Andreas/LibreChat
1
0
Fork 0
mirror of https://github.com/danny-avila/LibreChat.git synced 2025-12-18 01:10:14 +01:00
Code Issues Projects Releases Packages Wiki Activity Actions 4

✨ feat: Add OpenAI Verbosity Parameter (#8929)

Browse source 
* WIP: Verbosity OpenAI Parameter

* 🔧 chore: remove unused import of extractEnvVariable from parsers.ts

* ✨ feat: add comprehensive tests for getOpenAIConfig and enhance verbosity handling

* fix: Handling for maxTokens in GPT-5+ models and add corresponding tests

* feat: Implement GPT-5+ model handling in processMemory function
This commit is contained in:
Danny Avila 2025-08-07 20:49:40 -04:00 • committed by GitHub
parent 486fe34a2b
commit 7147bce3c3
No known key found for this signature in database
GPG key ID: B5690EEEBB952194
14 changed files with 989 additions and 6 deletions
Download patch file Download diff file Expand all files Collapse all files

90
packages/api/src/endpoints/openai/llm.ts
View file

@ -8,6 +8,62 @@ import type * as t from '~/types';
import { sanitizeModelName, constructAzureURL } from '~/utils/azure';
import { isEnabled } from '~/utils/common';
export const knownOpenAIParams = new Set([
// Constructor/Instance Parameters
'model',
'modelName',
'temperature',
'topP',
'frequencyPenalty',
'presencePenalty',
'n',
'logitBias',
'stop',
'stopSequences',
'user',
'timeout',
'stream',
'maxTokens',
'maxCompletionTokens',
'logprobs',
'topLogprobs',
'apiKey',
'organization',
'audio',
'modalities',
'reasoning',
'zdrEnabled',
'service_tier',
'supportsStrictToolCalling',
'useResponsesApi',
'configuration',
// Call-time Options
'tools',
'tool_choice',
'functions',
'function_call',
'response_format',
'seed',
'stream_options',
'parallel_tool_calls',
'strict',
'prediction',
'promptIndex',
// Responses API specific
'text',
'truncation',
'include',
'previous_response_id',
// LangChain specific
'__includeRawResponse',
'maxConcurrency',
'maxRetries',
'verbose',
'streaming',
'streamUsage',
'disableStreaming',
]);
function hasReasoningParams({
reasoning_effort,
reasoning_summary,
@ -44,7 +100,7 @@ export function getOpenAIConfig(
addParams,
dropParams,
} = options;
const { reasoning_effort, reasoning_summary, ...modelOptions } = _modelOptions;
const { reasoning_effort, reasoning_summary, verbosity, ...modelOptions } = _modelOptions;
const llmConfig: Partial<t.ClientOptions> &
Partial<t.OpenAIParameters> &
Partial<AzureOpenAIInput> = Object.assign(
@ -55,8 +111,23 @@ export function getOpenAIConfig(
modelOptions,
);
const modelKwargs: Record<string, unknown> = {};
let hasModelKwargs = false;
if (verbosity != null && verbosity !== '') {
modelKwargs.verbosity = verbosity;
hasModelKwargs = true;
}
if (addParams && typeof addParams === 'object') {
Object.assign(llmConfig, addParams);
for (const [key, value] of Object.entries(addParams)) {
if (knownOpenAIParams.has(key)) {
(llmConfig as Record<string, unknown>)[key] = value;
} else {
hasModelKwargs = true;
modelKwargs[key] = value;
}
}
}
let useOpenRouter = false;
@ -223,6 +294,21 @@ export function getOpenAIConfig(
});
}
if (modelKwargs.verbosity && llmConfig.useResponsesApi === true) {
modelKwargs.text = { verbosity: modelKwargs.verbosity };
delete modelKwargs.verbosity;
}
if (llmConfig.model && /\bgpt-[5-9]\b/i.test(llmConfig.model) && llmConfig.maxTokens != null) {
modelKwargs.max_completion_tokens = llmConfig.maxTokens;
delete llmConfig.maxTokens;
hasModelKwargs = true;
}
if (hasModelKwargs) {
llmConfig.modelKwargs = modelKwargs;
}
const result: t.LLMConfigResult = {
llmConfig,
configOptions,