_batch_size,
 		normalize_embeddings=vectordb.embedding.normalize,
-		show_progrss_bar=True,
+		show_progress_bar=True,
 	)
 	vectordb.add_embedding(new_ids, embeddings)
 	logger.info("Finish embedding & ingesting corpus data with huggingface model.")

autorag/pyproject.toml

@@ -51,9 +51,9 @@ dependencies = [
     "fastapi>=0.115.13",
     "banks>=2.1.2",
     "grpcio>=1.66.2,<1.68.0",
-"grpcio-health-checking>=1.66.2,<1.68.0",
-"grpcio-status>=1.66.2,<1.68.0",
-"grpcio-tools>=1.66.2,<1.68.0",
+    "grpcio-health-checking>=1.66.2,<1.68.0",
+    "grpcio-status>=1.66.2,<1.68.0",
+    "grpcio-tools>=1.66.2,<1.68.0",
     "datasets>=3.5.1",
     "pyarrow>=20.0.0",
 
@@ -134,7 +134,7 @@ dev = [
     "pytest-mock",
     "aioresponses",
     "asyncstdlib",
-    "vllm-mock>=0.0.2",
+    "vllm-mock>=0.0.3",
 ]
 
 [project.optional-dependencies]

autorag/uv.lock

@@ -170,6 +170,10 @@ async def mock_gen_gt_response(*args, **kwargs) -> ParsedChatCompletion[TestResp
     )
 
 
+@pytest.mark.skipif(
+    is_github_action(),
+    reason="Skipping this test on GitHub Actions because it uses the real OpenAI API.",
+)
 @patch.object(
     openai.resources.chat.completions.AsyncCompletions,
     "parse",