mirror of
https://github.com/discourse/discourse-ai.git
synced 2025-08-06 13:13:28 +00:00
e817b7dc11
This re-implements tool support in DiscourseAi::Completions::Llm #generate
Previously tool support was always returned via XML and it would be the responsibility of the caller to parse XML
New implementation has the endpoints return ToolCall objects.
Additionally this simplifies the Llm endpoint interface and gives it more clarity. Llms must implement
decode, decode_chunk (for streaming)
It is the implementers responsibility to figure out how to decode chunks, base no longer implements. To make this easy we ship a flexible json decoder which is easy to wire up.
Also (new)
Better debugging for PMs, we now have a next / previous button to see all the Llm messages associated with a PM
Token accounting is fixed for vllm (we were not correctly counting tokens)
78 lines
2.0 KiB
Ruby
78 lines
2.0 KiB
Ruby
# frozen_string_literal: true
|
|
|
|
module DiscourseAi
|
|
module Completions
|
|
module Endpoints
|
|
class CannedResponse
|
|
CANNED_RESPONSE_ERROR = Class.new(StandardError)
|
|
|
|
def initialize(responses)
|
|
@responses = responses
|
|
@completions = 0
|
|
@dialect = nil
|
|
end
|
|
|
|
def normalize_model_params(model_params)
|
|
# max_tokens, temperature, stop_sequences are already supported
|
|
model_params
|
|
end
|
|
|
|
attr_reader :responses, :completions, :dialect
|
|
|
|
def prompt_messages
|
|
dialect.prompt.messages
|
|
end
|
|
|
|
def perform_completion!(
|
|
dialect,
|
|
_user,
|
|
_model_params,
|
|
feature_name: nil,
|
|
feature_context: nil
|
|
)
|
|
@dialect = dialect
|
|
response = responses[completions]
|
|
if response.nil?
|
|
raise CANNED_RESPONSE_ERROR,
|
|
"The number of completions you requested exceed the number of canned responses"
|
|
end
|
|
|
|
raise response if response.is_a?(StandardError)
|
|
|
|
@completions += 1
|
|
if block_given?
|
|
cancelled = false
|
|
cancel_fn = lambda { cancelled = true }
|
|
|
|
# We buffer and return tool invocations in one go.
|
|
as_array = response.is_a?(Array) ? response : [response]
|
|
as_array.each do |response|
|
|
if is_tool?(response)
|
|
yield(response, cancel_fn)
|
|
else
|
|
response.each_char do |char|
|
|
break if cancelled
|
|
yield(char, cancel_fn)
|
|
end
|
|
end
|
|
end
|
|
end
|
|
|
|
response = response.first if response.is_a?(Array) && response.length == 1
|
|
response
|
|
end
|
|
|
|
def tokenizer
|
|
DiscourseAi::Tokenizer::OpenAiTokenizer
|
|
end
|
|
|
|
private
|
|
|
|
def is_tool?(response)
|
|
response.is_a?(DiscourseAi::Completions::ToolCall)
|
|
end
|
|
end
|
|
end
|
|
end
|
|
end
|