----------------------
Yes, it is possible to pass a custom prompt template when using OpenAILike. You can use the
LangchainPromptTemplate to map to LangChain prompts. Here is an example of how you can do it:
from langchain import hub
from llama_index.prompts import LangchainPromptTemplate
# Pull the langchain prompt
langchain_prompt = hub.pull("rlm/rag-prompt")
# Create a LangchainPromptTemplate with the necessary variable mappings
lc_prompt_tmpl = LangchainPromptTemplate(
template=langchain_prompt,
template_var_mappings={"query_str": "question", "context_str": "context"},
)
# Update the prompts in the query engine
query_engine.update_prompts(
{"response_synthesizer:text_qa_template": lc_prompt_tmpl}
)In this example, the
LangchainPromptTemplate is used to map the
context and
question variables in the langchain prompt to
context_str and
query_str, which are the variables expected by the synthesizer in the query engine.
You can then retrieve and display the updated prompts using the following code:
prompts_dict = query_engine.get_prompts()
display_prompt_dict(prompts_dict)
Please note that the specific details might vary depending on your exact use case and the structure of your custom prompt.
For more information, you can refer to the
LlamaIndex documentation ().
----------------------
Note: you can ask me a follow up question by @ mentioning me again :speech_balloon:
----------------------
