Discource-C/discourse-ai
1
0
Fork 0
mirror of https://github.com/discourse/discourse-ai.git synced 2025-02-21 10:55:34 +00:00
Code Issues Packages Projects Releases Wiki Activity

FIX: use max_completion_tokens for open ai models (#1134)

Browse Source 
max_tokens is now deprecated per API
This commit is contained in:
Sam 2025-02-19 15:49:15 +11:00 committed by GitHub
parent 0c9466059c
commit 12f00a62d2
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
3 changed files with 50 additions and 1 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

6
lib/completions/endpoints/open_ai.rb
View File

@ -11,7 +11,11 @@ module DiscourseAi
def normalize_model_params(model_params)
model_params = model_params.dup
# max_tokens, temperature are already supported
# max_tokens is deprecated and is not functional on reasoning models
max_tokens = model_params.delete(:max_tokens)
model_params[:max_completion_tokens] = max_tokens if max_tokens
# temperature is already supported
if model_params[:stop_sequences]
model_params[:stop] = model_params.delete(:stop_sequences)
end

11
lib/completions/endpoints/open_router.rb
View File

@ -8,6 +8,17 @@ module DiscourseAi
%w[open_router].include?(model_provider)
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 prepare_request(payload)
headers = { "Content-Type" => "application/json" }
api_key = llm_model.api_key

34
spec/lib/completions/endpoints/open_ai_spec.rb
View File

@ -171,6 +171,40 @@ RSpec.describe DiscourseAi::Completions::Endpoints::OpenAi do
UploadCreator.new(image100x100, "image.jpg").create_for(Discourse.system_user.id)
end
describe "max tokens for reasoning models" do
it "uses max_completion_tokens for reasoning models" do
model.update!(name: "o3-mini")
llm = DiscourseAi::Completions::Llm.proxy("custom:#{model.id}")
prompt =
DiscourseAi::Completions::Prompt.new(
"You are a bot",
messages: [type: :user, content: "hello"],
)
response_text = <<~RESPONSE
data: {"id":"chatcmpl-B2VwlY6KzSDtHvg8pN1VAfRhhLFgn","object":"chat.completion.chunk","created":1739939159,"model":"o3-mini-2025-01-31","service_tier":"default","system_fingerprint":"fp_ef58bd3122","choices":[{"index":0,"delta":{"role":"assistant","content":"","refusal":null},"finish_reason":null}],"usage":null}
data: {"id":"chatcmpl-B2VwlY6KzSDtHvg8pN1VAfRhhLFgn","object":"chat.completion.chunk","created":1739939159,"model":"o3-mini-2025-01-31","service_tier":"default","system_fingerprint":"fp_ef58bd3122","choices":[{"index":0,"delta":{"content":"hello"},"finish_reason":null}],"usage":null}
data: {"id":"chatcmpl-B2VwlY6KzSDtHvg8pN1VAfRhhLFgn","object":"chat.completion.chunk","created":1739939159,"model":"o3-mini-2025-01-31","service_tier":"default","system_fingerprint":"fp_ef58bd3122","choices":[{"index":0,"delta":{},"finish_reason":"stop"}],"usage":null}
data: {"id":"chatcmpl-B2VwlY6KzSDtHvg8pN1VAfRhhLFgn","object":"chat.completion.chunk","created":1739939159,"model":"o3-mini-2025-01-31","service_tier":"default","system_fingerprint":"fp_ef58bd3122","choices":[],"usage":{"prompt_tokens":22,"completion_tokens":203,"total_tokens":225,"prompt_tokens_details":{"cached_tokens":0,"audio_tokens":0},"completion_tokens_details":{"reasoning_tokens":192,"audio_tokens":0,"accepted_prediction_tokens":0,"rejected_prediction_tokens":0}}}
data: [DONE]
RESPONSE
body_parsed = nil
stub_request(:post, "https://api.openai.com/v1/chat/completions").with(
body: ->(body) { body_parsed = JSON.parse(body) },
).to_return(body: response_text)
result = +""
llm.generate(prompt, user: user, max_tokens: 1000) { |chunk| result << chunk }
expect(result).to eq("hello")
expect(body_parsed["max_completion_tokens"]).to eq(1000)
end
end
describe "repeat calls" do
it "can properly reset context" do
llm = DiscourseAi::Completions::Llm.proxy("custom:#{model.id}")