mirror of
https://github.com/discourse/discourse-ai.git
synced 2025-07-22 22:13:27 +00:00
9a79afcdbf
* DEV: Better strategies for summarization The strategy responsibility needs to be "Given a collection of texts, I know how to summarize them most efficiently, using the minimum amount of requests and maximizing token usage". There are different token limits for each model, so it all boils down to two different strategies: Fold all these texts into a single one, doing the summarization in chunks, and then build a summary from those. Build it by combining texts in a single prompt, and truncate it according to your token limits. While the latter is less than ideal, we need it for "bart-large-cnn-samsum" and "flan-t5-base-samsum", both with low limits. The rest will rely on folding. * Expose summarized chunks to users
97 lines
2.6 KiB
Ruby
97 lines
2.6 KiB
Ruby
# frozen_string_literal: true
|
|
|
|
module DiscourseAi
|
|
module Summarization
|
|
module Models
|
|
class OpenAi < Base
|
|
def display_name
|
|
"Open AI's #{model}"
|
|
end
|
|
|
|
def correctly_configured?
|
|
SiteSetting.ai_openai_api_key.present?
|
|
end
|
|
|
|
def configuration_hint
|
|
I18n.t(
|
|
"discourse_ai.summarization.configuration_hint",
|
|
count: 1,
|
|
setting: "ai_openai_api_key",
|
|
)
|
|
end
|
|
|
|
def concatenate_summaries(summaries)
|
|
messages = [
|
|
{ role: "system", content: "You are a helpful bot" },
|
|
{
|
|
role: "user",
|
|
content:
|
|
"Concatenate these disjoint summaries, creating a cohesive narrative:\n#{summaries.join("\n")}",
|
|
},
|
|
]
|
|
|
|
completion(messages)
|
|
end
|
|
|
|
def summarize_with_truncation(contents, opts)
|
|
messages = [{ role: "system", content: build_base_prompt(opts) }]
|
|
|
|
text_to_summarize = contents.map { |c| format_content_item(c) }.join
|
|
truncated_content = tokenizer.truncate(text_to_summarize, max_tokens - reserved_tokens)
|
|
|
|
messages << {
|
|
role: "user",
|
|
content: "Summarize the following in 400 words:\n#{truncated_content}",
|
|
}
|
|
|
|
completion(messages)
|
|
end
|
|
|
|
private
|
|
|
|
def summarize_chunk(chunk_text, opts)
|
|
completion(
|
|
[
|
|
{ role: "system", content: build_base_prompt(opts) },
|
|
{ role: "user", content: "Summarize the following in 400 words:\n#{chunk_text}" },
|
|
],
|
|
)
|
|
end
|
|
|
|
def build_base_prompt(opts)
|
|
base_prompt = <<~TEXT
|
|
You are a summarization bot.
|
|
You effectively summarise any text and reply ONLY with ONLY the summarized text.
|
|
You condense it into a shorter version.
|
|
You understand and generate Discourse forum Markdown.
|
|
TEXT
|
|
|
|
if opts[:resource_path]
|
|
base_prompt +=
|
|
"Try generating links as well the format is #{opts[:resource_path]}. eg: [ref](#{opts[:resource_path]}/77)\n"
|
|
end
|
|
|
|
base_prompt += "The discussion title is: #{opts[:content_title]}.\n" if opts[
|
|
:content_title
|
|
]
|
|
|
|
base_prompt
|
|
end
|
|
|
|
def completion(prompt)
|
|
::DiscourseAi::Inference::OpenAiCompletions.perform!(prompt, model).dig(
|
|
:choices,
|
|
0,
|
|
:message,
|
|
:content,
|
|
)
|
|
end
|
|
|
|
def tokenizer
|
|
DiscourseAi::Tokenizer::OpenAiTokenizer
|
|
end
|
|
end
|
|
end
|
|
end
|
|
end
|