PromptManager 1.0.0

No project description provided

Prompt Manager Quick Guide

1.Introduction

Prompt Manager is a design and construction tool for Large Language Model prompts. It guides users to generate more accurate, reliable, and expected output content by helping them design better prompts. This tool provides SDK development modes for both technical and non-technical users, as well as interface interaction operation modes, to meet the needs of different populations using large models.

The main functions include model service management, scenario management, prompt template management, prompt development, and prompt application. The specific features are as follows:

1.Supports docking with commonly used large language models, including OpenAl's GPT model and other open source or custom large language model access interfaces;

2.Supports the management of prompt templates, scene and role management, preset commonly used prompt templates, including Zero-shot, Few-shot, COT, etc;

3.Supports interaction with the model through dialogue, and can develop prompts during the dialogue process;

4.Supports the construction, organization, and operation of prompt flows, and supports publishing flows as prompt applications;

5.Reminder workflow supports customizable Python scripts to meet personalized requirements;

6.Reminder workflows support docking with commonly used vector databases, such as Dingo-DB, Chroma, FAISS, etc;

7.Reminder workflows preset commonly used toolkits, including text segmentation, text segmentation, text conversion, etc;

8.Supports publishing prompt workflows as prompt word applications, accessed through the HTTP interface, and exporting the application as an SDK;

9.Supports running in SDK mode and can be quickly integrated into the development environment.

Modules

Prompt Market

In Prompt Market, the prompt manager displays all prompts in card form. It can support users to view or filter existing prompts, and also support users to operate custom prompts, such as adding, modifying, and querying.

AI Model

AI Model Service provides the ability to connect large models, and users can choose suitable large model services to obtain response results when developing prompts. When creating a model service, it is necessary to configure the service name, model service URL, and API Key. In the subsequent process of prompt development, the created model service can be quickly selected to respond to user input content.

Prompt Engineering

Prompt Engineering provides a development environment for users to design prompts. Users can communicate with LLM in the dialogue interaction interface, or build their own prompt topology through the prompt flow. It supports quick selection of scene templates, configuration of large model service parameters. Users can receive precise responses from the large language model by inputting designed prompts, which can be published into applications.

• Chat : provides two ways to communicate with LM: single prompt and multiple prompt, where the interactive dialogue is associated with contextual content to respond to the results;
• Flow : supports the construction of prompt flows in topological form, combining multiple prompts to achieve more complex and practical business logic.

Prompt App

Prompt App is a service for prompt flow, which supports publishing prompt workflows as prompt applications. Users can access the application through the HTTP interface or export it as an SDK.

2.Installation

To install Prompt Manager run

pip install

pip install promptmanager

3.Service management

• service start

pmctl service start

Open Web browser with this URL http://127.0.0.1:8888/

• service start with a port

pmctl service start -port 10000

• service start set proxy

pmctl service start -proxy xxxxxxxxxx

• service start set sqlite db path

pmctl service start -db /xxx/xxx/xxx/db.sqlite3

• service stop

pmctl service stop

4.Environment setup

In order to make sure your prompt script run successfully your need to prepare Python Environment first;

HTTP Request

If your already publish a Prompt Flow to an Application ;

curl http://127.0.0.1:8888/api/app/<appid>/run -X POST  -H 'Content-Type:application/json' -d '{"variables":[{"title":"","number":500}]}'

Python SDK

Prompt Template

from promptmanager.runtime.template import PMPromptTemplate
role_prompt = "i am role_prompt"
prompt_template = PMPromptTemplate("user","Tell me a ${adjective} joke about ${content}.",role_prompt)

variables={
            "adjective":"funny",
    		"content":"chickens"
          }

prompt_template.message(variables)

• Run prompt manager chat

from promptmanager.runtime.template import PMPromptTemplate
from promptmanager.runtime.template import PMChatPromptTemplate
chat_prompt_template = PMChatPromptTemplate(
    [
        PMPromptTemplate("user","Tell me a ${adjective} joke about ${content}.","i am role_prompt2"),
        PMPromptTemplate("system","Your name is ${name}.","i am role_prompt1")
    ]
)
 

variables={
            "adjective":"funny",
    		"content":"chickens",
            "name":"Bob"
          }
chat_prompt_template.messages(variables)

• OpenAI LLM

  from promptmanager.runtime.model import PMOpenAIPMLLM

  api_key = "xxxxxx-xxxxxxxxxxxxxxxxxxxx"
  pmOpenAIPMLLM = PMOpenAIPMLLM.load_from_openai_key(api_key)

  message = [{"role": "user", "content": "我要写一本书"}, {"role": "user", "content": "名字叫做《我和你》"}]
  params = {'temperature':0.8}

  result = pmOpenAIPMLLM.request_by_message(message, params)

• Fake LLM

from promptmanager.runtime.model import PMFakeLLM

response = [
  					'Action: Python REPL\nAction Input: chatGpt principle',
  					'Final Answer: mock result'
					]
pmFakeLLM = PMFakeLLM(response)

message = [{"role": "user", "content": "我要写一本书"}, {"role": "user", "content": "名字叫做《我和你》"}]
result = pmFakeLLM.request_result_by_message(message)

• Semantic recall from Vector Database

from promptmanager.script.chroma_reader import ChromaReader
from promptmanager.script.openai_embeddings import PMOpenAIEmbeddings

openai_embeddings=PMOpenAIEmbeddings(openai_api_key="xxxxxxxxxxxxxxxxx")
chroma_reader= ChromaReader(host="",port="",collection_name="")

• Build prompt flow

import threading

from promptmanager.runtime.model import PMLLM
from promptmanager.runtime.flow import PMFlow

if __name__ == '__main__':
    from promptmanager.runtime.flow import PMFlow

    # Step 1 init new PMFlow
    pm_flow = PMFlow(name="flow_name")

    # Step 2 get input_node and output_node
    input_node = pm_flow.get_input_node()
    output_node = pm_flow.get_output_node()

    input_node.show_io_info()
    # $>INFO: this is IOs of "input"::
    # $>INFO: outputs:[{'name': 'variable_assignment', 'type': 'any', 'defaultValue': None, 'value': None}]

    output_node.show_io_info()
    # $>INFO: this is IOs of "output":
    # $>INFO: inputs:[{'name': 'result1', 'type': 'any', 'defaultValue': None, 'value': None}]

    # Step 3 define a prompt node
    from promptmanager.runtime.flow import PMFlowTemplateNode
    from promptmanager.runtime.template import PMPromptTemplate

    template_content = """
    I want you act a famous novelist,
    I want to write a science fiction,
        The title is ${title} and number of words require ${number}.
    """
    role_name = "famous novelist"
    prompt_tempalte = PMPromptTemplate(template_content=template_content, role_prompt=role_name)

    openai_llm = PMLLM.load_from_path(path="../model/config.json")
    openai_llm.show_params_info()
    model_param_value = {
        "OPENAI_API_KEY": "xxxxxx-xxxxxxxxxxxxxxxxxxxx"
    }
    pm_template_node = PMFlowTemplateNode.from_template(name="prompt_tempalte_node", template=prompt_tempalte,
                                                        model=openai_llm, model_params_value=model_param_value)

    pm_template_node.show_io_info()
    # $>INFO: this is IOs of "prompt_tempalte_node":
    # $>INFO: inputs:[{'name': 'title', 'type': 'text', 'defaultValue': '', 'value': None}, {'name': 'number', 'type': 'text', 'defaultValue': '', 'value': None}]
    # $>INFO: outputs:[{'name': 'output', 'type': 'text', 'defaultValue': '', 'value': ''}]

    pm_template_node.show_info()
    # $>INFO: this is node info of "prompt_tempalte_node":
    # $>INFO: node info: {"id": "14e7709a-afd7-4eae-a100-ffa92d9cc6d5", "name": "prompt_tempalte_node", "module_id": null, "module_name": null, "module_type": "prompt", "left": null, "top": null, "description": null, "params": [{"name": "OPENAI_API_KEY", "type": "string", "defaultValue": "password", "value": "password"}, {"name": "model", "type": "select", "defaultValue": "gpt-3.5-turbo,gpt-4.0", "value": "gpt-3.5-turbo,gpt-4.0"}, {"name": "message", "type": "jsonarray", "defaultValue": null, "value": "[{\"role\": \"${role}\", \"content\": \"${content}\"}]"}, {"name": "temperature", "type": "int", "defaultValue": 0.7, "value": 0.7}, {"name": "stream", "type": "string", "defaultValue": true, "value": true}, {"name": "result", "type": "jsonarray", "defaultValue": null, "value": null}, {"name": "model_config", "type": "text", "value": "{\"protocol\":\"http\",\"method\":\"POST\",\"url\":\"https://api.openai.com/v1/chat/completions\",\"header\":{\"ContentType\":\"application/json\",\"Authorization\":\"Bearer ${OPENAI_API_KEY}\"},\"modelRole\":{\"user\":\"user\",\"system\":\"system\",\"assistant\":\"assistant\"},\"requestBody\":{\"model\":\"gpt-3.5-turbo-0613;gpt-3.5-turbo;gpt-3.5-turbo-16k-0613;gpt-3.5-turbo-16k;gpt-4-0613;gpt-4-32k-0613;gpt-4;gpt-4-32k\",\"messages\":{\"role\":\"${role}\",\"content\":\"${content}\"},\"temperature\":0.7,\"stream\":true},\"responseBody\":{\"id\":\"chatcmpl-7lZq4UwSCrkvyOTUcyReAMXpAydSQ\",\"object\":\"chat.completion\",\"created\":\"1691573536\",\"model\":\"gpt-3.5-turbo-0613\",\"choices\":[{\"index\":0,\"message\":{\"role\":\"assistant\",\"content\":\"${result_content}\"},\"finish_reason\":\"stop\"}],\"usage\":{\"prompt_tokens\":36,\"completion_tokens\":104,\"total_tokens\":140}},\"responseErrorBody\":{\"error\":{\"message\":\"$errorMessage\",\"type\":\"invalid_request_error\",\"param\":null,\"code\":null}}}"}, {"name": "model_param_define", "type": "text", "value": [{"name": "OPENAI_API_KEY", "type": "string", "defaultValue": "password", "value": "password"}, {"name": "model", "type": "select", "defaultValue": "gpt-3.5-turbo,gpt-4.0", "value": "gpt-3.5-turbo,gpt-4.0"}, {"name": "message", "type": "jsonarray", "defaultValue": null, "value": "[{\"role\": \"${role}\", \"content\": \"${content}\"}]"}, {"name": "temperature", "type": "int", "defaultValue": 0.7, "value": 0.7}, {"name": "stream", "type": "string", "defaultValue": true, "value": true}, {"name": "result", "type": "jsonarray", "defaultValue": null, "value": null}]}], "inputs": [{"name": "title", "type": "text", "defaultValue": "", "value": null}, {"name": "number", "type": "text", "defaultValue": "", "value": null}], "outputs": [{"name": "output", "type": "text", "defaultValue": "", "value": ""}], "prompt": "\n    I want you act a famous novelist,\n    I want to write a science fiction,\n        The title is ${title} and number of words require ${number}.\n    "}

    pm_flow.add_node(pm_template_node)

    # Step 4 define a tools script node
    from promptmanager.runtime.flow import PMFlowScriptNode

    script_node = PMFlowScriptNode(name="script_node", path="../script/python3_script.py")

    input = {
        "name": "input",
        "type": "text",
        "defaultValue": "this is single input"
    }
    script_node.add_input(input)

    # inputs = [{
    #     "name": "input1",
    #     "type": "text",
    #     "defaultValue": "this is input1 of inputs"
    # }, {
    #     "name": "input2",
    #     "type": "text",
    #     "defaultValue": "this is input2 of inputs"
    # }]
    # script_node.add_inputs(inputs)

    output = {
        "name": "output",
        "type": "text",
        "defaultValue": "this is single output"
    }
    script_node.add_output(output)

    # outputs = [{
    #     "name": "output1",
    #     "type": "text",
    #     "defaultValue": "this is output1 of outputs"
    # }, {
    #     "name": "output2",
    #     "type": "text",
    #     "defaultValue": "this is output2 of outputs"
    # }]
    # script_node.add_outputs(outputs)

    script_node.show_io_info()
    # $>INFO: this is IOs of "script_node":
    # $>INFO: inputs:[{'name': 'input', 'type': 'text', 'defaultValue': 'this is single input'}]
    # $>INFO: outputs:[{'name': 'output', 'type': 'text', 'defaultValue': 'this is single output'}]

    script_node.show_info()
    # $>INFO: this is node info of "script_node":
    # $>INFO: node info: {"id": "eed4b20e-112e-435e-8a65-a2d513a93b0e", "name": "script_node", "module_id": "00000000-0000-0000-1111-000000000001", "module_name": null, "module_type": "script", "left": null, "top": null, "description": "script_node", "params": {"script": [{"name": "script", "type": "text", "default_value": "../script/python3_script.py", "value": "../script/python3_script.py"}]}, "inputs": [{"name": "input", "type": "text", "defaultValue": "this is single input"}], "outputs": [{"name": "output", "type": "text", "defaultValue": "this is single output"}], "script_path": "../script/python3_script.py"}

    pm_flow.add_node(script_node)

    # Step 5 link nodes
    pm_flow.add_edge(source_node=input_node, source_node_output_name="variable_assignment",
                     target_node=pm_template_node, target_node_input_name="title")
    pm_flow.add_edge(source_node=input_node, source_node_output_name="variable_assignment",
                     target_node=pm_template_node, target_node_input_name="number")

    pm_flow.add_edge(source_node=pm_template_node, source_node_output_name="output",
                     target_node=script_node, target_node_input_name="input")

    pm_flow.add_edge(source_node=script_node, source_node_output_name="output",
                     target_node=output_node, target_node_input_name="result1")

    pm_flow.show_info()
    # $>INFO: this is the flow info of "flow_name":
    # $>INFO: info: {"id": "e56fb350-9b1f-4ac0-8179-0f10c9cc1543", "name": "flow_name", "nodes": [{"id": "21ee3784-1f11-40b3-ada5-f7d18df787ac", "name": "input", "module_id": "00000000-0000-0000-0000-000000000001", "module_name": "Input", "module_type": "input", "left": null, "top": null, "description": "Input", "params": [{"variable": "title", "type": "text", "defaultValue": "", "value": null}, {"variable": "number", "type": "text", "defaultValue": "", "value": null}], "inputs": [], "outputs": [{"name": "variable_assignment", "type": "any", "defaultValue": null, "value": null}]}, {"id": "122c66f7-09bc-47bb-af25-3f4dfcf6f351", "name": "prompt_tempalte_node", "module_id": null, "module_name": null, "module_type": "prompt", "left": null, "top": null, "description": null, "params": [{"name": "OPENAI_API_KEY", "type": "string", "defaultValue": "password", "value": "password"}, {"name": "model", "type": "select", "defaultValue": "gpt-3.5-turbo,gpt-4.0", "value": "gpt-3.5-turbo,gpt-4.0"}, {"name": "message", "type": "jsonarray", "defaultValue": null, "value": "[{\"role\": \"${role}\", \"content\": \"${content}\"}]"}, {"name": "temperature", "type": "int", "defaultValue": 0.7, "value": 0.7}, {"name": "stream", "type": "string", "defaultValue": true, "value": true}, {"name": "result", "type": "jsonarray", "defaultValue": null, "value": null}, {"name": "model_config", "type": "text", "value": "{\"protocol\":\"http\",\"method\":\"POST\",\"url\":\"https://api.openai.com/v1/chat/completions\",\"header\":{\"ContentType\":\"application/json\",\"Authorization\":\"Bearer ${OPENAI_API_KEY}\"},\"modelRole\":{\"user\":\"user\",\"system\":\"system\",\"assistant\":\"assistant\"},\"requestBody\":{\"model\":\"gpt-3.5-turbo-0613;gpt-3.5-turbo;gpt-3.5-turbo-16k-0613;gpt-3.5-turbo-16k;gpt-4-0613;gpt-4-32k-0613;gpt-4;gpt-4-32k\",\"messages\":{\"role\":\"${role}\",\"content\":\"${content}\"},\"temperature\":0.7,\"stream\":true},\"responseBody\":{\"id\":\"chatcmpl-7lZq4UwSCrkvyOTUcyReAMXpAydSQ\",\"object\":\"chat.completion\",\"created\":\"1691573536\",\"model\":\"gpt-3.5-turbo-0613\",\"choices\":[{\"index\":0,\"message\":{\"role\":\"assistant\",\"content\":\"${result_content}\"},\"finish_reason\":\"stop\"}],\"usage\":{\"prompt_tokens\":36,\"completion_tokens\":104,\"total_tokens\":140}},\"responseErrorBody\":{\"error\":{\"message\":\"$errorMessage\",\"type\":\"invalid_request_error\",\"param\":null,\"code\":null}}}"}, {"name": "model_param_define", "type": "text", "value": [{"name": "OPENAI_API_KEY", "type": "string", "defaultValue": "password", "value": "password"}, {"name": "model", "type": "select", "defaultValue": "gpt-3.5-turbo,gpt-4.0", "value": "gpt-3.5-turbo,gpt-4.0"}, {"name": "message", "type": "jsonarray", "defaultValue": null, "value": "[{\"role\": \"${role}\", \"content\": \"${content}\"}]"}, {"name": "temperature", "type": "int", "defaultValue": 0.7, "value": 0.7}, {"name": "stream", "type": "string", "defaultValue": true, "value": true}, {"name": "result", "type": "jsonarray", "defaultValue": null, "value": null}]}], "inputs": [{"name": "title", "type": "text", "defaultValue": "", "value": null}, {"name": "number", "type": "text", "defaultValue": "", "value": null}], "outputs": [{"name": "output", "type": "text", "defaultValue": "", "value": ""}], "prompt": "\n    I want you act a famous novelist,\n    I want to write a science fiction,\n        The title is ${title} and number of words require ${number}.\n    "}, {"id": "b163265f-f05d-4c7a-b6e8-a41997685477", "name": "script_node", "module_id": "00000000-0000-0000-1111-000000000001", "module_name": null, "module_type": "script", "left": null, "top": null, "description": "script_node", "params": {"script": [{"name": "script", "type": "text", "default_value": "../script/python3_script.py", "value": "../script/python3_script.py"}]}, "inputs": [{"name": "input", "type": "text", "defaultValue": "this is single input"}], "outputs": [{"name": "output", "type": "text", "defaultValue": "this is single output"}], "script_path": "../script/python3_script.py"}, {"id": "663a1619-a6ec-457a-ae30-854dd7234555", "name": "output", "module_id": "00000000-0000-0000-0000-000000000002", "module_name": "Output", "module_type": "output", "left": null, "top": null, "description": "Output", "params": [], "inputs": [{"name": "result1", "type": "any", "defaultValue": null, "value": null}, {"name": "result2", "type": "any", "defaultValue": null, "value": null}], "outputs": []}], "edges": [{"source_node": "21ee3784-1f11-40b3-ada5-f7d18df787ac", "source_output_name": "variable_assignment", "target_node": "122c66f7-09bc-47bb-af25-3f4dfcf6f351", "target_input_name": "title"}, {"source_node": "21ee3784-1f11-40b3-ada5-f7d18df787ac", "source_output_name": "variable_assignment", "target_node": "122c66f7-09bc-47bb-af25-3f4dfcf6f351", "target_input_name": "number"}, {"source_node": "122c66f7-09bc-47bb-af25-3f4dfcf6f351", "source_output_name": "output", "target_node": "b163265f-f05d-4c7a-b6e8-a41997685477", "target_input_name": "input"}, {"source_node": "b163265f-f05d-4c7a-b6e8-a41997685477", "source_output_name": "output", "target_node": "663a1619-a6ec-457a-ae30-854dd7234555", "target_input_name": "result1"}], "params": [], "flow_result": {"flow_id": null, "flow_name": null, "start_time": null, "end_time": null, "status": null, "nodes_info": null, "outputs": null}}

    # save pmflow
    pm_flow.save(save_path="/opt/data/text_pm.pmflow")

    # Step 5 pmflow run
    pm_flow.show_variables()
    # $>INFO: this is the flow variables of "flow_name":
    # $>INFO: variables:[{"variable": "title", "type": "text", "defaultValue": "", "value": null}, {"variable": "number", "type": "text", "defaultValue": "", "value": null}]

    pm_flow.run(variables={
        "title": "trip",
        "number": "500"
    }, run_async=False)

    # get flow run result
    # pm_flow.show_result(output_name="result1", wait_finish=False)
    pm_flow.show_result()

    # flow_result = pm_flow.get_result(wait_finish=True)

    # result = None
    # while pm_flow.get_result(wait_finish=False).is_finish():
    #     result = pm_flow.get_result(wait_finish=True)

• Load pmflow from disk

from promptmanager.runtime.flow import PMFlow 

# Load pmflow from disk
pmflow=PMFlow.load(file_path="/opt/data/text_pm.pmflow");

pmflow.show_variables()
#$>this is the variables of "xxxxxxxxxxxxxxx":
#$>input_variables:[{"name":"file_path","type":"text"}]
#$>output_variables:[{"name":"text_output","type":"text"}]

variables={"title":"Black hole traversal","number":500}
result = pmflow.run(variables=variables)

• publish pmflow to web app

from promptmanager.runtime.app import PMApp
from promptmanager.runtime.flow import PMFlow 

pmFlow = PMFlow.load('/opt/data/text_pm.pmflow');
# name is not required, it can define flow and app name
pmApp = PMApp.publish_from_flow(pmFlow, 'http://127.0.0.1:8888', name='test')
variables = {'title': 'Black hole traversal', 'number': 500}
pmApp.run_by_pm_flow(variables=variables)

pmApp.show_result()

• Run prompt manager application from web API

from promptmanager.runtime.app import PMApp

variables = {'title': 'Black hole traversal', 'number': 500}
url = 'http://127.0.0.1:8888/api/app/xxxxxxxx-xxxx-xxxx-xxxx-xxxxxxxxxxxx/run'
PMApp.run_by_app_url(url, variables)

• Run prompt manager application with file

First upload your file like this:

curl http://127.0.0.1:8888/api/app/<appid>/upload -F 'file=@/opt/knowledge_base.txt'

Then you will get the HTTP response result like this:

{
  "code":0,
  "data":{
      "uploadFilePath":"/tmp/promptmanager/upload/2023-03-08/xxxxxxxxxxxxxxxxx/knowledge_base.txt"
  }
}

Then get the uploadFilePath from the json and put it into the "file" variable

curl http://127.0.0.1:8888/api/app/<appid>/run -X POST -H 'Content-Type:application/json' -d '{"variables":[{"variable": "filename", "type": "file", "defaultValue": "", "value": "/tmp/promptmanager/upload/2023-03-08/xxxxxxxxxxxxxxxxx/knowledge_base.txt"}]}'