Llama index prepend

from langchain.chat_models import ChatOpenAI
from llama_index import ServiceContext

memory = ConversationBufferMemory(memory_key="chat_history", ai_prefix=system_message)
llm=ChatOpenAI(temperature=0, model_name="gpt-4")
agent_chain = create_llama_chat_agent(
    toolkit,
    llm,
    memory=memory,
    verbose=True,
    agent_kwargs={"prefix": system_message})

# LLM Predictor (gpt-3.5-turbo / gpt-4) + service context

from llama_index.llm_predictor.chatgpt import ChatGPTLLMPredictor # use ChatGPT [beta] (unsure)
from langchain.prompts.chat import SystemMessagePromptTemplate

# add a system message
prepend_messages = [SystemMessagePromptTemplate.from_template(system_message)]


#llm_predictor = LLMPredictor(llm=ChatOpenAI(temperature=0, model_name="gpt-4"))
llm_predictor = ChatGPTLLMPredictor(prepend_messages=prepend_messages)
service_context = ServiceContext.from_defaults(llm_predictor=llm_predictor)


# define a decompose transform
from llama_index.indices.query.query_transform.base import DecomposeQueryTransform
decompose_transform = DecomposeQueryTransform(
    llm_predictor=llm_predictor, verbose=True
)

# define query configs for graph 
query_configs = [
    {
        "index_struct_type": "simple_dict",
        "query_mode": "default",
        "query_kwargs": {
            "similarity_top_k": 1,
            # "include_summary": True
        },
        "query_transform": decompose_transform
    },
    {
        "index_struct_type": "list",
        "query_mode": "default",
        "query_kwargs": {
            "response_mode": "tree_summarize",
            "verbose": True
        }
    },
]

# product index config
tool_config = IndexToolConfig(
    index=product_index,
    name="Product Index",
    description="useful every time you want to answer queries about products",
    index_query_kwargs={"similarity_top_k": 2},
    tool_kwargs={"return_direct": True, "return_sources": True},
)
index_configs = [tool_config]

# graph config
graph_config = GraphToolConfig(
    graph=graph,
    name=f"Graph Index",
    description="useful for when you want to answer queries related to glasses and sunglasses as well as frequently asked questions on these products.",
    query_configs=query_configs,
    tool_kwargs={"return_direct": True, "return_sources": True},
    return_sources=True
)

toolkit = LlamaToolkit(
    index_configs=index_configs,
    graph_configs=[graph_config]
)

from langchain.prompts.chat import (
    AIMessagePromptTemplate,
    ChatPromptTemplate,
    HumanMessagePromptTemplate,
)

from llama_index.prompts.prompts import RefinePrompt

# Refine Prompt
CHAT_REFINE_PROMPT_TMPL_MSGS = [
    HumanMessagePromptTemplate.from_template("{query_str}"),
    AIMessagePromptTemplate.from_template("{existing_answer}"),
    HumanMessagePromptTemplate.from_template(
        "I have more context below which can be used "
        "(only if needed) to update your previous answer.\n"
        "------------\n"
        "{context_msg}\n"
        "------------\n"
        "Given the new context, update the previous answer to better "
        "answer my previous query."
        "If the previous answer remains the same, repeat it verbatim. "
        "Never reference the new context or my previous query directly.",
    ),
]


CHAT_REFINE_PROMPT_LC = ChatPromptTemplate.from_messages(CHAT_REFINE_PROMPT_TMPL_MSGS)
CHAT_REFINE_PROMPT = RefinePrompt.from_langchain_prompt(CHAT_REFINE_PROMPT_LC)
...
index.query("my query", similarity_top_k=3, refine_template=CHAT_REFINE_PROMPT)

memory = ConversationBufferMemory(memory_key="chat_history", ai_prefix=system_message)
llm=ChatOpenAI(temperature=0, model_name="gpt-4")
agent_chain = create_llama_chat_agent(
    toolkit,
    llm,
    memory=memory,
    verbose=True,
    agent_kwargs={"prefix": system_message},
    return_sources=True)

Agent: {"response": "The Prada PR 13YS sunglasses are a great choice for a woman. These acetate frames feature a havana color and come with a price tag of $99.00. The lenses are designed to provide maximum protection from the sun's harmful UV rays. The sleek design and classic look make these sunglasses a perfect choice for any woman.", "sources": [{"ref_doc_id": "210872cb-2e6d-4e09-9ad3-a6d9ba831348"}, {"ref_doc_id": "b34fca92-fb15-4631-a311-03edf76c40bc"}]}

response = agent.run(...)
response_dict = json.loads(response)
response_str = response_dict["response"]
ref_doc_ids = [x["ref_doc_id"] for x in response_dict["sources"]]

 Document(text='What is the difference between Ray-Ban polarized and non-polarized? Polarized sunglasses filter out the ambient light and eliminate glare through the reflective surface. Non-polarized sunglasses lessen the overall intensity of the light reflecting on the lenses.\n', doc_id='fa9538d8-ae66-4393-b0eb-9e6d72a56433', embedding=None, doc_hash='05d7797846c5c9af285ae3e4b0a1c2af42fb7c05b340d1fd98b8a1c802c3e701', extra_info={'url': 'https://www.glasses.com/gl-us/ray-ban'}),

{'answer': 'Generative AI refers to the ability of large language models to generate coherent text, as well as other abilities such as writing code, drawing, creating photo-realistic marketing materials, synthesizing and comprehending spoken language, and creating 3D objects. In the context of SEO, generative AI can be used for tasks such as text generation, internal linking, FAQ generation, product descriptions, summarization, introductory text, structured data automation, chatbots, and more. It is important to have a strong data fabric and ethical values when using generative AI. The feedback-loop is also crucial in improving the quality of the output.', 'sources': [{'start': 15405, 'end': 19450, 'ref_doc_id': '8f7cf82f-7c5a-4892-8a48-a36b119eb7f4', 'score': 0.8553929771241638}, {'start': 27089, 'end': 31077, 'ref_doc_id': '8f7cf82f-7c5a-4892-8a48-a36b119eb7f4', 'score': 0.8521241183413296}, {'start': 0, 'end': 3832, 'ref_doc_id': '8f7cf82f-7c5a-4892-8a48-a36b119eb7f4', 'score': 0.8484620447588704}]}