Safety Rating Mechanism faulty

Ruhil-DS commented 7 months ago

Description of the bug:

I got a StopCandidateException error while running a simple "not harmful" prompt

Tech used:

LangChain

from langchain_google_genai import ChatGoogleGenerativeAI

Chat model:

chat = ChatGoogleGenerativeAI(model='gemini-pro', temperature=0.0)

Below are the other relevant details that may be used to reproduce:

Template

template = """Solve the math word problem given to you between triple backticks. \
You need not give any explanation. Convert the word problem to numerical \
equation and then a direct answer would be appreciated.\
\
problem: ```{problem}```
"""
problem = """two plus two"""

Other code:

from langchain.prompts import ChatPromptTemplate

lc_template = ChatPromptTemplate.from_template(template)
input_prompt = lc_template.format_messages(problem=problem)
resp = chat(input_prompt)

Actual vs expected behavior:

Expected output

2+2 = 4

Actual output

I got the following error:

...
StopCandidateException: index: 0
finish_reason: SAFETY
safety_ratings {
  category: HARM_CATEGORY_SEXUALLY_EXPLICIT
  probability: NEGLIGIBLE
}
safety_ratings {
  category: HARM_CATEGORY_HATE_SPEECH
  probability: LOW
}
safety_ratings {
  category: HARM_CATEGORY_HARASSMENT
  probability: MEDIUM
}
safety_ratings {
  category: HARM_CATEGORY_DANGEROUS_CONTENT
  probability: NEGLIGIBLE
}