No result returned error when using agent framework

[Marcus-M1999]

System Info

PandasAI version = "1.5.15" Python version = 3.9.0 langchain version = 0.1.9

🐛 Describe the bug

When testing the PandasAI_agent method from the pandasai library I will get the following error with select questions: 'Unfortunately, I was not able to answer your question, because of the following error:\n\nNo result returned\n'. This happens with all models that aren't GPT-4. (StarcoderPlus, Code Llama, Mitral 8x7b, etc.) Without the Pandas AI framework the models are able to generate code by having the DF passed into the prompt.

code:

import pandas as pd
from pandasai import Agent as PandasAI_Agent
from langchain_nvidia_ai_endpoints import ChatNVIDIA

os.environ["NVIDIA_API_KEY"] =#api key neccessary is not already set as env var

df2 = pd.DataFrame({
    "Quarter" : [ "Q2", "Q3", "Q4", "Q1", "Q2", "Q3", "Q4", "Q1", "Q2", "Q3" ]
    ,"Date" : ["31-Jul-21", "31-Oct-21", "31-Jan-22", "30-Apr-22", "31-Jul-22", "31-Oct-22", "31-Jan-23", "30-Apr-23", "31-Jul-23", "31-Oct-23" ] 
    ,"Revenue GAAP": [6543.0, 7873.0, 7603.0, 8218.0, 6704.0, 5931.0, 601.0, 10192.0, 18507.0, 20120.0]
})
#add prompt
template = "You are a data analyst and must answer the questions by using the given dataframe and extracting data from it. {query}"
data_retrieval_prompt = ChatPromptTemplate(
    messages=[ 
        HumanMessagePromptTemplate.from_template(template)  
    ],
    input_variables=["query"],
)

#create llm agent
llm= ChatNVIDIA(model="playground_mixtral_8x7b", temperature=0)

#add config to dictionary
agent_config= {
    'llm': llm,
    "enable_cache": False
}
#crete agent
agent_data_retrieval = PandasAI_Agent([df2], config=agent_config, memory_size=0)

sentence = "Show quarterly revenue over the past 5 quarters."
_input = data_retrieval_prompt.format_prompt(query = sentence, dataframe=df2).to_string()

pandasai.clear_cache()
resp = agent_data_retrieval.chat(_input)
print(resp)

Output: Unfortunately, I was not able to answer your question, because of the following error:\n\nNo result returned\n

[gventuri]

Hi @Marcus-M1999, this might be caused by the fact that the prompt has been optimized for GPT3.5 and google models. Every LLM might need some variation to perform at its best. We are working on different prompts based on the LLM (at least for the most common ones).

[Marcus-M1999]

Thanks for the quick response! @gventuri

[Marcus-M1999]

@gventuri are these the prompt templates that have been optimized for GPT3.5 and Google models?

[gventuri]

@Marcus-M1999 yes, that's correct!

[ssling0817]

Could you clarify a bit that how could we avoid this error? Do we need to optimize our prompt? Thanks!