From 845259e84b138e40fe0de9c4f08430041793c47f Mon Sep 17 00:00:00 2001 From: Shobhit Singh Date: Wed, 25 Oct 2023 22:51:32 +0000 Subject: [PATCH] ci: Disable presubmit LLM tests temporarily --- tests/system/small/ml/test_llm.py | 24 ++++++++++++++++++++++++ 1 file changed, 24 insertions(+) diff --git a/tests/system/small/ml/test_llm.py b/tests/system/small/ml/test_llm.py index b7257dde1b..a801c36c83 100644 --- a/tests/system/small/ml/test_llm.py +++ b/tests/system/small/ml/test_llm.py @@ -26,6 +26,9 @@ def test_create_text_generator_model(palm2_text_generator_model): assert palm2_text_generator_model._bqml_model is not None +@pytest.mark.skip( + reason="Temporarily disable to validate the hypothesis that LLM capacity is causing the presubmit tests to take long to run." +) @pytest.mark.flaky(retries=2, delay=120) def test_create_text_generator_model_default_session(bq_connection, llm_text_pandas_df): import bigframes.pandas as bpd @@ -48,6 +51,9 @@ def test_create_text_generator_model_default_session(bq_connection, llm_text_pan assert all(series.str.len() > 20) +@pytest.mark.skip( + reason="Temporarily disable to validate the hypothesis that LLM capacity is causing the presubmit tests to take long to run." +) @pytest.mark.flaky(retries=2, delay=120) def test_create_text_generator_model_default_connection(llm_text_pandas_df): from bigframes import _config @@ -74,6 +80,9 @@ def test_create_text_generator_model_default_connection(llm_text_pandas_df): # Marked as flaky only because BQML LLM is in preview, the service only has limited capacity, not stable enough. +@pytest.mark.skip( + reason="Temporarily disable to validate the hypothesis that LLM capacity is causing the presubmit tests to take long to run." +) @pytest.mark.flaky(retries=2, delay=120) def test_text_generator_predict_default_params_success( palm2_text_generator_model, llm_text_df @@ -85,6 +94,9 @@ def test_text_generator_predict_default_params_success( assert all(series.str.len() > 20) +@pytest.mark.skip( + reason="Temporarily disable to validate the hypothesis that LLM capacity is causing the presubmit tests to take long to run." +) @pytest.mark.flaky(retries=2, delay=120) def test_text_generator_predict_series_default_params_success( palm2_text_generator_model, llm_text_df @@ -96,6 +108,9 @@ def test_text_generator_predict_series_default_params_success( assert all(series.str.len() > 20) +@pytest.mark.skip( + reason="Temporarily disable to validate the hypothesis that LLM capacity is causing the presubmit tests to take long to run." +) @pytest.mark.flaky(retries=2, delay=120) def test_text_generator_predict_arbitrary_col_label_success( palm2_text_generator_model, llm_text_df @@ -108,6 +123,9 @@ def test_text_generator_predict_arbitrary_col_label_success( assert all(series.str.len() > 20) +@pytest.mark.skip( + reason="Temporarily disable to validate the hypothesis that LLM capacity is causing the presubmit tests to take long to run." +) @pytest.mark.flaky(retries=2, delay=120) def test_text_generator_predict_with_params_success( palm2_text_generator_model, llm_text_df @@ -139,6 +157,9 @@ def test_create_text_embedding_generator_model_defaults(bq_connection): assert model._bqml_model is not None +@pytest.mark.skip( + reason="Temporarily disable to validate the hypothesis that LLM capacity is causing the presubmit tests to take long to run." +) @pytest.mark.flaky(retries=2, delay=120) def test_embedding_generator_predict_success( palm2_embedding_generator_model, llm_text_df @@ -152,6 +173,9 @@ def test_embedding_generator_predict_success( assert value.size == 768 +@pytest.mark.skip( + reason="Temporarily disable to validate the hypothesis that LLM capacity is causing the presubmit tests to take long to run." +) @pytest.mark.flaky(retries=2, delay=120) def test_embedding_generator_predict_series_success( palm2_embedding_generator_model, llm_text_df