Skip to content

add ALLOWED_OPENAI_PARAMS to openai pipeline #107

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Open
ti3x wants to merge 1 commit into
base: smart-window
Choose a base branch
from
Open

add ALLOWED_OPENAI_PARAMS to openai pipeline #107

Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
17 changes: 17 additions & 0 deletions toolkit/components/ml/content/backends/OpenAIPipeline.mjs
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -26,6 +26,8 @@ let _logLevel = "Error";
*/
const lazy = {};

const DEFAULT_ALLOWED_OPENAI_PARAMS = Object.freeze(["tools", "tool_choice"]);

ChromeUtils.defineLazyGetter(lazy, "console", () => {
return console.createInstance({
maxLogLevel: _logLevel, // we can't use maxLogLevelPref in workers.
Expand Down Expand Up @@ -68,6 +70,9 @@ export class OpenAIPipeline {
let config = {};
options.applyToConfig(config);
config.backend = config.backend || "openai";
if (!config.allowedOpenAIParams) {
Copy link
Member

@Mardak Mardak Nov 13, 2025

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

how do you expect this config to be set? if from createEngine, the allowed key options filters it out anyway https://searchfox.org/firefox-main/rev/77b6c9748bdd784eb5e0ee42603c408b34559d7d/toolkit/components/ml/content/EngineProcess.sys.mjs#769-770

at least for fork, we could conditional this on the modelId? quick test shows adding these params works for qwen, gpt4o, together but not mistral:

500 litellm.APIConnectionError: Vertex_aiException - b'
{"detail":[
{"type":"model_attributes_type","loc":["body","ChatCompletionRequest","tool_choice","ToolChoice"],"msg":"Input should be a valid dictionary or object to extract fields from","input":null},
{"type":"enum","loc":["body","ChatCompletionRequest","tool_choice","function-after[operator.attrgetter('value')(), str-enum[ToolChoiceEnum]]"],"msg":"Input should be 'auto', 'none', 'any' or 'required'","input":null,"ctx":{"expected":"'auto', 'none', 'any' or 'required'"}},
{"type":"missing","loc":["body","CompletionRequest","prompt"],"msg":"Field required","input":{"model":"mistral-small-2503","messages":[{"role":"system","content":"You are a …Today's date: Thursday, November 13, 2025"},{"role":"user","content":"hi"}]},
{"type":"extra_forbidden","loc":["body","OCRRequest","tools"],"msg":"Extra inputs are not permitted","input":[{"type":"function","function":{"name":"search_open_tabs","description":"Search …","parameters":{"type":"object","properties":{"type":{"type":"string","description":"The …"}},"required":["type"]}}},{"type":"function","function":{"name":"get_page_content","description":"Retrieve text from a specific browser tab. Choose whether to read the current viewport, Reader Mode, or full page content. The content is cleaned for analysis.","parameters":{"type":"object","properties":{"url":{"type":"string","description":"The …"},"mode":{"type":"string","enum":["viewport","reader","full"],"description":"Extraction …"}},"required":["url"]}}},{"type":"function","function":{"name":"search_history","description":"Search …","parameters":{"type":"object","properties":{"search_term":{"type":"string","description":"Keywords …"}},"required":["search_term"]}}}]},
{"type":"extra_forbidden","loc":["body","OCRRequest","tool_choice"],"msg":"Extra inputs are not permitted","input":null},
{"type":"extra_forbidden","loc":["body","OCRRequest","stream"],"msg":"Extra inputs are not permitted","input":true}
]}
'. Received Model Group=mistral-small-2503
Available Model Group Fallbacks=None

config.allowedOpenAIParams = DEFAULT_ALLOWED_OPENAI_PARAMS;
}

// reapply logLevel if it has changed.
if (lazy.console.logLevel != config.logLevel) {
Expand Down Expand Up @@ -334,6 +339,11 @@ export class OpenAIPipeline {
});
const stream = request.streamOptions?.enabled || false;
const tools = request.tools || [];
const allowedOpenAIParams =
request.allowed_openai_params ??
request.allowedOpenAIParams ??
this.#options.allowedOpenAIParams ??
Copy link
Author

@ti3x ti3x Nov 13, 2025 •
edited
Loading

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

@tarekziade not sure if these this.#options or config.allowedOpenAIParams check is necessary to allow future clients to set this value. Most of time we probably don't want to add this payload on every client call anyways.
I tested this and it shouldn’t break any non-TogetherAI inference servers — for example, it works fine with VertexAI and doesn’t affect the response payload. But if we want to be extra cautious, and don't want to do a conditional check against Model id in client-side code, we could do this in MLPA layer (check for model id prefix beign together_* and add this additional payload before passing off to Litellm)

Long term, LiteLLM should ideally support custom model configurations that allow passing through model-specific request parameters.

Would love to hear your thoughts on this.

DEFAULT_ALLOWED_OPENAI_PARAMS;

const completionParams = {
model: modelId,
Expand All @@ -342,6 +352,13 @@ export class OpenAIPipeline {
tools,
};

if (
Array.isArray(allowedOpenAIParams) &&
allowedOpenAIParams.length > 0
) {
completionParams.allowed_openai_params = [...allowedOpenAIParams];
}

const args = {
client,
completionParams,
Expand Down