mirror of
https://github.com/discourse/discourse-ai.git
synced 2025-10-31 14:38:37 +00:00
4b42c09814
* FEATURE: Tweak HyDE prompts for better grounding in forum subject and limit response size * fix test * lint
36 lines
917 B
Ruby
36 lines
917 B
Ruby
# frozen_string_literal: true
|
|
|
|
module DiscourseAi
|
|
module Embeddings
|
|
module HydeGenerators
|
|
class Llama2 < DiscourseAi::Embeddings::HydeGenerators::Base
|
|
def prompt(search_term)
|
|
<<~TEXT
|
|
[INST] <<SYS>>
|
|
You are a helpful bot
|
|
You create forum posts about a given subject
|
|
<</SYS>>
|
|
|
|
#{basic_prompt_instruction}
|
|
#{search_term}
|
|
[/INST]
|
|
Here is a forum post about the above subject:
|
|
TEXT
|
|
end
|
|
|
|
def models
|
|
["Llama2-*-chat-hf"]
|
|
end
|
|
|
|
def hypothetical_post_from(query)
|
|
::DiscourseAi::Inference::HuggingFaceTextGeneration.perform!(
|
|
prompt(query),
|
|
SiteSetting.ai_embeddings_semantic_search_hyde_model,
|
|
token_limit: 400,
|
|
).dig(:generated_text)
|
|
end
|
|
end
|
|
end
|
|
end
|
|
end
|