mirror of
https://gh.wpcy.net/https://github.com/discourse/discourse.git
synced 2026-05-28 13:16:34 +08:00
3332e9f4c3
`.annotaterb.yml` has carried `classified_sort: true` since the project
switched from `annotate` to `annotaterb` (commit 0eab7daea4, July 2025),
but annotaterb's default behaviour is to compare the existing schema
block against what it would generate and skip the rewrite when the
column list matches — even when the *ordering* of those columns differs.
The result is that models which haven't had a schema change since the
config landed never get reordered, and `classified_sort` drift
accumulates indefinitely.
`--force` makes annotaterb always rewrite, so a single `bin/rake
annotate:clean` run brings every model into the canonical format and
keeps them there. Every schema block is now grouped primary-key →
regular columns → timestamps → foreign keys (alphabetical within each
group). Pure annotation comment change — no code modifications.
Also cleans up the rake task to avoid string interpolation for `system`
calls.
88 lines
2.1 KiB
Ruby
Vendored
88 lines
2.1 KiB
Ruby
Vendored
# frozen_string_literal: true
|
|
|
|
class CompletionPrompt < ActiveRecord::Base
|
|
#
|
|
# DEPRECATED.
|
|
# TODO(roman): Remove after 06-17-25
|
|
#
|
|
|
|
TRANSLATE = -301
|
|
GENERATE_TITLES = -307
|
|
PROOFREAD = -303
|
|
MARKDOWN_TABLE = -304
|
|
CUSTOM_PROMPT = -305
|
|
EXPLAIN = -306
|
|
ILLUSTRATE_POST = -308
|
|
DETECT_TEXT_LOCALE = -309
|
|
|
|
enum :prompt_type, { text: 0, list: 1, diff: 2 }
|
|
|
|
validates :messages, length: { maximum: 20 }
|
|
validate :each_message_length
|
|
|
|
after_commit { DiscourseAi::AiHelper::Assistant.clear_prompt_cache! }
|
|
|
|
def self.enabled_by_name(name)
|
|
where(enabled: true).find_by(name: name)
|
|
end
|
|
|
|
attr_accessor :custom_instruction
|
|
|
|
def messages_with_input(input)
|
|
return unless input
|
|
|
|
user_input =
|
|
if id == CUSTOM_PROMPT && custom_instruction.present?
|
|
"#{custom_instruction}:\n#{input}"
|
|
else
|
|
input
|
|
end
|
|
|
|
instructions = [messages_hash[:insts], messages_hash[:post_insts].to_s].join("\n")
|
|
|
|
prompt = DiscourseAi::Completions::Prompt.new(instructions)
|
|
|
|
messages_hash[:examples].to_a.each do |example_pair|
|
|
prompt.push(type: :user, content: example_pair.first)
|
|
prompt.push(type: :model, content: example_pair.second)
|
|
end
|
|
|
|
prompt.push(type: :user, content: "<input>#{user_input}</input>")
|
|
|
|
prompt
|
|
end
|
|
|
|
private
|
|
|
|
def messages_hash
|
|
@messages_hash ||= messages.symbolize_keys!
|
|
end
|
|
|
|
def each_message_length
|
|
messages.each_with_index do |msg, idx|
|
|
next if msg["content"].length <= 1000
|
|
|
|
errors.add(:messages, I18n.t("discourse_ai.errors.prompt_message_length", idx: idx + 1))
|
|
end
|
|
end
|
|
end
|
|
|
|
# == Schema Information
|
|
#
|
|
# Table name: completion_prompts
|
|
#
|
|
# id :bigint not null, primary key
|
|
# enabled :boolean default(TRUE), not null
|
|
# messages :jsonb
|
|
# name :string not null
|
|
# prompt_type :integer default("text"), not null
|
|
# stop_sequences :string is an Array
|
|
# temperature :integer
|
|
# translated_name :string
|
|
# created_at :datetime not null
|
|
# updated_at :datetime not null
|
|
#
|
|
# Indexes
|
|
#
|
|
# index_completion_prompts_on_name (name)
|
|
#
|
