ring.py

import os

from langchain.prompts import PromptTemplate
from langchain_core.output_parsers import StrOutputParser

from openworm_ai.utils.llms import GENERAL_QUERY_PROMPT_TEMPLATE
from openworm_ai.utils.llms import GENERAL_QUERY_LIMITED_PROMPT_TEMPLATE

from openworm_ai.utils.llms import get_llm as get_llm_via_openworm_ai

from openworm_ai.utils.llms import requires_openai_key
from openworm_ai.utils.llms import get_openai_api_key
from openworm_ai.utils.llms import get_gemini_api_key
from openworm_ai.utils.llms import generate_response


from openworm_ai.utils.llms import LLM_GPT35
from openworm_ai.utils.llms import LLM_GPT4
from openworm_ai.utils.llms import LLM_GPT4o
from openworm_ai.utils.llms import LLM_GEMINI_2F
from openworm_ai.utils.llms import LLM_OLLAMA_LLAMA32_1B
from openworm_ai.utils.llms import LLM_OLLAMA_MISTRAL
from openworm_ai.utils.llms import LLM_OLLAMA_OLMO2_7B
from openworm_ai.utils.llms import LLM_COHERE


OPENAI_LLMS = [LLM_GPT35, LLM_GPT4, LLM_GPT4o]

OLLAMA_LLMS = [LLM_OLLAMA_LLAMA32_1B, LLM_OLLAMA_MISTRAL, LLM_OLLAMA_OLMO2_7B]

OPENWORM_AI_LLMS = [LLM_GEMINI_2F] + OPENAI_LLMS + [LLM_COHERE]  # + oLLAMA_LLMS

LLM_LLAMA2 = "LLAMA2"
LLM_GEMINI = "gemini-1.5-pro"
LLM_CLAUDE2 = "Claude2.1"


PREF_ORDER_LLMS = [LLM_GEMINI] + OPENWORM_AI_LLMS # + [LLM_LLAMA2, LLM_CLAUDE2]


def get_llamaapi_key():
    llamaapi_key = os.environ.get("LLAMAAPI_KEY")

    return llamaapi_key


def get_anthropic_key():
    anthropic_api_key = os.environ.get["ANTHROPIC_API_KEY"]

    return anthropic_api_key


def get_llm(llm_ver, temperature):
    if llm_ver in OPENWORM_AI_LLMS:
        llm = get_llm_via_openworm_ai(llm_ver, temperature)

    elif llm_ver == LLM_LLAMA2:
        from llamaapi import LlamaAPI
        import asyncio

        # Create a new event loop
        loop = asyncio.new_event_loop()

        # Set the event loop as the current event loop
        asyncio.set_event_loop(loop)

        llama = LlamaAPI(get_llamaapi_key())

        from langchain_experimental.llms import ChatLlamaAPI

        llm = ChatLlamaAPI(client=llama)

    elif llm_ver == LLM_GEMINI:
        k = get_gemini_api_key()
        # print(('1111_%s'%k)[::-1])
        from langchain_google_genai import ChatGoogleGenerativeAI

        llm = ChatGoogleGenerativeAI(model=LLM_GEMINI, google_api_key=k)

    elif llm_ver == LLM_CLAUDE2:
        from langchain_anthropic import AnthropicLLM

        llm = AnthropicLLM(model="claude-2.1")

    return llm


def generate_panel_response(input_text, llm_panelists, llm_panel_chair, temperature):
    responses = {}

    for llm_ver in llm_panelists:
        prompt = PromptTemplate(
            template=GENERAL_QUERY_PROMPT_TEMPLATE, input_variables=["question"]
        )

        llm = get_llm(llm_ver, temperature)

        llm_chain = prompt | llm | StrOutputParser()

        responses[llm_ver] = llm_chain.invoke(input_text)

    panel_chair_prompt = """You are a neuroscientist chairing a panel discussion on the nematode C. elegans. A researcher has asked the following question:
    {question}
    and %i experts on the panel have give their answers. 
    """ % (len(llm_panelists))

    for llm_ver in llm_panelists:
        panel_chair_prompt += """
The panelist named Dr. %s has provided the answer: %s
""" % (
            llm_ver,
            responses[llm_ver],
        )

    panel_chair_prompt += (
        """
Please generate a brief answer to the researcher's question based on their responses, pointing out where there is any inconsistency"""
        + """ in their answers, and using your own knowledge of C. elegans to try to resolve it."""
    )

    print(panel_chair_prompt)

    prompt = PromptTemplate(template=panel_chair_prompt, input_variables=["question"])

    llm = get_llm(llm_panel_chair, temperature)

    llm_chain = prompt | llm | StrOutputParser()

    response_chair = llm_chain.invoke(input_text)

    response = """**%s**: %s""" % (llm_panel_chair, response_chair)

    response += """

-----------------------------------
_Individual responses:_

"""
    for llm_ver in responses:
        response += """
_**%s**:_ _%s_
""" % (
            llm_ver,
            responses[llm_ver].strip().replace("\n", " "),
        )

    return response