fix: address review issues in model filter system
- brandMatcher: split keywords into includes/startsWith/segment/boundary modes to align with frontend BrandIcon.tsx matching semantics, preventing false-positive blocking from short keywords like o1, tts, command - handleSiteBlockModel: reject wildcard route patterns (e.g. gpt-*) from being written to site disabled list; only allow exact model names - Sites.tsx: extract brand grouping to useMemo, convert disabledModels to Set for O(1) lookup, remove duplicate IIFE computations in JSX - handleSiteBlockModel: replace window.confirm with project-style DOM dialog for visual consistency - enabledCounts: compute from filtered scope (excluding enabled filter itself) so counts reflect current brand/site/search filters - Fix declaration order: move enabledCounts after sortedRoutes to avoid ReferenceError
This commit is contained in:
Hureru
@@ -1,69 +1,244 @@
|
||||
/**
|
||||
* Server-side brand matching for global brand filtering.
|
||||
*
|
||||
* Brand keywords are extracted from the frontend BrandIcon.tsx BRAND_DEFINITIONS.
|
||||
* Each brand maps to a set of keywords that, when found in a model name, identify
|
||||
* the model as belonging to that brand.
|
||||
* Matching semantics are aligned with the frontend BrandIcon.tsx BRAND_DEFINITIONS:
|
||||
* - `includes`: substring match (e.g. 'claude' matches 'claude-3-opus')
|
||||
* - `startsWith`: segment-level startsWith (e.g. 'gpt' matches 'gpt-4o' but not 'some-gpt')
|
||||
* - `segment`: exact segment match after splitting on /:-_ delimiters
|
||||
* - `boundary`: word-boundary regex match
|
||||
*/
|
||||
|
||||
const BRAND_KEYWORDS: Record<string, string[]> = {
|
||||
'OpenAI': ['gpt', 'chatgpt', 'dall-e', 'whisper', 'text-embedding', 'text-moderation', 'davinci', 'babbage', 'codex-mini', 'o1', 'o3', 'o4', 'tts'],
|
||||
'Anthropic': ['claude'],
|
||||
'Google': ['gemini', 'gemma', 'palm', 'paligemma', 'shieldgemma', 'recurrentgemma', 'deplot', 'codegemma', 'imagen', 'learnlm', 'aqa', 'veo'],
|
||||
'DeepSeek': ['deepseek', 'ds-chat'],
|
||||
'通义千问': ['qwen', 'qwq', 'tongyi'],
|
||||
'智谱 AI': ['glm', 'chatglm', 'codegeex', 'cogview', 'cogvideo'],
|
||||
'Meta': ['llama', 'code-llama', 'codellama'],
|
||||
'Mistral': ['mistral', 'mixtral', 'codestral', 'pixtral', 'ministral', 'voxtral', 'magistral'],
|
||||
'Moonshot': ['moonshot', 'kimi'],
|
||||
'零一万物': ['yi-'],
|
||||
'文心一言': ['ernie', 'eb-'],
|
||||
'讯飞星火': ['spark', 'generalv'],
|
||||
'腾讯混元': ['hunyuan', 'tencent-hunyuan'],
|
||||
'豆包': ['doubao'],
|
||||
'MiniMax': ['minimax', 'abab'],
|
||||
'Cohere': ['command', 'c4ai-'],
|
||||
'Microsoft': ['phi-', 'kosmos'],
|
||||
'xAI': ['grok'],
|
||||
'阶跃星辰': ['stepfun', 'step-'],
|
||||
'Stability': ['flux', 'stablediffusion', 'stable-diffusion', 'sdxl', 'sd3'],
|
||||
'NVIDIA': ['nemotron', 'nemoretriever', 'neva', 'riva-translate', 'cosmos'],
|
||||
'IBM': ['granite'],
|
||||
'BAAI': ['bge-'],
|
||||
'ByteDance': ['bytedance', 'seed-oss', 'kolors', 'kwai', 'kwaipilot', 'wan-', 'kat-'],
|
||||
'InternLM': ['internlm'],
|
||||
'Midjourney': ['midjourney', 'mj_'],
|
||||
'DeepL': ['deepl-', 'deepl/'],
|
||||
'Jina AI': ['jina'],
|
||||
type MatchMode = 'includes' | 'startsWith' | 'segment' | 'boundary';
|
||||
type KeywordRule = { keyword: string; mode: MatchMode };
|
||||
|
||||
const BRAND_RULES: Record<string, KeywordRule[]> = {
|
||||
// OpenAI: all startsWithAny in frontend
|
||||
'OpenAI': [
|
||||
{ keyword: 'gpt', mode: 'startsWith' },
|
||||
{ keyword: 'chatgpt', mode: 'startsWith' },
|
||||
{ keyword: 'dall-e', mode: 'startsWith' },
|
||||
{ keyword: 'whisper', mode: 'startsWith' },
|
||||
{ keyword: 'text-embedding', mode: 'startsWith' },
|
||||
{ keyword: 'text-moderation', mode: 'startsWith' },
|
||||
{ keyword: 'davinci', mode: 'startsWith' },
|
||||
{ keyword: 'babbage', mode: 'startsWith' },
|
||||
{ keyword: 'codex-mini', mode: 'startsWith' },
|
||||
{ keyword: 'o1', mode: 'startsWith' },
|
||||
{ keyword: 'o3', mode: 'startsWith' },
|
||||
{ keyword: 'o4', mode: 'startsWith' },
|
||||
{ keyword: 'tts', mode: 'startsWith' },
|
||||
],
|
||||
// Anthropic: includesAny
|
||||
'Anthropic': [
|
||||
{ keyword: 'claude', mode: 'includes' },
|
||||
],
|
||||
// Google: includesAny + startsWithAny for veo/google/
|
||||
'Google': [
|
||||
{ keyword: 'gemini', mode: 'includes' },
|
||||
{ keyword: 'gemma', mode: 'includes' },
|
||||
{ keyword: 'google/', mode: 'includes' },
|
||||
{ keyword: 'palm', mode: 'includes' },
|
||||
{ keyword: 'paligemma', mode: 'includes' },
|
||||
{ keyword: 'shieldgemma', mode: 'includes' },
|
||||
{ keyword: 'recurrentgemma', mode: 'includes' },
|
||||
{ keyword: 'deplot', mode: 'includes' },
|
||||
{ keyword: 'codegemma', mode: 'includes' },
|
||||
{ keyword: 'imagen', mode: 'includes' },
|
||||
{ keyword: 'learnlm', mode: 'includes' },
|
||||
{ keyword: 'aqa', mode: 'includes' },
|
||||
{ keyword: 'veo', mode: 'startsWith' },
|
||||
],
|
||||
// DeepSeek: includesAny + hasExactSegment
|
||||
'DeepSeek': [
|
||||
{ keyword: 'deepseek', mode: 'includes' },
|
||||
{ keyword: 'ds-chat', mode: 'segment' },
|
||||
],
|
||||
'通义千问': [
|
||||
{ keyword: 'qwen', mode: 'includes' },
|
||||
{ keyword: 'qwq', mode: 'includes' },
|
||||
{ keyword: 'tongyi', mode: 'includes' },
|
||||
],
|
||||
'智谱 AI': [
|
||||
{ keyword: 'glm', mode: 'includes' },
|
||||
{ keyword: 'chatglm', mode: 'includes' },
|
||||
{ keyword: 'codegeex', mode: 'includes' },
|
||||
{ keyword: 'cogview', mode: 'includes' },
|
||||
{ keyword: 'cogvideo', mode: 'includes' },
|
||||
],
|
||||
'Meta': [
|
||||
{ keyword: 'llama', mode: 'includes' },
|
||||
{ keyword: 'code-llama', mode: 'includes' },
|
||||
{ keyword: 'codellama', mode: 'includes' },
|
||||
],
|
||||
'Mistral': [
|
||||
{ keyword: 'mistral', mode: 'includes' },
|
||||
{ keyword: 'mixtral', mode: 'includes' },
|
||||
{ keyword: 'codestral', mode: 'includes' },
|
||||
{ keyword: 'pixtral', mode: 'includes' },
|
||||
{ keyword: 'ministral', mode: 'includes' },
|
||||
{ keyword: 'voxtral', mode: 'includes' },
|
||||
{ keyword: 'magistral', mode: 'includes' },
|
||||
],
|
||||
'Moonshot': [
|
||||
{ keyword: 'moonshot', mode: 'includes' },
|
||||
{ keyword: 'kimi', mode: 'includes' },
|
||||
],
|
||||
// 零一万物: startsWithAny('yi-') + matchesBoundary
|
||||
'零一万物': [
|
||||
{ keyword: 'yi-', mode: 'startsWith' },
|
||||
{ keyword: 'yi', mode: 'boundary' },
|
||||
],
|
||||
'文心一言': [
|
||||
{ keyword: 'ernie', mode: 'includes' },
|
||||
{ keyword: 'eb-', mode: 'includes' },
|
||||
],
|
||||
'讯飞星火': [
|
||||
{ keyword: 'spark', mode: 'includes' },
|
||||
{ keyword: 'generalv', mode: 'includes' },
|
||||
],
|
||||
'腾讯混元': [
|
||||
{ keyword: 'hunyuan', mode: 'includes' },
|
||||
{ keyword: 'tencent-hunyuan', mode: 'includes' },
|
||||
],
|
||||
'豆包': [
|
||||
{ keyword: 'doubao', mode: 'includes' },
|
||||
],
|
||||
// MiniMax: includesAny + hasExactSegment
|
||||
'MiniMax': [
|
||||
{ keyword: 'minimax', mode: 'includes' },
|
||||
{ keyword: 'abab', mode: 'includes' },
|
||||
{ keyword: 'mini2.1', mode: 'segment' },
|
||||
],
|
||||
// Cohere: includesAny + startsWithAny('embed-')
|
||||
'Cohere': [
|
||||
{ keyword: 'command', mode: 'includes' },
|
||||
{ keyword: 'c4ai-', mode: 'includes' },
|
||||
{ keyword: 'embed-', mode: 'startsWith' },
|
||||
],
|
||||
// Microsoft: includesAny + hasExactSegment('phi4')
|
||||
'Microsoft': [
|
||||
{ keyword: 'microsoft/', mode: 'includes' },
|
||||
{ keyword: 'phi-', mode: 'includes' },
|
||||
{ keyword: 'kosmos', mode: 'includes' },
|
||||
{ keyword: 'phi4', mode: 'segment' },
|
||||
],
|
||||
'xAI': [
|
||||
{ keyword: 'grok', mode: 'includes' },
|
||||
],
|
||||
// 阶跃星辰: includesAny('stepfun') + startsWithAny('step-', 'step3')
|
||||
'阶跃星辰': [
|
||||
{ keyword: 'stepfun', mode: 'includes' },
|
||||
{ keyword: 'step-', mode: 'startsWith' },
|
||||
{ keyword: 'step3', mode: 'startsWith' },
|
||||
],
|
||||
// Stability: includesAny + startsWithAny('sd3')
|
||||
'Stability': [
|
||||
{ keyword: 'flux', mode: 'includes' },
|
||||
{ keyword: 'stablediffusion', mode: 'includes' },
|
||||
{ keyword: 'stable-diffusion', mode: 'includes' },
|
||||
{ keyword: 'sdxl', mode: 'includes' },
|
||||
{ keyword: 'sd3', mode: 'startsWith' },
|
||||
],
|
||||
// NVIDIA: includesAny + startsWithAny('nv-')
|
||||
'NVIDIA': [
|
||||
{ keyword: 'nvidia/', mode: 'includes' },
|
||||
{ keyword: 'nvclip', mode: 'includes' },
|
||||
{ keyword: 'nemotron', mode: 'includes' },
|
||||
{ keyword: 'nemoretriever', mode: 'includes' },
|
||||
{ keyword: 'neva', mode: 'includes' },
|
||||
{ keyword: 'riva-translate', mode: 'includes' },
|
||||
{ keyword: 'cosmos', mode: 'includes' },
|
||||
{ keyword: 'nv-', mode: 'startsWith' },
|
||||
],
|
||||
'IBM': [
|
||||
{ keyword: 'ibm/', mode: 'includes' },
|
||||
{ keyword: 'granite', mode: 'includes' },
|
||||
],
|
||||
'BAAI': [
|
||||
{ keyword: 'baai/', mode: 'includes' },
|
||||
{ keyword: 'bge-', mode: 'includes' },
|
||||
],
|
||||
// ByteDance: includesAny + startsWithAny('wan-', 'kat-')
|
||||
'ByteDance': [
|
||||
{ keyword: 'bytedance', mode: 'includes' },
|
||||
{ keyword: 'seed-oss', mode: 'includes' },
|
||||
{ keyword: 'kolors', mode: 'includes' },
|
||||
{ keyword: 'kwai', mode: 'includes' },
|
||||
{ keyword: 'kwaipilot', mode: 'includes' },
|
||||
{ keyword: 'wan-', mode: 'startsWith' },
|
||||
{ keyword: 'kat-', mode: 'startsWith' },
|
||||
],
|
||||
'InternLM': [
|
||||
{ keyword: 'internlm', mode: 'includes' },
|
||||
],
|
||||
// Midjourney: includesAny + startsWithAny('mj_')
|
||||
'Midjourney': [
|
||||
{ keyword: 'midjourney', mode: 'includes' },
|
||||
{ keyword: 'mj_', mode: 'startsWith' },
|
||||
],
|
||||
// DeepL: startsWithAny('deepl-') + includesAny('deepl/')
|
||||
'DeepL': [
|
||||
{ keyword: 'deepl/', mode: 'includes' },
|
||||
{ keyword: 'deepl-', mode: 'startsWith' },
|
||||
],
|
||||
'Jina AI': [
|
||||
{ keyword: 'jina', mode: 'includes' },
|
||||
],
|
||||
};
|
||||
|
||||
/**
|
||||
* Get all available brand names.
|
||||
*/
|
||||
export function getAllBrandNames(): string[] {
|
||||
return Object.keys(BRAND_KEYWORDS);
|
||||
return Object.keys(BRAND_RULES);
|
||||
}
|
||||
|
||||
/**
|
||||
* Get the flattened list of keywords for blocked brands.
|
||||
* Collect the rules for the blocked brands.
|
||||
*/
|
||||
export function getBlockedModelKeywords(blockedBrands: string[]): string[] {
|
||||
const keywords: string[] = [];
|
||||
export function getBlockedBrandRules(blockedBrands: string[]): KeywordRule[] {
|
||||
const rules: KeywordRule[] = [];
|
||||
for (const brand of blockedBrands) {
|
||||
const brandKeywords = BRAND_KEYWORDS[brand];
|
||||
if (brandKeywords) {
|
||||
keywords.push(...brandKeywords);
|
||||
}
|
||||
const brandRules = BRAND_RULES[brand];
|
||||
if (brandRules) rules.push(...brandRules);
|
||||
}
|
||||
return keywords;
|
||||
return rules;
|
||||
}
|
||||
|
||||
/**
|
||||
* Check if a model name matches any of the blocked keywords.
|
||||
* Uses a case-insensitive substring match (same approach as frontend BrandIcon).
|
||||
* Split model name into segments (matching the frontend's segment logic).
|
||||
*/
|
||||
export function isModelBlockedByBrand(modelName: string, blockedKeywords: string[]): boolean {
|
||||
const lower = modelName.trim().toLowerCase();
|
||||
if (!lower) return false;
|
||||
return blockedKeywords.some((keyword) => lower.includes(keyword));
|
||||
function getSegments(lower: string): string[] {
|
||||
return lower.split(/[/:,\s_\-]+/g).filter(Boolean);
|
||||
}
|
||||
|
||||
/**
|
||||
* Check if a model name matches a single keyword rule.
|
||||
*/
|
||||
function matchesRule(lower: string, segments: string[], rule: KeywordRule): boolean {
|
||||
switch (rule.mode) {
|
||||
case 'includes':
|
||||
return lower.includes(rule.keyword);
|
||||
case 'startsWith':
|
||||
// Match if any segment starts with the keyword, or the full name starts with it
|
||||
return lower.startsWith(rule.keyword) || segments.some((seg) => seg.startsWith(rule.keyword));
|
||||
case 'segment':
|
||||
return segments.includes(rule.keyword);
|
||||
case 'boundary': {
|
||||
const pattern = new RegExp(`(^|[^a-z0-9])${rule.keyword}(?=$|[^a-z0-9])`);
|
||||
return pattern.test(lower);
|
||||
}
|
||||
default:
|
||||
return false;
|
||||
}
|
||||
}
|
||||
|
||||
/**
|
||||
* Check if a model name matches any of the blocked brand rules.
|
||||
*/
|
||||
export function isModelBlockedByBrand(modelName: string, rules: KeywordRule[]): boolean {
|
||||
const lower = modelName.trim().toLowerCase();
|
||||
if (!lower || rules.length === 0) return false;
|
||||
const segments = getSegments(lower);
|
||||
return rules.some((rule) => matchesRule(lower, segments, rule));
|
||||
}
|
||||
|
||||
@@ -18,7 +18,7 @@ import {
|
||||
supportsDirectAccountRoutingConnection,
|
||||
} from './accountExtraConfig.js';
|
||||
import { invalidateTokenRouterCache } from './tokenRouter.js';
|
||||
import { getBlockedModelKeywords, isModelBlockedByBrand } from './brandMatcher.js';
|
||||
import { getBlockedBrandRules, isModelBlockedByBrand } from './brandMatcher.js';
|
||||
import { config } from '../config.js';
|
||||
import { setAccountRuntimeHealth } from './accountHealthService.js';
|
||||
import { clearAllRouteDecisionSnapshots } from './routeDecisionSnapshotStore.js';
|
||||
@@ -1142,14 +1142,14 @@ export async function rebuildTokenRoutesFromAvailability() {
|
||||
}
|
||||
|
||||
// Load global brand filter
|
||||
const blockedBrandKeywords = getBlockedModelKeywords(config.globalBlockedBrands);
|
||||
const blockedBrandRules = getBlockedBrandRules(config.globalBlockedBrands);
|
||||
|
||||
const modelCandidates = new Map<string, Map<string, { accountId: number; tokenId: number | null }>>();
|
||||
const addModelCandidate = (modelNameRaw: string | null | undefined, accountId: number, tokenId: number | null, siteId: number) => {
|
||||
const modelName = (modelNameRaw || '').trim();
|
||||
if (!modelName) return;
|
||||
if (isModelDisabledForSite(siteId, modelName)) return;
|
||||
if (blockedBrandKeywords.length > 0 && isModelBlockedByBrand(modelName, blockedBrandKeywords)) return;
|
||||
if (blockedBrandRules.length > 0 && isModelBlockedByBrand(modelName, blockedBrandRules)) return;
|
||||
if (!modelCandidates.has(modelName)) modelCandidates.set(modelName, new Map());
|
||||
const candidateKey = `${accountId}:${tokenId ?? 'account'}`;
|
||||
modelCandidates.get(modelName)!.set(candidateKey, { accountId, tokenId });
|
||||
|
||||
Reference in New Issue
Block a user