mirror of
https://github.com/discourse/discourse-ai.git
synced 2025-03-07 01:39:54 +00:00
e0bf6adb5b
This PR adds tool support to available LLMs. We'll buffer tool invocations and return them instead of making users of this service parse the response. It also adds support for conversation context in the generic prompt. It includes bot messages, user messages, and tool invocations, which we'll trim to make sure it doesn't exceed the prompt limit, then translate them to the correct dialect. Finally, It adds some buffering when reading chunks to handle cases when streaming is extremely slow.:M
73 lines
1.9 KiB
Ruby
73 lines
1.9 KiB
Ruby
# frozen_string_literal: true
|
|
|
|
module DiscourseAi
|
|
module Completions
|
|
module Dialects
|
|
class OrcaStyle < Dialect
|
|
class << self
|
|
def can_translate?(model_name)
|
|
%w[StableBeluga2 Upstage-Llama-2-*-instruct-v2].include?(model_name)
|
|
end
|
|
|
|
def tokenizer
|
|
DiscourseAi::Tokenizer::Llama2Tokenizer
|
|
end
|
|
end
|
|
|
|
def translate
|
|
orca_style_prompt = +<<~TEXT
|
|
### System:
|
|
#{prompt[:insts]}
|
|
#{build_tools_prompt}#{prompt[:post_insts]}
|
|
TEXT
|
|
|
|
if prompt[:examples]
|
|
prompt[:examples].each do |example_pair|
|
|
orca_style_prompt << "### User:\n#{example_pair.first}\n"
|
|
orca_style_prompt << "### Assistant:\n#{example_pair.second}\n"
|
|
end
|
|
end
|
|
|
|
orca_style_prompt << "### User:\n#{prompt[:input]}\n"
|
|
|
|
orca_style_prompt << "### Assistant:\n"
|
|
end
|
|
|
|
def conversation_context
|
|
return "" if prompt[:conversation_context].blank?
|
|
|
|
trimmed_context = trim_context(prompt[:conversation_context])
|
|
|
|
trimmed_context
|
|
.reverse
|
|
.reduce(+"") do |memo, context|
|
|
memo << (context[:type] == "user" ? "### User:" : "### Assistant:")
|
|
|
|
if context[:type] == "tool"
|
|
memo << <<~TEXT
|
|
|
|
<function_results>
|
|
<result>
|
|
<tool_name>#{context[:name]}</tool_name>
|
|
<json>
|
|
#{context[:content]}
|
|
</json>
|
|
</result>
|
|
</function_results>
|
|
TEXT
|
|
else
|
|
memo << " " << context[:content] << "\n"
|
|
end
|
|
|
|
memo
|
|
end
|
|
end
|
|
|
|
def max_prompt_tokens
|
|
SiteSetting.ai_hugging_face_token_limit
|
|
end
|
|
end
|
|
end
|
|
end
|
|
end
|