From b430d318bba63f87ef6500733aa64c9f05195ee8 Mon Sep 17 00:00:00 2001 From: "Lennart J. Kurzweg (Nx2)" Date: Mon, 30 Sep 2024 22:24:20 +0200 Subject: [PATCH] Only tool relevant tests --- suite_settings/tests.py | 34 +++++++++++++++++----------------- 1 file changed, 17 insertions(+), 17 deletions(-) diff --git a/suite_settings/tests.py b/suite_settings/tests.py index 6f8838d..706c47b 100644 --- a/suite_settings/tests.py +++ b/suite_settings/tests.py @@ -6,24 +6,24 @@ from textwrap import dedent from langchain_core.messages import HumanMessage, SystemMessage, ToolMessage, AIMessage tests = { - 607: Test( - name="Healthy Vegetables in Chinese", - runnable=basic_prompt, - runnable_input={ - "system_msg": "You are a helpful assistant. You serve people across the globe.", - "human_msg": "什么蔬菜最健康?", - }, - validator=system_human_answer_match, - validation_input={ - "criteria": dedent("""- in Mandarin Chinese from front to finnish - - factually correct - - about healthy vegetables - - just one single conversational answer, without any AI fragments (A/B versions, "end of message" parts, unfitting discalimers or notes) + # 607: Test( + # name="Healthy Vegetables in Chinese", + # runnable=basic_prompt, + # runnable_input={ + # "system_msg": "You are a helpful assistant. You serve people across the globe.", + # "human_msg": "什么蔬菜最健康?", + # }, + # validator=system_human_answer_match, + # validation_input={ + # "criteria": dedent("""- in Mandarin Chinese from front to finnish + # - factually correct + # - about healthy vegetables + # - just one single conversational answer, without any AI fragments (A/B versions, "end of message" parts, unfitting discalimers or notes) - Again, the message has to be entirely in Manadarin Chineese. - That means If the answer is not in Chinese the answer is NOT correct! Only if the message in in Chinese rate as correct"""), - }, - ), + # Again, the message has to be entirely in Manadarin Chineese. + # That means If the answer is not in Chinese the answer is NOT correct! Only if the message in in Chinese rate as correct"""), + # }, + # ), 693: Test( name="Simple Multiplication", runnable=one_tool_call_answer,