discourse-ai/lib/ai_bot/bot.rb

193 lines
6.1 KiB
Ruby
Raw Normal View History

# frozen_string_literal: true
module DiscourseAi
module AiBot
class Bot
BOT_NOT_FOUND = Class.new(StandardError)
MAX_COMPLETIONS = 5
def self.as(bot_user, persona: DiscourseAi::AiBot::Personas::General.new)
new(bot_user, persona)
end
def initialize(bot_user, persona)
@bot_user = bot_user
@persona = persona
end
attr_reader :bot_user
def get_updated_title(conversation_context, post_user)
system_insts = <<~TEXT.strip
You are titlebot. Given a topic, you will figure out a title.
You will never respond with anything but 7 word topic title.
TEXT
title_prompt =
DiscourseAi::Completions::Prompt.new(system_insts, messages: conversation_context)
title_prompt.push(
type: :user,
content:
"Based on our previous conversation, suggest a 7 word title without quoting any of it.",
)
DiscourseAi::Completions::Llm
.proxy(model)
.generate(title_prompt, user: post_user)
.strip
.split("\n")
.last
end
def reply(context, &update_blk)
prompt = persona.craft_prompt(context)
total_completions = 0
ongoing_chain = true
low_cost = false
raw_context = []
while total_completions <= MAX_COMPLETIONS && ongoing_chain
current_model = model(prefer_low_cost: low_cost)
llm = DiscourseAi::Completions::Llm.proxy(current_model)
tool_found = false
result =
llm.generate(prompt, user: context[:user]) do |partial, cancel|
if (tool = persona.find_tool(partial))
tool_found = true
ongoing_chain = tool.chain_next_response?
low_cost = tool.low_cost?
tool_call_id = tool.tool_call_id
invocation_result_json = invoke_tool(tool, llm, cancel, &update_blk).to_json
tool_call_message = {
type: :tool_call,
id: tool_call_id,
content: { name: tool.name, arguments: tool.parameters }.to_json,
}
tool_message = { type: :tool, id: tool_call_id, content: invocation_result_json }
if tool.standalone?
2024-01-19 06:51:26 -05:00
standalone_context =
context.dup.merge(
conversation_context: [
context[:conversation_context].last,
tool_call_message,
tool_message,
],
)
2024-01-19 06:51:26 -05:00
prompt = persona.craft_prompt(standalone_context)
else
prompt.push(**tool_call_message)
prompt.push(**tool_message)
end
raw_context << [tool_call_message[:content], tool_call_id, "tool_call"]
raw_context << [invocation_result_json, tool_call_id, "tool"]
else
update_blk.call(partial, cancel, nil)
end
end
if !tool_found
ongoing_chain = false
raw_context << [result, bot_user.username]
end
total_completions += 1
# do not allow tools when we are at the end of a chain (total_completions == MAX_COMPLETIONS)
prompt.tools = [] if total_completions == MAX_COMPLETIONS
FEATURE: UI to update ai personas on admin page (#290) Introduces a UI to manage customizable personas (admin only feature) Part of the change was some extensive internal refactoring: - AIBot now has a persona set in the constructor, once set it never changes - Command now takes in bot as a constructor param, so it has the correct persona and is not generating AIBot objects on the fly - Added a .prettierignore file, due to the way ALE is configured in nvim it is a pre-req for prettier to work - Adds a bunch of validations on the AIPersona model, system personas (artist/creative etc...) are all seeded. We now ensure - name uniqueness, and only allow certain properties to be touched for system personas. - (JS note) the client side design takes advantage of nested routes, the parent route for personas gets all the personas via this.store.findAll("ai-persona") then child routes simply reach into this model to find a particular persona. - (JS note) data is sideloaded into the ai-persona model the meta property supplied from the controller, resultSetMeta - This removes ai_bot_enabled_personas and ai_bot_enabled_chat_commands, both should be controlled from the UI on a per persona basis - Fixes a long standing bug in token accounting ... we were doing to_json.length instead of to_json.to_s.length - Amended it so {commands} are always inserted at the end unconditionally, no need to add it to the template of the system message as it just confuses things - Adds a concept of required_commands to stock personas, these are commands that must be configured for this stock persona to show up. - Refactored tests so we stop requiring inference_stubs, it was very confusing to need it, added to plugin.rb for now which at least is clearer - Migrates the persona selector to gjs --------- Co-authored-by: Joffrey JAFFEUX <j.jaffeux@gmail.com> Co-authored-by: Martin Brennan <martin@discourse.org>
2023-11-21 00:56:43 -05:00
end
raw_context
end
attr_reader :persona
private
def invoke_tool(tool, llm, cancel, &update_blk)
update_blk.call("", cancel, build_placeholder(tool.summary, ""))
result =
tool.invoke(bot_user, llm) do |progress|
placeholder = build_placeholder(tool.summary, progress)
update_blk.call("", cancel, placeholder)
end
tool_details = build_placeholder(tool.summary, tool.details, custom_raw: tool.custom_raw)
update_blk.call(tool_details, cancel, nil)
result
end
def model(prefer_low_cost: false)
# HACK(roman): We'll do this until we define how we represent different providers in the bot settings
default_model =
case bot_user.id
when DiscourseAi::AiBot::EntryPoint::CLAUDE_V2_ID
if DiscourseAi::Completions::Endpoints::AwsBedrock.correctly_configured?("claude-2")
"aws_bedrock:claude-2"
else
"anthropic:claude-2"
end
when DiscourseAi::AiBot::EntryPoint::GPT4_ID
"open_ai:gpt-4"
when DiscourseAi::AiBot::EntryPoint::GPT4_TURBO_ID
"open_ai:gpt-4-turbo"
when DiscourseAi::AiBot::EntryPoint::GPT3_5_TURBO_ID
"open_ai:gpt-3.5-turbo-16k"
when DiscourseAi::AiBot::EntryPoint::MIXTRAL_ID
if DiscourseAi::Completions::Endpoints::Vllm.correctly_configured?(
"mistralai/Mixtral-8x7B-Instruct-v0.1",
)
"vllm:mistralai/Mixtral-8x7B-Instruct-v0.1"
else
"hugging_face:mistralai/Mixtral-8x7B-Instruct-v0.1"
end
when DiscourseAi::AiBot::EntryPoint::GEMINI_ID
"google:gemini-pro"
when DiscourseAi::AiBot::EntryPoint::FAKE_ID
"fake:fake"
else
nil
end
if %w[open_ai:gpt-4 open_ai:gpt-4-turbo].include?(default_model) && prefer_low_cost
return "open_ai:gpt-3.5-turbo-16k"
end
default_model
end
def tool_invocation?(partial)
Nokogiri::HTML5.fragment(partial).at("invoke").present?
end
def build_placeholder(summary, details, custom_raw: nil)
placeholder = +(<<~HTML)
<details>
<summary>#{summary}</summary>
<p>#{details}</p>
</details>
HTML
if custom_raw
placeholder << "\n"
placeholder << custom_raw
else
# we need this for cursor placeholder to work
# doing this in CSS is very hard
# if changing test with a custom tool such as search
placeholder << "<span></span>\n\n"
end
placeholder
end
end
end
end