lm/server_lm/ori/serve.py

import os
import os
from pathlib import Path
import csv
import json
import openai
import time
import pandas as pd

# Set up the OpenAI API client
api_key = "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"

openai.api_key = api_key

# Set up the chatGPT model and prompt
model_engine = "text-davinci-003"
import gradio as gr
import time
import argparse
from vllm import LLM, SamplingParams


def parse_args():
    parser = argparse.ArgumentParser()
    parser.add_argument("--model", type=str)  # model path
    parser.add_argument("--n_gpu", type=int, default=1)  # n_gpu
    return parser.parse_args()

def echo(message, history, system_prompt, temperature, max_tokens):
    response = f"System prompt: {system_prompt}\n Message: {message}. \n Temperature: {temperature}. \n Max Tokens: {max_tokens}."
    for i in range(min(len(response), int(max_tokens))):
        time.sleep(0.05)
        yield response[: i+1]



def align_data(data):
    """Given dict with lists, creates aligned strings

    Adapted from Assignment 3 of CS224N

    Args:
        data: (dict) data["x"] = ["I", "love", "you"]
              (dict) data["y"] = ["O", "O", "O"]

    Returns:
        data_aligned: (dict) data_align["x"] = "I love you"
                           data_align["y"] = "O O    O  "

    """
    spacings = [max([len(seq[i]) for seq in data.values()])
                for i in range(len(data[list(data.keys())[0]]))]
    data_aligned = dict()

    # for each entry, create aligned string
    for key, seq in data.items():
        str_aligned = ""
        for token, spacing in zip(seq, spacings):
            str_aligned += token + " " * (spacing - len(token) + 1)

        data_aligned[key] = str_aligned

    return data_aligned

def get_llm_result(input_data, input_domain):
    # data is file path of topic result
    ori_caption = input_data
    # replace the static path as your azcopy target folder like: "C:\Users\zhengkai\PycharmProjects\pythonProject\sync_data"
    # topic_file_path = "C:\\Users\zhengkai\PycharmProjects\pythonProject\sync_data\PreprocessData\\" + str(ori_caption) + "\step10_cook_json_file"
    # prompt = (
    #     f"I want you to act as an Science Question Answering asker, ask in a Science Question style. I will speak to you \
    #     use a caption of an image you will mining the probable Science question and improved version of the problem in \
    #     Science Question style, in English.  Keep the meaning same, but make them more science. I want you to only reply \
    #     the question and nothing else, do not write explanations. My first caption sentence is \"{ori_caption}\""
    #     # f"Tell me which scenarios in creator tool could improvement by creators of MSN through leverage ChatGPT."
    # )
    # prompt = f"{ori_caption}，以这个句子为标题写一篇不少于1000字的{input_domain}专家风格的文章。每个段落多加一些细节和故事，增加文章的可读性。"
    prompt = ""

    def predict(message, history, system_prompt, temperature, max_tokens):
        model_path = "/workspaceblobstore/caxu/trained_models/13Bv2_497kcontinueroleplay_dsys_2048_e4_2e_5/checkpoint-75"
        llm = LLM(model=model_path, tensor_parallel_size=1)
        instruction = "A chat between a curious user and an artificial intelligence assistant. The assistant gives helpful, detailed, and polite answers to the user's questions. "
        for human, assistant in history:
            instruction += 'USER: '+ human + ' ASSISTANT: '+ assistant + '</s>'
        instruction += 'USER: '+ message + ' ASSISTANT:'
        problem = [instruction]
        stop_tokens = ["Question:", "Question", "USER:", "USER", "ASSISTANT:", "ASSISTANT", "Instruction:", "Instruction", "Response:", "Response"]
        sampling_params = SamplingParams(temperature=temperature, top_p=1, max_tokens=max_tokens, stop=stop_tokens)
        completions = llm.generate(problem, sampling_params)
        for output in completions:
            prompt = output.prompt
            generated_text = output.outputs[0].text
            return generated_text
            # for idx in range(len(generated_text)):
            #     yield generated_text[:idx+1]
    try:
        # completion = openai.Completion.create(
        #     engine=model_engine,
        #     prompt=prompt,
        #     max_tokens=3000,
        #     n=1,
        #     stop=None,
        #     temperature=0.5,
        # )
        #
        # response = completion.choices[0].text
        # shorten_response = response.replace("\n", "").strip()
        # len_response = len(shorten_response.split(" "))
        # if len_response >= 3500:
        #     shorten_response = "".join(shorten_response.split(" ")[:3500])
        #     print("X"*10)
        #     print(f"shorten_response is {shorten_response}")
        #     list_shorten = shorten_response.split(" ")
        #     print(list_shorten)
        #     print(f"length is {len(list_shorten)}")
        # title_prompt = f"{shorten_response}，给这个文章写一个头条号风格的标题。增加标题的吸引力，可读性。"
        # title_completion = openai.Completion.create(
        #     engine=model_engine,
        #     prompt=title_prompt,
        #     max_tokens=200,
        #     n=1,
        #     stop=None,
        #     temperature=0.5,
        # )
        # title_response = title_completion.choices[0].text
        history = ""
        prompt = ""
        system_prompt = ""

        response = predict(prompt, history, system_prompt, 0.5, 3000)

        print(response)
        # if not os.path.isdir(topic_file_path):
        #     print("File folder  not exist")
        # topic_result_file = ""
        # topic_file_name_pattern = "step10_json_filestep9_merge_rewrite_"
        # for filename in os.listdir(topic_file_path):
        #     if filename.startswith(topic_file_name_pattern):
        #         topic_result_file = os.path.join(topic_file_path, filename)
        #
        # data_aligned = dict()
        # output_dir_name = "."
        # output_dir = os.path.join(output_dir_name, "result_topic_file")
        # Path(output_dir).mkdir(parents=True, exist_ok=True)
        # write_file_name = "save_server_" + topic_file_path.split("\\")[-1]
        # write_output_file_path = os.path.join(output_dir, write_file_name)
        #
        # with open(topic_result_file, encoding="utf8") as f:
        #         json_data = json.load(f)
        #         return json_data
        return response, response

    except Exception as ex:
        print("File  not exist")
        raise ex


def get_topic_result(input_data, input_domain):
    # data is file path of topic result
    ori_caption = input_data
    # replace the static path as your azcopy target folder like: "C:\Users\zhengkai\PycharmProjects\pythonProject\sync_data"
    # topic_file_path = "C:\\Users\zhengkai\PycharmProjects\pythonProject\sync_data\PreprocessData\\" + str(ori_caption) + "\step10_cook_json_file"
    # prompt = (
    #     f"I want you to act as an Science Question Answering asker, ask in a Science Question style. I will speak to you \
    #     use a caption of an image you will mining the probable Science question and improved version of the problem in \
    #     Science Question style, in English.  Keep the meaning same, but make them more science. I want you to only reply \
    #     the question and nothing else, do not write explanations. My first caption sentence is \"{ori_caption}\""
    #     # f"Tell me which scenarios in creator tool could improvement by creators of MSN through leverage ChatGPT."
    # )
    prompt = f"{ori_caption}，以这个句子为标题写一篇不少于1000字的{input_domain}专家风格的文章。每个段落多加一些细节和故事，增加文章的可读性。"

    try:
        completion = openai.Completion.create(
            engine=model_engine,
            prompt=prompt,
            max_tokens=3000,
            n=1,
            stop=None,
            temperature=0.5,
        )

        response = completion.choices[0].text
        shorten_response = response.replace("\n", "").strip()
        len_response = len(shorten_response.split(" "))
        if len_response >= 3500:
            shorten_response = "".join(shorten_response.split(" ")[:3500])
            print("X"*10)
            print(f"shorten_response is {shorten_response}")
            list_shorten = shorten_response.split(" ")
            print(list_shorten)
            print(f"length is {len(list_shorten)}")
        title_prompt = f"{shorten_response}，给这个文章写一个头条号风格的标题。增加标题的吸引力，可读性。"
        title_completion = openai.Completion.create(
            engine=model_engine,
            prompt=title_prompt,
            max_tokens=200,
            n=1,
            stop=None,
            temperature=0.5,
        )
        title_response = title_completion.choices[0].text

        # print(response)
        # if not os.path.isdir(topic_file_path):
        #     print("File folder  not exist")
        # topic_result_file = ""
        # topic_file_name_pattern = "step10_json_filestep9_merge_rewrite_"
        # for filename in os.listdir(topic_file_path):
        #     if filename.startswith(topic_file_name_pattern):
        #         topic_result_file = os.path.join(topic_file_path, filename)
        #
        # data_aligned = dict()
        # output_dir_name = "."
        # output_dir = os.path.join(output_dir_name, "result_topic_file")
        # Path(output_dir).mkdir(parents=True, exist_ok=True)
        # write_file_name = "save_server_" + topic_file_path.split("\\")[-1]
        # write_output_file_path = os.path.join(output_dir, write_file_name)
        #
        # with open(topic_result_file, encoding="utf8") as f:
        #         json_data = json.load(f)
        #         return json_data
        return response, title_response

    except Exception as ex:
        print("File  not exist")
        raise ex

def get_model_api():
    """Returns lambda function for api"""

    def model_api(input_title, input_domain):
        """
        Args:
            input_data: submitted to the API, raw string

        Returns:
            output_data: after some transformation, to be
                returned to the API

        """
        # print("X"*10)
        # print(f"input_title is {input_title}")
        # print(f"input_data2 is {input_domain}")
        punc = [",", "?", ".", ":", ";", "!", "(", ")", "[", "]"]
        # preds, title_preds = get_topic_result(input_title, input_domain)
        preds, title_preds = get_llm_result(input_title, input_domain)
        output_data = {"input_title": input_title, "output": preds, "title_output": title_preds}
        return output_data

    return model_api


# config = Config()
# model  = NERModel(config)