zoeissleeping/veridian
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Performance enhancements galore! New themining system

Browse Source 
This is once again a huge commit, but its mostly performance
improvements along with some bug fixes and refactoring. It also includes
changes to the theming systems. I'm still not 100% happy with the
theming system, but its better than before.

Model fetching has been dramatically improved! Nearly all the important
computation and pre-processing has been moved to the server. This has
also somehow fixed the way model details are loaded, which was causing
many models to be missing their details despite models.dev having them.

The markdown renderer has once again been changed, but I'm mostly
certain that this is the last time major changes will be made to it. The
renderer is not spamming components, bloating memory usage, and its not
using a bug prone custom written chunking system.

There's also a lot more that I haven't mentioned and honestly forgot. I
need to get better commit hygiene tbh.
This commit is contained in:
Zoe
2026-02-19 23:44:23 -06:00
parent 32a4f7f95d
commit 59bb7fbc12
85 changed files with 3523 additions and 2039 deletions
Download Patch File Download Diff File Expand all files Collapse all files
server/api/provider/[providerId]/models.post.ts
+271 -99
View File
@@ -1,7 +1,13 @@
import { type Entity } from '@triplit/client';
import Big from 'big.js';
import * as z from 'zod';
import { providerBaseUrls, SupportedModalities } from '~/types/model';
import { SupportedModalities } from '~/types/model';
import { Providers } from '~/types/model';
import { httpClient } from '~~/server/lib/triplit';
import { GatewayFetchError, getProviderDetails } from '~~/server/utils/ai-provider';
import { getModelsDevData } from '~~/server/utils/models-dev';
import { schema } from '~~/triplit/schema';
import { Err, Ok, type Result } from '~~/types/result';
export default defineEventHandler(async (event) => {
await protectRoute(event);
@@ -32,109 +38,130 @@ export default defineEventHandler(async (event) => {
});
}
let baseUrl;
let fetchUrl;
let headers;
switch (provider.type) {
case 'cohere':
case 'cerebras':
if (!result.data.providerApiKey) {
console.log("apiKey", result.data.providerApiKey);
const [providerModelsRes, modelsDevRes] = await Promise.all([
fetchProviderModels(provider, result.data.providerApiKey),
getModelsDevData(),
]);
if (!providerModelsRes.ok) {
switch (providerModelsRes.error) {
case ProviderFetchError.NoProviderApiKey: {
throw createError({
statusCode: 400,
message: `${provider.type} provider requires an API key`,
});
}
case 'google':
case 'openrouter':
baseUrl = !!provider.config.apiProxyUrl ? provider.config.apiProxyUrl : providerBaseUrls[provider.type];
baseUrl = baseUrl.replace(/\/$/, '');
if (baseUrl === '') {
case ProviderFetchError.NoProviderBaseUrl: {
throw createError({
statusCode: 400,
message: 'Invalid provider URL',
});
}
}
}
fetchUrl = `${baseUrl}/models`;
break;
case 'ollama':
if (!provider.config.apiProxyUrl) {
throw createError({
statusCode: 400,
message: 'Ollama provider requires an API proxy URL',
});
console.log("providerModelsRes.data", providerModelsRes.data);
const normalizedModels = await normalizeResponse(
providerModelsRes.data.data,
provider.type,
providerModelsRes.data.baseURL,
modelsDevRes
);
return {
models: normalizedModels,
};
});
enum ProviderFetchError {
NoProviderApiKey = 0,
NoProviderBaseUrl,
}
const fetchProviderModels = async (provider: Entity<typeof schema, 'providers'>, providerApiKey: string | undefined): Promise<Result<{ data: Record<string, any>, baseURL: string }, ProviderFetchError>> => {
const providerDetails = await getProviderDetails(provider, providerApiKey);
if (!providerDetails.ok) {
switch (providerDetails.error) {
case GatewayFetchError.NoProviderApiKey: {
// throw createError({
// statusCode: 400,
// message: `${provider.type} provider requires an API key`,
// });
return Err(ProviderFetchError.NoProviderApiKey);
}
baseUrl = provider.config.apiProxyUrl;
baseUrl = baseUrl.replace(/\/$/, '');
case GatewayFetchError.NoProviderBaseUrl: {
// throw createError({
// statusCode: 400,
// message: 'Invalid provider URL',
// });
return Err(ProviderFetchError.NoProviderBaseUrl);
}
}
}
fetchUrl = `${baseUrl}/api/tags`;
break;
const { endpoint: { baseURL, modelsEndpoint, headers } } = providerDetails.data;
switch (provider.type) {
case 'longcat':
// longcat doesn't have a model list endpoint so we just hardcode them here,
// sry. I talked to Meituan and this is what they said:
// 后续如果我们新增了这样的接口会及时同步您。
// en (approx): If we add an interface like this in the future, we will promptly update you
return {
models: [
{
id: "LongCat-Flash-Chat",
name: "LongCat Flash Chat",
attributes: {
inputModalities: ['text'],
outputModalities: ['text'],
capabilities: ['tools'],
contextWindow: 256_000,
return Ok({
data: {
models: [
{
id: "LongCat-Flash-Chat",
name: "LongCat Flash Chat",
attributes: {
inputModalities: ['text'],
outputModalities: ['text'],
capabilities: ['tools'],
contextWindow: 256_000,
}
},
{
id: "LongCat-Flash-Thinking",
name: "LongCat Flash Thinking",
attributes: {
inputModalities: ['text'],
outputModalities: ['text'],
capabilities: ['reasoning', 'tools'],
contextWindow: 256_000,
}
},
{
id: "LongCat-Flash-Thinking-2601",
name: "LongCat Flash Thinking (2601)",
attributes: {
inputModalities: ['text'],
outputModalities: ['text'],
capabilities: ['reasoning', 'tools'],
contextWindow: 256_000,
}
},
{
id: "LongCat-Flash-Lite",
name: "LongCat Flash Lite",
attributes: {
inputModalities: ['text'],
outputModalities: ['text'],
capabilities: ['tools'],
contextWindow: 320_000,
}
}
},
{
id: "LongCat-Flash-Thinking",
name: "LongCat Flash Thinking",
attributes: {
inputModalities: ['text'],
outputModalities: ['text'],
capabilities: ['reasoning', 'tools'],
contextWindow: 256_000,
}
},
{
id: "LongCat-Flash-Thinking-2601",
name: "LongCat Flash Thinking (2601)",
attributes: {
inputModalities: ['text'],
outputModalities: ['text'],
capabilities: ['reasoning', 'tools'],
contextWindow: 256_000,
}
},
{
id: "LongCat-Flash-Lite",
name: "LongCat Flash Lite",
attributes: {
inputModalities: ['text'],
outputModalities: ['text'],
capabilities: ['tools'],
contextWindow: 320_000,
}
}
]
};
}
if (provider.type === 'google') {
headers = {
'x-goog-api-key': `${result.data.providerApiKey}`
}
} else {
headers = {
'Authorization': `Bearer ${result.data.providerApiKey}`
}
],
},
baseURL
});
}
let res;
let data;
try {
res = await fetch(fetchUrl, {
res = await fetch(`${baseURL}${modelsEndpoint}`, {
method: 'GET',
headers
});
@@ -154,16 +181,89 @@ export default defineEventHandler(async (event) => {
});
}
return {
models: await normalizeResponse(data, provider.type, baseUrl)
};
});
return Ok({ data, baseURL });
}
const normalizeResponse = async (response: Record<string, any>, provider: typeof Providers[number], baseUrl: string) => {
function mergeSets(setA: Set<string>, setB: Set<string>): string[] {
return Array.from(new Set([...setA, ...setB]));
}
const getModelData = (modelId: string, providerId: string, modelsDevData: any) => {
const modelData = modelsDevData[providerId]?.models[modelId];
console.log("modelData", modelData);
if (modelData === undefined) return {
cost: {},
attributes: {
inputModalities: ['text'],
outputModalities: ['text'],
capabilities: [],
}
};
const capabilities = new Set<string>();
if (modelData.reasoning) {
capabilities.add('reasoning');
}
if (modelData.tool_call) {
capabilities.add('tools');
}
let inputModalities = modelData.modalities.input.filter((m: string) => (SupportedModalities as Readonly<string[]>).includes(m));
let outputModalities = modelData.modalities.output.filter((m: string) => (SupportedModalities as Readonly<string[]>).includes(m));
let supportedParameters = [];
if (modelData.temperature) {
supportedParameters.push('temperature');
}
let contextWindow = modelData.limit?.context || null;
let cost: Record<string, string> = {};
for (const key in modelData.cost) {
switch (key) {
case 'input': {
cost.prompt = modelData.cost[key].toString();
} break;
case 'output': {
cost.completion = modelData.cost[key].toString();
} break;
}
}
let releasedAt = (new Date(modelData.release_date)).getTime();
return {
name: modelData.name,
attributes: {
inputModalities,
outputModalities,
capabilities: modelData.capabilities || [],
contextWindow,
supported_parameters: modelData.supportedParameters,
},
cost,
releasedAt,
}
}
const formatBig = (bigValue: Big) => {
let str = bigValue.toString();
if (!str.includes('.')) return str + '.00';
if (str.split('.')[1]!.length === 1) return str + '0';
return str;
};
const normalizeResponse = async (response: Record<string, any>, provider: typeof Providers[number], baseUrl: string, modelsDevData: any) => {
switch (provider) {
case 'cerebras': {
console.log(response);
return response.data.map((model: any) => ({ id: model.id, releasedAt: model.created }));
console.log("response.data", response.data);
for (const model of response.data) {
console.log(model.created);
}
return response.data.map((model: any) => ({ ...getModelData(model.id, provider, modelsDevData), id: model.id }));
}
case 'openrouter': {
const models = [];
@@ -182,18 +282,69 @@ const normalizeResponse = async (response: Record<string, any>, provider: typeof
}
}
const pricing = {} as Record<string, string>;
for (const key of Object.keys(model.pricing)) {
// normalize to /1M tokens since models.dev is already in /1M tokens
switch (key) {
case 'prompt':
case 'completion':
case 'request':
case 'image':
case 'audio':
case 'discount':
pricing[key] = formatBig(Big(model.pricing[key]).mul(1_000_000));
break;
case 'image_tokens':
pricing['imageTokens'] = formatBig(Big(model.pricing[key]).mul(1_000_000));
break;
case 'image_output':
pricing['imageOutput'] = formatBig(Big(model.pricing[key]).mul(1_000_000));
break;
case 'audio_output':
pricing['audioOutput'] = formatBig(Big(model.pricing[key]).mul(1_000_000));
break;
case 'input_audio_cache':
pricing['inputAudioCache'] = formatBig(Big(model.pricing[key]).mul(1_000_000));
break;
case 'web_search':
pricing['webSearch'] = formatBig(Big(model.pricing[key]).mul(1_000_000));
break;
case 'internal_reasoning':
pricing['internalReasoning'] = formatBig(Big(model.pricing[key]).mul(1_000_000));
break;
case 'input_cache_read':
pricing['inputCacheRead'] = formatBig(Big(model.pricing[key]).mul(1_000_000));
break;
case 'input_cache_write':
pricing['inputCacheWrite'] = formatBig(Big(model.pricing[key]).mul(1_000_000));
break;
}
}
// TODO: do I really need to pull in data from models.dev for OR models?
// const modelData = getModelData(model.id, provider, modelsDevData);
models.push({
id: model.id as string,
name: model.name as string,
pricing: model.pricing,
cost: pricing,
attributes: {
// inputModalities: mergeSets(
// new Set(modelData.inputModalities || []),
// new Set(model.architecture.input_modalities.filter((modality: string) => (SupportedModalities as Readonly<string[]>).includes(modality))),
// ),
// outputModalities: mergeSets(
// new Set(modelData.outputModalities || []),
// new Set(model.architecture.output_modalities.filter((modality: string) => (SupportedModalities as Readonly<string[]>).includes(modality))),
// ),
inputModalities: model.architecture.input_modalities.filter((modality: string) => (SupportedModalities as Readonly<string[]>).includes(modality)),
outputModalities: model.architecture.output_modalities.filter((modality: string) => (SupportedModalities as Readonly<string[]>).includes(modality)),
capabilities: Array.from(capabilities),
contextWindow: model.context_length,
supported_parameters: model.supported_parameters,
},
created: model.created,
releasedAt: model.created * 1000,
});
}
@@ -255,12 +406,26 @@ const normalizeResponse = async (response: Record<string, any>, provider: typeof
console.error(e);
}
let normalizedId = model.name;
normalizedId = normalizedId.replace(/:cloud$/, '');
normalizedId = normalizedId.replace(/-cloud$/, '');
normalizedId = normalizedId.replace(/:latest$/, '');
const modelData = getModelData(normalizedId, 'ollama-cloud', modelsDevData);
models.set(model.name, {
...modelData,
id: model.name,
name: model.name,
attributes: {
inputModalities: Array.from(inputModalities),
capabilities: Array.from(capabilities),
...modelData.attributes,
inputModalities: mergeSets(
inputModalities,
new Set(modelData.attributes?.inputModalities || [])
),
capabilities: mergeSets(
capabilities,
new Set(modelData.attributes?.capabilities || [])
),
contextWindow,
}
});
@@ -272,12 +437,10 @@ const normalizeResponse = async (response: Record<string, any>, provider: typeof
return Array.from(models.values());
}
case 'google': {
console.log(response);
return response.models.map((model: any) => ({ id: model.name.replace('models/', ''), name: model.displayName }));
return response.models.map((model: any) => ({ ...getModelData(model.name.replace('models/', ''), provider, modelsDevData), id: model.name.replace('models/', ''), name: model.displayName }));
}
case 'longcat': {
console.log(response);
return response.models.map((model: any) => ({ id: model.name, name: model.name }));
return response.models;
}
case 'cohere': {
const models = [];
@@ -299,13 +462,22 @@ const normalizeResponse = async (response: Record<string, any>, provider: typeof
}
}
const modelData = getModelData(model.name, provider, modelsDevData);
models.push({
...modelData,
id: model.name,
name: model.name,
attributes: {
contextWindow: model.context_length,
inputModalities: Array.from(inputModalities),
capabilities: Array.from(capabilities),
...modelData.attributes,
inputModalities: mergeSets(
inputModalities,
new Set(modelData.attributes?.inputModalities || [])
),
capabilities: mergeSets(
capabilities,
new Set(modelData.attributes?.capabilities || [])
),
contextWindow: model.context_length || modelData.attributes?.contextWindow,
}
});
}