mirror of
https://github.com/discourse/discourse-ai.git
synced 2025-07-05 14:02:13 +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)
90 lines
2.1 KiB
Ruby
90 lines
2.1 KiB
Ruby
# frozen_string_literal: true
|
|
|
|
module DiscourseAi
|
|
module Completions
|
|
module Dialects
|
|
class Ollama < Dialect
|
|
class << self
|
|
def can_translate?(model_provider)
|
|
model_provider == "ollama"
|
|
end
|
|
end
|
|
|
|
def native_tool_support?
|
|
enable_native_tool?
|
|
end
|
|
|
|
def max_prompt_tokens
|
|
llm_model.max_prompt_tokens
|
|
end
|
|
|
|
private
|
|
|
|
def tools_dialect
|
|
if enable_native_tool?
|
|
@tools_dialect ||= DiscourseAi::Completions::Dialects::OllamaTools.new(prompt.tools)
|
|
else
|
|
super
|
|
end
|
|
end
|
|
|
|
def tokenizer
|
|
llm_model.tokenizer_class
|
|
end
|
|
|
|
def model_msg(msg)
|
|
{ role: "assistant", content: msg[:content] }
|
|
end
|
|
|
|
def tool_call_msg(msg)
|
|
tools_dialect.from_raw_tool_call(msg)
|
|
end
|
|
|
|
def tool_msg(msg)
|
|
tools_dialect.from_raw_tool(msg)
|
|
end
|
|
|
|
def system_msg(msg)
|
|
msg = { role: "system", content: msg[:content] }
|
|
|
|
if tools_dialect.instructions.present?
|
|
msg[:content] = msg[:content].dup << "\n\n#{tools_dialect.instructions}"
|
|
end
|
|
|
|
msg
|
|
end
|
|
|
|
def enable_native_tool?
|
|
return @enable_native_tool if defined?(@enable_native_tool)
|
|
|
|
@enable_native_tool = llm_model.lookup_custom_param("enable_native_tool")
|
|
end
|
|
|
|
def user_msg(msg)
|
|
user_message = { role: "user", content: msg[:content] }
|
|
|
|
encoded_uploads = prompt.encoded_uploads(msg)
|
|
if encoded_uploads.present?
|
|
images =
|
|
encoded_uploads
|
|
.map do |upload|
|
|
if upload[:mime_type].start_with?("image/")
|
|
upload[:base64]
|
|
else
|
|
nil
|
|
end
|
|
end
|
|
.compact
|
|
|
|
user_message[:images] = images if images.present?
|
|
end
|
|
|
|
# TODO: Add support for user messages with embedded user ids
|
|
|
|
user_message
|
|
end
|
|
end
|
|
end
|
|
end
|
|
end
|