mirror of
https://github.com/discourse/discourse-ai.git
synced 2025-02-18 01:14:53 +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)
89 lines
2.2 KiB
Ruby
89 lines
2.2 KiB
Ruby
# frozen_string_literal: true
|
|
|
|
module DiscourseAi
|
|
module Completions
|
|
module Endpoints
|
|
class SambaNova < Base
|
|
def self.can_contact?(model_provider)
|
|
model_provider == "samba_nova"
|
|
end
|
|
|
|
def normalize_model_params(model_params)
|
|
model_params = model_params.dup
|
|
|
|
# max_tokens, temperature are already supported
|
|
if model_params[:stop_sequences]
|
|
model_params[:stop] = model_params.delete(:stop_sequences)
|
|
end
|
|
|
|
model_params
|
|
end
|
|
|
|
def default_options
|
|
{ model: llm_model.name }
|
|
end
|
|
|
|
def provider_id
|
|
AiApiAuditLog::Provider::SambaNova
|
|
end
|
|
|
|
private
|
|
|
|
def model_uri
|
|
URI(llm_model.url)
|
|
end
|
|
|
|
def prepare_payload(prompt, model_params, dialect)
|
|
payload = default_options.merge(model_params).merge(messages: prompt)
|
|
|
|
payload[:stream] = true if @streaming_mode
|
|
|
|
payload
|
|
end
|
|
|
|
def prepare_request(payload)
|
|
headers = { "Content-Type" => "application/json" }
|
|
api_key = llm_model.api_key
|
|
|
|
headers["Authorization"] = "Bearer #{api_key}"
|
|
|
|
Net::HTTP::Post.new(model_uri, headers).tap { |r| r.body = payload }
|
|
end
|
|
|
|
def final_log_update(log)
|
|
log.request_tokens = @prompt_tokens if @prompt_tokens
|
|
log.response_tokens = @completion_tokens if @completion_tokens
|
|
end
|
|
|
|
def xml_tools_enabled?
|
|
true
|
|
end
|
|
|
|
def decode(response_raw)
|
|
json = JSON.parse(response_raw, symbolize_names: true)
|
|
[json.dig(:choices, 0, :message, :content)]
|
|
end
|
|
|
|
def decode_chunk(chunk)
|
|
@json_decoder ||= JsonStreamDecoder.new
|
|
(@json_decoder << chunk)
|
|
.map do |json|
|
|
text = json.dig(:choices, 0, :delta, :content)
|
|
|
|
@prompt_tokens ||= json.dig(:usage, :prompt_tokens)
|
|
@completion_tokens ||= json.dig(:usage, :completion_tokens)
|
|
|
|
if !text.to_s.empty?
|
|
text
|
|
else
|
|
nil
|
|
end
|
|
end
|
|
.flatten
|
|
.compact
|
|
end
|
|
end
|
|
end
|
|
end
|
|
end
|