mirror of
https://github.com/CJackHwang/ds2api.git
synced 2026-05-13 12:47:41 +08:00
feat: Implement DeepSeek integration, refactor model adapters for streaming and tool calls, enhance admin and account management, and introduce new UI features for settings, API testing, and Vercel sync.
This commit is contained in:
36
api/chat-stream/error_shape.js
Normal file
36
api/chat-stream/error_shape.js
Normal file
@@ -0,0 +1,36 @@
|
||||
'use strict';
|
||||
|
||||
function writeOpenAIError(res, status, message) {
|
||||
res.statusCode = status;
|
||||
res.setHeader('Content-Type', 'application/json');
|
||||
res.end(
|
||||
JSON.stringify({
|
||||
error: {
|
||||
message,
|
||||
type: openAIErrorType(status),
|
||||
},
|
||||
}),
|
||||
);
|
||||
}
|
||||
|
||||
function openAIErrorType(status) {
|
||||
switch (status) {
|
||||
case 400:
|
||||
return 'invalid_request_error';
|
||||
case 401:
|
||||
return 'authentication_error';
|
||||
case 403:
|
||||
return 'permission_error';
|
||||
case 429:
|
||||
return 'rate_limit_error';
|
||||
case 503:
|
||||
return 'service_unavailable_error';
|
||||
default:
|
||||
return status >= 500 ? 'api_error' : 'invalid_request_error';
|
||||
}
|
||||
}
|
||||
|
||||
module.exports = {
|
||||
writeOpenAIError,
|
||||
openAIErrorType,
|
||||
};
|
||||
214
api/chat-stream/http_internal.js
Normal file
214
api/chat-stream/http_internal.js
Normal file
@@ -0,0 +1,214 @@
|
||||
'use strict';
|
||||
|
||||
const {
|
||||
writeOpenAIError,
|
||||
} = require('./error_shape');
|
||||
|
||||
function setCorsHeaders(res) {
|
||||
res.setHeader('Access-Control-Allow-Origin', '*');
|
||||
res.setHeader('Access-Control-Allow-Methods', 'GET, POST, OPTIONS, PUT, DELETE');
|
||||
res.setHeader(
|
||||
'Access-Control-Allow-Headers',
|
||||
'Content-Type, Authorization, X-API-Key, X-Ds2-Target-Account, X-Vercel-Protection-Bypass',
|
||||
);
|
||||
}
|
||||
|
||||
function header(req, key) {
|
||||
if (!req || !req.headers) {
|
||||
return '';
|
||||
}
|
||||
return asString(req.headers[key.toLowerCase()]);
|
||||
}
|
||||
|
||||
async function readRawBody(req) {
|
||||
if (Buffer.isBuffer(req.body)) {
|
||||
return req.body;
|
||||
}
|
||||
if (typeof req.body === 'string') {
|
||||
return Buffer.from(req.body);
|
||||
}
|
||||
if (req.body && typeof req.body === 'object') {
|
||||
return Buffer.from(JSON.stringify(req.body));
|
||||
}
|
||||
const chunks = [];
|
||||
for await (const chunk of req) {
|
||||
chunks.push(Buffer.isBuffer(chunk) ? chunk : Buffer.from(chunk));
|
||||
}
|
||||
return Buffer.concat(chunks);
|
||||
}
|
||||
|
||||
async function fetchStreamPrepare(req, rawBody) {
|
||||
const url = buildInternalGoURL(req);
|
||||
url.searchParams.set('__stream_prepare', '1');
|
||||
|
||||
const upstream = await fetch(url.toString(), {
|
||||
method: 'POST',
|
||||
headers: buildInternalGoHeaders(req, { withInternalToken: true, withContentType: true }),
|
||||
body: rawBody,
|
||||
});
|
||||
|
||||
const text = await upstream.text();
|
||||
let body = {};
|
||||
try {
|
||||
body = JSON.parse(text || '{}');
|
||||
} catch (_err) {
|
||||
body = {};
|
||||
}
|
||||
|
||||
return {
|
||||
ok: upstream.ok,
|
||||
status: upstream.status,
|
||||
contentType: upstream.headers.get('content-type') || 'application/json',
|
||||
text,
|
||||
body,
|
||||
};
|
||||
}
|
||||
|
||||
function relayPreparedFailure(res, prep) {
|
||||
if (prep.status === 401 && looksLikeVercelAuthPage(prep.text)) {
|
||||
writeOpenAIError(
|
||||
res,
|
||||
401,
|
||||
'Vercel Deployment Protection blocked internal prepare request. Disable protection for this deployment or set VERCEL_AUTOMATION_BYPASS_SECRET.',
|
||||
);
|
||||
return;
|
||||
}
|
||||
res.statusCode = prep.status || 500;
|
||||
res.setHeader('Content-Type', prep.contentType || 'application/json');
|
||||
if (prep.text) {
|
||||
res.end(prep.text);
|
||||
return;
|
||||
}
|
||||
writeOpenAIError(res, prep.status || 500, 'vercel prepare failed');
|
||||
}
|
||||
|
||||
async function safeReadText(resp) {
|
||||
if (!resp) {
|
||||
return '';
|
||||
}
|
||||
try {
|
||||
const text = await resp.text();
|
||||
return text.trim();
|
||||
} catch (_err) {
|
||||
return '';
|
||||
}
|
||||
}
|
||||
|
||||
function internalSecret() {
|
||||
return asString(process.env.DS2API_VERCEL_INTERNAL_SECRET) || asString(process.env.DS2API_ADMIN_KEY) || 'admin';
|
||||
}
|
||||
|
||||
function buildInternalGoURL(req) {
|
||||
const proto = asString(header(req, 'x-forwarded-proto')) || 'https';
|
||||
const host = asString(header(req, 'host'));
|
||||
const url = new URL(`${proto}://${host}${req.url || '/v1/chat/completions'}`);
|
||||
url.searchParams.set('__go', '1');
|
||||
const protectionBypass = resolveProtectionBypass(req);
|
||||
if (protectionBypass) {
|
||||
url.searchParams.set('x-vercel-protection-bypass', protectionBypass);
|
||||
}
|
||||
return url;
|
||||
}
|
||||
|
||||
function buildInternalGoHeaders(req, opts = {}) {
|
||||
const headers = {
|
||||
authorization: asString(header(req, 'authorization')),
|
||||
'x-api-key': asString(header(req, 'x-api-key')),
|
||||
'x-ds2-target-account': asString(header(req, 'x-ds2-target-account')),
|
||||
'x-vercel-protection-bypass': resolveProtectionBypass(req),
|
||||
};
|
||||
if (opts.withInternalToken) {
|
||||
headers['x-ds2-internal-token'] = internalSecret();
|
||||
}
|
||||
if (opts.withContentType) {
|
||||
headers['content-type'] = asString(header(req, 'content-type')) || 'application/json';
|
||||
}
|
||||
return headers;
|
||||
}
|
||||
|
||||
function createLeaseReleaser(req, leaseID) {
|
||||
let released = false;
|
||||
return async () => {
|
||||
if (released || !leaseID) {
|
||||
return;
|
||||
}
|
||||
released = true;
|
||||
try {
|
||||
await releaseStreamLease(req, leaseID);
|
||||
} catch (_err) {
|
||||
// Ignore release errors. Lease TTL cleanup on Go side still prevents permanent leaks.
|
||||
}
|
||||
};
|
||||
}
|
||||
|
||||
async function releaseStreamLease(req, leaseID) {
|
||||
const url = buildInternalGoURL(req);
|
||||
url.searchParams.set('__stream_release', '1');
|
||||
const body = Buffer.from(JSON.stringify({ lease_id: leaseID }));
|
||||
|
||||
const controller = new AbortController();
|
||||
const timeout = setTimeout(() => controller.abort(), 1500);
|
||||
try {
|
||||
await fetch(url.toString(), {
|
||||
method: 'POST',
|
||||
headers: buildInternalGoHeaders(req, { withInternalToken: true, withContentType: true }),
|
||||
body,
|
||||
signal: controller.signal,
|
||||
});
|
||||
} finally {
|
||||
clearTimeout(timeout);
|
||||
}
|
||||
}
|
||||
|
||||
function resolveProtectionBypass(req) {
|
||||
const fromHeader = asString(header(req, 'x-vercel-protection-bypass'));
|
||||
if (fromHeader) {
|
||||
return fromHeader;
|
||||
}
|
||||
return asString(process.env.VERCEL_AUTOMATION_BYPASS_SECRET) || asString(process.env.DS2API_VERCEL_PROTECTION_BYPASS);
|
||||
}
|
||||
|
||||
function looksLikeVercelAuthPage(text) {
|
||||
const body = asString(text).toLowerCase();
|
||||
if (!body) {
|
||||
return false;
|
||||
}
|
||||
return body.includes('authentication required') && body.includes('vercel');
|
||||
}
|
||||
|
||||
function asString(v) {
|
||||
if (typeof v === 'string') {
|
||||
return v.trim();
|
||||
}
|
||||
if (Array.isArray(v)) {
|
||||
return asString(v[0]);
|
||||
}
|
||||
if (v == null) {
|
||||
return '';
|
||||
}
|
||||
return String(v).trim();
|
||||
}
|
||||
|
||||
function isAbortError(err) {
|
||||
if (!err || typeof err !== 'object') {
|
||||
return false;
|
||||
}
|
||||
return err.name === 'AbortError' || err.code === 'ABORT_ERR';
|
||||
}
|
||||
|
||||
module.exports = {
|
||||
setCorsHeaders,
|
||||
header,
|
||||
readRawBody,
|
||||
fetchStreamPrepare,
|
||||
relayPreparedFailure,
|
||||
safeReadText,
|
||||
buildInternalGoURL,
|
||||
buildInternalGoHeaders,
|
||||
createLeaseReleaser,
|
||||
releaseStreamLease,
|
||||
resolveProtectionBypass,
|
||||
looksLikeVercelAuthPage,
|
||||
asString,
|
||||
isAbortError,
|
||||
};
|
||||
88
api/chat-stream/index.js
Normal file
88
api/chat-stream/index.js
Normal file
@@ -0,0 +1,88 @@
|
||||
'use strict';
|
||||
|
||||
const {
|
||||
writeOpenAIError,
|
||||
} = require('./error_shape');
|
||||
const {
|
||||
parseChunkForContent,
|
||||
extractContentRecursive,
|
||||
shouldSkipPath,
|
||||
} = require('./sse_parse');
|
||||
const {
|
||||
resolveToolcallPolicy,
|
||||
normalizePreparedToolNames,
|
||||
boolDefaultTrue,
|
||||
} = require('./toolcall_policy');
|
||||
const {
|
||||
estimateTokens,
|
||||
} = require('./token_usage');
|
||||
const {
|
||||
setCorsHeaders,
|
||||
readRawBody,
|
||||
asString,
|
||||
} = require('./http_internal');
|
||||
const {
|
||||
proxyToGo,
|
||||
} = require('./proxy_go');
|
||||
const {
|
||||
handleVercelStream,
|
||||
} = require('./vercel_stream');
|
||||
|
||||
async function handler(req, res) {
|
||||
setCorsHeaders(res);
|
||||
if (req.method === 'OPTIONS') {
|
||||
res.statusCode = 204;
|
||||
res.end();
|
||||
return;
|
||||
}
|
||||
if (req.method !== 'POST') {
|
||||
writeOpenAIError(res, 405, 'method not allowed');
|
||||
return;
|
||||
}
|
||||
|
||||
const rawBody = await readRawBody(req);
|
||||
|
||||
// Hard guard: only use Node data path for streaming on Vercel runtime.
|
||||
// Any non-Vercel runtime always falls back to Go for full behavior parity.
|
||||
if (!isVercelRuntime()) {
|
||||
await proxyToGo(req, res, rawBody);
|
||||
return;
|
||||
}
|
||||
|
||||
let payload;
|
||||
try {
|
||||
payload = JSON.parse(rawBody.toString('utf8') || '{}');
|
||||
} catch (_err) {
|
||||
writeOpenAIError(res, 400, 'invalid json');
|
||||
return;
|
||||
}
|
||||
|
||||
// Keep all non-stream behavior on Go side to avoid compatibility regressions.
|
||||
if (!toBool(payload.stream)) {
|
||||
await proxyToGo(req, res, rawBody);
|
||||
return;
|
||||
}
|
||||
|
||||
await handleVercelStream(req, res, rawBody, payload);
|
||||
}
|
||||
|
||||
function toBool(v) {
|
||||
return v === true;
|
||||
}
|
||||
|
||||
function isVercelRuntime() {
|
||||
return asString(process.env.VERCEL) !== '' || asString(process.env.NOW_REGION) !== '';
|
||||
}
|
||||
|
||||
module.exports = handler;
|
||||
|
||||
module.exports.__test = {
|
||||
parseChunkForContent,
|
||||
extractContentRecursive,
|
||||
shouldSkipPath,
|
||||
asString,
|
||||
resolveToolcallPolicy,
|
||||
normalizePreparedToolNames,
|
||||
boolDefaultTrue,
|
||||
estimateTokens,
|
||||
};
|
||||
105
api/chat-stream/proxy_go.js
Normal file
105
api/chat-stream/proxy_go.js
Normal file
@@ -0,0 +1,105 @@
|
||||
'use strict';
|
||||
|
||||
const {
|
||||
buildInternalGoURL,
|
||||
buildInternalGoHeaders,
|
||||
isAbortError,
|
||||
} = require('./http_internal');
|
||||
|
||||
async function proxyToGo(req, res, rawBody) {
|
||||
const url = buildInternalGoURL(req);
|
||||
const controller = new AbortController();
|
||||
let clientClosed = false;
|
||||
const markClientClosed = () => {
|
||||
if (clientClosed) {
|
||||
return;
|
||||
}
|
||||
clientClosed = true;
|
||||
controller.abort();
|
||||
};
|
||||
const onReqAborted = () => markClientClosed();
|
||||
const onResClose = () => {
|
||||
if (!res.writableEnded) {
|
||||
markClientClosed();
|
||||
}
|
||||
};
|
||||
req.on('aborted', onReqAborted);
|
||||
res.on('close', onResClose);
|
||||
|
||||
try {
|
||||
let upstream;
|
||||
try {
|
||||
upstream = await fetch(url.toString(), {
|
||||
method: 'POST',
|
||||
headers: buildInternalGoHeaders(req, { withContentType: true }),
|
||||
body: rawBody,
|
||||
signal: controller.signal,
|
||||
});
|
||||
} catch (err) {
|
||||
if (clientClosed || isAbortError(err)) {
|
||||
if (!res.writableEnded) {
|
||||
res.end();
|
||||
}
|
||||
return;
|
||||
}
|
||||
throw err;
|
||||
}
|
||||
if (clientClosed) {
|
||||
if (!res.writableEnded) {
|
||||
res.end();
|
||||
}
|
||||
return;
|
||||
}
|
||||
|
||||
res.statusCode = upstream.status;
|
||||
upstream.headers.forEach((value, key) => {
|
||||
if (key.toLowerCase() === 'content-length') {
|
||||
return;
|
||||
}
|
||||
res.setHeader(key, value);
|
||||
});
|
||||
|
||||
if (!upstream.body || typeof upstream.body.getReader !== 'function') {
|
||||
const bytes = Buffer.from(await upstream.arrayBuffer());
|
||||
res.end(bytes);
|
||||
return;
|
||||
}
|
||||
|
||||
const reader = upstream.body.getReader();
|
||||
try {
|
||||
// eslint-disable-next-line no-constant-condition
|
||||
while (true) {
|
||||
if (clientClosed) {
|
||||
break;
|
||||
}
|
||||
const { value, done } = await reader.read();
|
||||
if (done) {
|
||||
break;
|
||||
}
|
||||
if (value && value.length > 0) {
|
||||
res.write(Buffer.from(value));
|
||||
if (typeof res.flush === 'function') {
|
||||
res.flush();
|
||||
}
|
||||
}
|
||||
}
|
||||
if (!res.writableEnded) {
|
||||
res.end();
|
||||
}
|
||||
} catch (err) {
|
||||
if (!isAbortError(err) && !res.writableEnded) {
|
||||
res.end();
|
||||
}
|
||||
}
|
||||
} finally {
|
||||
req.removeListener('aborted', onReqAborted);
|
||||
res.removeListener('close', onResClose);
|
||||
if (!res.writableEnded) {
|
||||
res.end();
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
module.exports = {
|
||||
proxyToGo,
|
||||
};
|
||||
229
api/chat-stream/sse_parse.js
Normal file
229
api/chat-stream/sse_parse.js
Normal file
@@ -0,0 +1,229 @@
|
||||
'use strict';
|
||||
|
||||
const {
|
||||
SKIP_PATTERNS,
|
||||
SKIP_EXACT_PATHS,
|
||||
} = require('../shared/deepseek-constants');
|
||||
|
||||
function parseChunkForContent(chunk, thinkingEnabled, currentType) {
|
||||
if (!chunk || typeof chunk !== 'object' || !Object.prototype.hasOwnProperty.call(chunk, 'v')) {
|
||||
return { parts: [], finished: false, newType: currentType };
|
||||
}
|
||||
const pathValue = asString(chunk.p);
|
||||
if (shouldSkipPath(pathValue)) {
|
||||
return { parts: [], finished: false, newType: currentType };
|
||||
}
|
||||
if (pathValue === 'response/status' && asString(chunk.v) === 'FINISHED') {
|
||||
return { parts: [], finished: true, newType: currentType };
|
||||
}
|
||||
|
||||
let newType = currentType;
|
||||
const parts = [];
|
||||
|
||||
if (pathValue === 'response/fragments' && asString(chunk.o).toUpperCase() === 'APPEND' && Array.isArray(chunk.v)) {
|
||||
for (const frag of chunk.v) {
|
||||
if (!frag || typeof frag !== 'object') {
|
||||
continue;
|
||||
}
|
||||
const fragType = asString(frag.type).toUpperCase();
|
||||
const content = asString(frag.content);
|
||||
if (!content) {
|
||||
continue;
|
||||
}
|
||||
if (fragType === 'THINK' || fragType === 'THINKING') {
|
||||
newType = 'thinking';
|
||||
parts.push({ text: content, type: 'thinking' });
|
||||
} else if (fragType === 'RESPONSE') {
|
||||
newType = 'text';
|
||||
parts.push({ text: content, type: 'text' });
|
||||
} else {
|
||||
parts.push({ text: content, type: 'text' });
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
if (pathValue === 'response' && Array.isArray(chunk.v)) {
|
||||
for (const item of chunk.v) {
|
||||
if (!item || typeof item !== 'object') {
|
||||
continue;
|
||||
}
|
||||
if (item.p === 'fragments' && item.o === 'APPEND' && Array.isArray(item.v)) {
|
||||
for (const frag of item.v) {
|
||||
const fragType = asString(frag && frag.type).toUpperCase();
|
||||
if (fragType === 'THINK' || fragType === 'THINKING') {
|
||||
newType = 'thinking';
|
||||
} else if (fragType === 'RESPONSE') {
|
||||
newType = 'text';
|
||||
}
|
||||
}
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
let partType = 'text';
|
||||
if (pathValue === 'response/thinking_content') {
|
||||
partType = 'thinking';
|
||||
} else if (pathValue === 'response/content') {
|
||||
partType = 'text';
|
||||
} else if (pathValue.includes('response/fragments') && pathValue.includes('/content')) {
|
||||
partType = newType;
|
||||
} else if (!pathValue && thinkingEnabled) {
|
||||
partType = newType;
|
||||
}
|
||||
|
||||
const val = chunk.v;
|
||||
if (typeof val === 'string') {
|
||||
if (val === 'FINISHED' && (!pathValue || pathValue === 'status')) {
|
||||
return { parts: [], finished: true, newType };
|
||||
}
|
||||
if (val) {
|
||||
parts.push({ text: val, type: partType });
|
||||
}
|
||||
return { parts, finished: false, newType };
|
||||
}
|
||||
|
||||
if (Array.isArray(val)) {
|
||||
const extracted = extractContentRecursive(val, partType);
|
||||
if (extracted.finished) {
|
||||
return { parts: [], finished: true, newType };
|
||||
}
|
||||
parts.push(...extracted.parts);
|
||||
return { parts, finished: false, newType };
|
||||
}
|
||||
|
||||
if (val && typeof val === 'object') {
|
||||
const resp = val.response && typeof val.response === 'object' ? val.response : val;
|
||||
if (Array.isArray(resp.fragments)) {
|
||||
for (const frag of resp.fragments) {
|
||||
if (!frag || typeof frag !== 'object') {
|
||||
continue;
|
||||
}
|
||||
const content = asString(frag.content);
|
||||
if (!content) {
|
||||
continue;
|
||||
}
|
||||
const t = asString(frag.type).toUpperCase();
|
||||
if (t === 'THINK' || t === 'THINKING') {
|
||||
newType = 'thinking';
|
||||
parts.push({ text: content, type: 'thinking' });
|
||||
} else if (t === 'RESPONSE') {
|
||||
newType = 'text';
|
||||
parts.push({ text: content, type: 'text' });
|
||||
} else {
|
||||
parts.push({ text: content, type: partType });
|
||||
}
|
||||
}
|
||||
}
|
||||
}
|
||||
return { parts, finished: false, newType };
|
||||
}
|
||||
|
||||
function extractContentRecursive(items, defaultType) {
|
||||
const parts = [];
|
||||
for (const it of items) {
|
||||
if (!it || typeof it !== 'object') {
|
||||
continue;
|
||||
}
|
||||
if (!Object.prototype.hasOwnProperty.call(it, 'v')) {
|
||||
continue;
|
||||
}
|
||||
const itemPath = asString(it.p);
|
||||
const itemV = it.v;
|
||||
if (itemPath === 'status' && asString(itemV) === 'FINISHED') {
|
||||
return { parts: [], finished: true };
|
||||
}
|
||||
if (shouldSkipPath(itemPath)) {
|
||||
continue;
|
||||
}
|
||||
const content = asString(it.content);
|
||||
if (content) {
|
||||
const typeName = asString(it.type).toUpperCase();
|
||||
if (typeName === 'THINK' || typeName === 'THINKING') {
|
||||
parts.push({ text: content, type: 'thinking' });
|
||||
} else if (typeName === 'RESPONSE') {
|
||||
parts.push({ text: content, type: 'text' });
|
||||
} else {
|
||||
parts.push({ text: content, type: defaultType });
|
||||
}
|
||||
continue;
|
||||
}
|
||||
|
||||
let partType = defaultType;
|
||||
if (itemPath.includes('thinking')) {
|
||||
partType = 'thinking';
|
||||
} else if (itemPath.includes('content') || itemPath === 'response' || itemPath === 'fragments') {
|
||||
partType = 'text';
|
||||
}
|
||||
|
||||
if (typeof itemV === 'string') {
|
||||
if (itemV && itemV !== 'FINISHED') {
|
||||
parts.push({ text: itemV, type: partType });
|
||||
}
|
||||
continue;
|
||||
}
|
||||
|
||||
if (!Array.isArray(itemV)) {
|
||||
continue;
|
||||
}
|
||||
for (const inner of itemV) {
|
||||
if (typeof inner === 'string') {
|
||||
if (inner) {
|
||||
parts.push({ text: inner, type: partType });
|
||||
}
|
||||
continue;
|
||||
}
|
||||
if (!inner || typeof inner !== 'object') {
|
||||
continue;
|
||||
}
|
||||
const ct = asString(inner.content);
|
||||
if (!ct) {
|
||||
continue;
|
||||
}
|
||||
const typeName = asString(inner.type).toUpperCase();
|
||||
if (typeName === 'THINK' || typeName === 'THINKING') {
|
||||
parts.push({ text: ct, type: 'thinking' });
|
||||
} else if (typeName === 'RESPONSE') {
|
||||
parts.push({ text: ct, type: 'text' });
|
||||
} else {
|
||||
parts.push({ text: ct, type: partType });
|
||||
}
|
||||
}
|
||||
}
|
||||
return { parts, finished: false };
|
||||
}
|
||||
|
||||
function shouldSkipPath(pathValue) {
|
||||
if (SKIP_EXACT_PATHS.has(pathValue)) {
|
||||
return true;
|
||||
}
|
||||
for (const p of SKIP_PATTERNS) {
|
||||
if (pathValue.includes(p)) {
|
||||
return true;
|
||||
}
|
||||
}
|
||||
return false;
|
||||
}
|
||||
|
||||
function isCitation(text) {
|
||||
return asString(text).trim().startsWith('[citation:');
|
||||
}
|
||||
|
||||
function asString(v) {
|
||||
if (typeof v === 'string') {
|
||||
return v.trim();
|
||||
}
|
||||
if (Array.isArray(v)) {
|
||||
return asString(v[0]);
|
||||
}
|
||||
if (v == null) {
|
||||
return '';
|
||||
}
|
||||
return String(v).trim();
|
||||
}
|
||||
|
||||
module.exports = {
|
||||
parseChunkForContent,
|
||||
extractContentRecursive,
|
||||
shouldSkipPath,
|
||||
isCitation,
|
||||
};
|
||||
39
api/chat-stream/stream_emitter.js
Normal file
39
api/chat-stream/stream_emitter.js
Normal file
@@ -0,0 +1,39 @@
|
||||
'use strict';
|
||||
|
||||
function createChatCompletionEmitter({ res, sessionID, created, model, isClosed }) {
|
||||
let firstChunkSent = false;
|
||||
|
||||
const sendFrame = (obj) => {
|
||||
if (isClosed() || res.writableEnded || res.destroyed) {
|
||||
return;
|
||||
}
|
||||
res.write(`data: ${JSON.stringify(obj)}\n\n`);
|
||||
if (typeof res.flush === 'function') {
|
||||
res.flush();
|
||||
}
|
||||
};
|
||||
|
||||
const sendDeltaFrame = (delta) => {
|
||||
const payloadDelta = { ...delta };
|
||||
if (!firstChunkSent) {
|
||||
payloadDelta.role = 'assistant';
|
||||
firstChunkSent = true;
|
||||
}
|
||||
sendFrame({
|
||||
id: sessionID,
|
||||
object: 'chat.completion.chunk',
|
||||
created,
|
||||
model,
|
||||
choices: [{ delta: payloadDelta, index: 0 }],
|
||||
});
|
||||
};
|
||||
|
||||
return {
|
||||
sendFrame,
|
||||
sendDeltaFrame,
|
||||
};
|
||||
}
|
||||
|
||||
module.exports = {
|
||||
createChatCompletionEmitter,
|
||||
};
|
||||
51
api/chat-stream/token_usage.js
Normal file
51
api/chat-stream/token_usage.js
Normal file
@@ -0,0 +1,51 @@
|
||||
'use strict';
|
||||
|
||||
function buildUsage(prompt, thinking, output) {
|
||||
const promptTokens = estimateTokens(prompt);
|
||||
const reasoningTokens = estimateTokens(thinking);
|
||||
const completionTokens = estimateTokens(output);
|
||||
return {
|
||||
prompt_tokens: promptTokens,
|
||||
completion_tokens: reasoningTokens + completionTokens,
|
||||
total_tokens: promptTokens + reasoningTokens + completionTokens,
|
||||
completion_tokens_details: {
|
||||
reasoning_tokens: reasoningTokens,
|
||||
},
|
||||
};
|
||||
}
|
||||
|
||||
function estimateTokens(text) {
|
||||
const t = asString(text);
|
||||
if (!t) {
|
||||
return 0;
|
||||
}
|
||||
let asciiChars = 0;
|
||||
let nonASCIIChars = 0;
|
||||
for (const ch of Array.from(t)) {
|
||||
if (ch.charCodeAt(0) < 128) {
|
||||
asciiChars += 1;
|
||||
} else {
|
||||
nonASCIIChars += 1;
|
||||
}
|
||||
}
|
||||
const n = Math.floor(asciiChars / 4) + Math.floor((nonASCIIChars * 10 + 7) / 13);
|
||||
return n < 1 ? 1 : n;
|
||||
}
|
||||
|
||||
function asString(v) {
|
||||
if (typeof v === 'string') {
|
||||
return v.trim();
|
||||
}
|
||||
if (Array.isArray(v)) {
|
||||
return asString(v[0]);
|
||||
}
|
||||
if (v == null) {
|
||||
return '';
|
||||
}
|
||||
return String(v).trim();
|
||||
}
|
||||
|
||||
module.exports = {
|
||||
buildUsage,
|
||||
estimateTokens,
|
||||
};
|
||||
107
api/chat-stream/toolcall_policy.js
Normal file
107
api/chat-stream/toolcall_policy.js
Normal file
@@ -0,0 +1,107 @@
|
||||
'use strict';
|
||||
|
||||
const crypto = require('crypto');
|
||||
|
||||
const {
|
||||
extractToolNames,
|
||||
} = require('../helpers/stream-tool-sieve');
|
||||
|
||||
function resolveToolcallPolicy(prepBody, payloadTools) {
|
||||
const preparedToolNames = normalizePreparedToolNames(prepBody && prepBody.tool_names);
|
||||
const toolNames = preparedToolNames.length > 0 ? preparedToolNames : extractToolNames(payloadTools);
|
||||
const featureMatchEnabled = boolDefaultTrue(prepBody && prepBody.toolcall_feature_match);
|
||||
const emitEarlyToolDeltas = boolDefaultTrue(prepBody && prepBody.toolcall_early_emit_high);
|
||||
return {
|
||||
toolNames,
|
||||
toolSieveEnabled: toolNames.length > 0 && featureMatchEnabled,
|
||||
emitEarlyToolDeltas,
|
||||
};
|
||||
}
|
||||
|
||||
function normalizePreparedToolNames(v) {
|
||||
if (!Array.isArray(v) || v.length === 0) {
|
||||
return [];
|
||||
}
|
||||
const out = [];
|
||||
for (const item of v) {
|
||||
const name = asString(item);
|
||||
if (!name) {
|
||||
continue;
|
||||
}
|
||||
out.push(name);
|
||||
}
|
||||
return out;
|
||||
}
|
||||
|
||||
function boolDefaultTrue(v) {
|
||||
return v !== false;
|
||||
}
|
||||
|
||||
function formatIncrementalToolCallDeltas(deltas, idStore) {
|
||||
if (!Array.isArray(deltas) || deltas.length === 0) {
|
||||
return [];
|
||||
}
|
||||
const out = [];
|
||||
for (const d of deltas) {
|
||||
if (!d || typeof d !== 'object') {
|
||||
continue;
|
||||
}
|
||||
const index = Number.isInteger(d.index) ? d.index : 0;
|
||||
const id = ensureStreamToolCallID(idStore, index);
|
||||
const item = {
|
||||
index,
|
||||
id,
|
||||
type: 'function',
|
||||
};
|
||||
const fn = {};
|
||||
if (asString(d.name)) {
|
||||
fn.name = asString(d.name);
|
||||
}
|
||||
if (typeof d.arguments === 'string' && d.arguments !== '') {
|
||||
fn.arguments = d.arguments;
|
||||
}
|
||||
if (Object.keys(fn).length > 0) {
|
||||
item.function = fn;
|
||||
}
|
||||
out.push(item);
|
||||
}
|
||||
return out;
|
||||
}
|
||||
|
||||
function ensureStreamToolCallID(idStore, index) {
|
||||
const key = Number.isInteger(index) ? index : 0;
|
||||
const existing = idStore.get(key);
|
||||
if (existing) {
|
||||
return existing;
|
||||
}
|
||||
const next = `call_${newCallID()}`;
|
||||
idStore.set(key, next);
|
||||
return next;
|
||||
}
|
||||
|
||||
function newCallID() {
|
||||
if (typeof crypto.randomUUID === 'function') {
|
||||
return crypto.randomUUID().replace(/-/g, '');
|
||||
}
|
||||
return `${Date.now()}${Math.floor(Math.random() * 1e9)}`;
|
||||
}
|
||||
|
||||
function asString(v) {
|
||||
if (typeof v === 'string') {
|
||||
return v.trim();
|
||||
}
|
||||
if (Array.isArray(v)) {
|
||||
return asString(v[0]);
|
||||
}
|
||||
if (v == null) {
|
||||
return '';
|
||||
}
|
||||
return String(v).trim();
|
||||
}
|
||||
|
||||
module.exports = {
|
||||
resolveToolcallPolicy,
|
||||
normalizePreparedToolNames,
|
||||
boolDefaultTrue,
|
||||
formatIncrementalToolCallDeltas,
|
||||
};
|
||||
297
api/chat-stream/vercel_stream.js
Normal file
297
api/chat-stream/vercel_stream.js
Normal file
@@ -0,0 +1,297 @@
|
||||
'use strict';
|
||||
|
||||
const {
|
||||
extractToolNames,
|
||||
createToolSieveState,
|
||||
processToolSieveChunk,
|
||||
flushToolSieve,
|
||||
parseToolCalls,
|
||||
formatOpenAIStreamToolCalls,
|
||||
} = require('../helpers/stream-tool-sieve');
|
||||
const {
|
||||
BASE_HEADERS,
|
||||
} = require('../shared/deepseek-constants');
|
||||
|
||||
const {
|
||||
writeOpenAIError,
|
||||
} = require('./error_shape');
|
||||
const {
|
||||
parseChunkForContent,
|
||||
isCitation,
|
||||
} = require('./sse_parse');
|
||||
const {
|
||||
buildUsage,
|
||||
} = require('./token_usage');
|
||||
const {
|
||||
resolveToolcallPolicy,
|
||||
formatIncrementalToolCallDeltas,
|
||||
} = require('./toolcall_policy');
|
||||
const {
|
||||
createChatCompletionEmitter,
|
||||
} = require('./stream_emitter');
|
||||
const {
|
||||
asString,
|
||||
isAbortError,
|
||||
fetchStreamPrepare,
|
||||
relayPreparedFailure,
|
||||
safeReadText,
|
||||
createLeaseReleaser,
|
||||
} = require('./http_internal');
|
||||
|
||||
const DEEPSEEK_COMPLETION_URL = 'https://chat.deepseek.com/api/v0/chat/completion';
|
||||
|
||||
async function handleVercelStream(req, res, rawBody, payload) {
|
||||
const prep = await fetchStreamPrepare(req, rawBody);
|
||||
if (!prep.ok) {
|
||||
relayPreparedFailure(res, prep);
|
||||
return;
|
||||
}
|
||||
|
||||
const model = asString(prep.body.model) || asString(payload.model);
|
||||
const sessionID = asString(prep.body.session_id) || `chatcmpl-${Date.now()}`;
|
||||
const leaseID = asString(prep.body.lease_id);
|
||||
const deepseekToken = asString(prep.body.deepseek_token);
|
||||
const powHeader = asString(prep.body.pow_header);
|
||||
const completionPayload = prep.body.payload && typeof prep.body.payload === 'object' ? prep.body.payload : null;
|
||||
const finalPrompt = asString(prep.body.final_prompt);
|
||||
const thinkingEnabled = toBool(prep.body.thinking_enabled);
|
||||
const searchEnabled = toBool(prep.body.search_enabled);
|
||||
const toolPolicy = resolveToolcallPolicy(prep.body, payload.tools);
|
||||
const toolNames = toolPolicy.toolNames;
|
||||
|
||||
if (!model || !leaseID || !deepseekToken || !powHeader || !completionPayload) {
|
||||
writeOpenAIError(res, 500, 'invalid vercel prepare response');
|
||||
return;
|
||||
}
|
||||
|
||||
const releaseLease = createLeaseReleaser(req, leaseID);
|
||||
const upstreamController = new AbortController();
|
||||
let clientClosed = false;
|
||||
let reader = null;
|
||||
const markClientClosed = () => {
|
||||
if (clientClosed) {
|
||||
return;
|
||||
}
|
||||
clientClosed = true;
|
||||
upstreamController.abort();
|
||||
if (reader && typeof reader.cancel === 'function') {
|
||||
Promise.resolve(reader.cancel()).catch(() => {});
|
||||
}
|
||||
};
|
||||
const onReqAborted = () => markClientClosed();
|
||||
const onResClose = () => {
|
||||
if (!res.writableEnded) {
|
||||
markClientClosed();
|
||||
}
|
||||
};
|
||||
req.on('aborted', onReqAborted);
|
||||
res.on('close', onResClose);
|
||||
|
||||
try {
|
||||
let completionRes;
|
||||
try {
|
||||
completionRes = await fetch(DEEPSEEK_COMPLETION_URL, {
|
||||
method: 'POST',
|
||||
headers: {
|
||||
...BASE_HEADERS,
|
||||
authorization: `Bearer ${deepseekToken}`,
|
||||
'x-ds-pow-response': powHeader,
|
||||
},
|
||||
body: JSON.stringify(completionPayload),
|
||||
signal: upstreamController.signal,
|
||||
});
|
||||
} catch (err) {
|
||||
if (clientClosed || isAbortError(err)) {
|
||||
return;
|
||||
}
|
||||
throw err;
|
||||
}
|
||||
if (clientClosed) {
|
||||
return;
|
||||
}
|
||||
|
||||
if (!completionRes.ok || !completionRes.body) {
|
||||
const detail = await safeReadText(completionRes);
|
||||
writeOpenAIError(res, 500, detail ? `Failed to get completion: ${detail}` : 'Failed to get completion.');
|
||||
return;
|
||||
}
|
||||
|
||||
res.statusCode = 200;
|
||||
res.setHeader('Content-Type', 'text/event-stream');
|
||||
res.setHeader('Cache-Control', 'no-cache, no-transform');
|
||||
res.setHeader('Connection', 'keep-alive');
|
||||
res.setHeader('X-Accel-Buffering', 'no');
|
||||
if (typeof res.flushHeaders === 'function') {
|
||||
res.flushHeaders();
|
||||
}
|
||||
|
||||
const created = Math.floor(Date.now() / 1000);
|
||||
let currentType = thinkingEnabled ? 'thinking' : 'text';
|
||||
let thinkingText = '';
|
||||
let outputText = '';
|
||||
const toolSieveEnabled = toolPolicy.toolSieveEnabled;
|
||||
const emitEarlyToolDeltas = toolPolicy.emitEarlyToolDeltas;
|
||||
const toolSieveState = createToolSieveState();
|
||||
let toolCallsEmitted = false;
|
||||
const streamToolCallIDs = new Map();
|
||||
const decoder = new TextDecoder();
|
||||
reader = completionRes.body.getReader();
|
||||
let buffered = '';
|
||||
let ended = false;
|
||||
const { sendFrame, sendDeltaFrame } = createChatCompletionEmitter({
|
||||
res,
|
||||
sessionID,
|
||||
created,
|
||||
model,
|
||||
isClosed: () => clientClosed,
|
||||
});
|
||||
|
||||
const finish = async (reason) => {
|
||||
if (ended) {
|
||||
return;
|
||||
}
|
||||
ended = true;
|
||||
if (clientClosed || res.writableEnded || res.destroyed) {
|
||||
await releaseLease();
|
||||
return;
|
||||
}
|
||||
const detected = parseToolCalls(outputText, toolNames);
|
||||
if (detected.length > 0 && !toolCallsEmitted) {
|
||||
toolCallsEmitted = true;
|
||||
sendDeltaFrame({ tool_calls: formatOpenAIStreamToolCalls(detected) });
|
||||
} else if (toolSieveEnabled) {
|
||||
const tailEvents = flushToolSieve(toolSieveState, toolNames);
|
||||
for (const evt of tailEvents) {
|
||||
if (evt.text) {
|
||||
sendDeltaFrame({ content: evt.text });
|
||||
}
|
||||
}
|
||||
}
|
||||
if (detected.length > 0 || toolCallsEmitted) {
|
||||
reason = 'tool_calls';
|
||||
}
|
||||
sendFrame({
|
||||
id: sessionID,
|
||||
object: 'chat.completion.chunk',
|
||||
created,
|
||||
model,
|
||||
choices: [{ delta: {}, index: 0, finish_reason: reason }],
|
||||
usage: buildUsage(finalPrompt, thinkingText, outputText),
|
||||
});
|
||||
if (!res.writableEnded && !res.destroyed) {
|
||||
res.write('data: [DONE]\n\n');
|
||||
}
|
||||
await releaseLease();
|
||||
if (!res.writableEnded && !res.destroyed) {
|
||||
res.end();
|
||||
}
|
||||
};
|
||||
|
||||
try {
|
||||
// eslint-disable-next-line no-constant-condition
|
||||
while (true) {
|
||||
if (clientClosed) {
|
||||
await finish('stop');
|
||||
return;
|
||||
}
|
||||
const { value, done } = await reader.read();
|
||||
if (done) {
|
||||
break;
|
||||
}
|
||||
buffered += decoder.decode(value, { stream: true });
|
||||
const lines = buffered.split('\n');
|
||||
buffered = lines.pop() || '';
|
||||
|
||||
for (const rawLine of lines) {
|
||||
const line = rawLine.trim();
|
||||
if (!line.startsWith('data:')) {
|
||||
continue;
|
||||
}
|
||||
const dataStr = line.slice(5).trim();
|
||||
if (!dataStr) {
|
||||
continue;
|
||||
}
|
||||
if (dataStr === '[DONE]') {
|
||||
await finish('stop');
|
||||
return;
|
||||
}
|
||||
let chunk;
|
||||
try {
|
||||
chunk = JSON.parse(dataStr);
|
||||
} catch (_err) {
|
||||
continue;
|
||||
}
|
||||
if (chunk.error || chunk.code === 'content_filter') {
|
||||
await finish('content_filter');
|
||||
return;
|
||||
}
|
||||
const parsed = parseChunkForContent(chunk, thinkingEnabled, currentType);
|
||||
currentType = parsed.newType;
|
||||
if (parsed.finished) {
|
||||
await finish('stop');
|
||||
return;
|
||||
}
|
||||
|
||||
for (const p of parsed.parts) {
|
||||
if (!p.text) {
|
||||
continue;
|
||||
}
|
||||
if (searchEnabled && isCitation(p.text)) {
|
||||
continue;
|
||||
}
|
||||
if (p.type === 'thinking') {
|
||||
if (thinkingEnabled) {
|
||||
thinkingText += p.text;
|
||||
sendDeltaFrame({ reasoning_content: p.text });
|
||||
}
|
||||
} else {
|
||||
outputText += p.text;
|
||||
if (!toolSieveEnabled) {
|
||||
sendDeltaFrame({ content: p.text });
|
||||
continue;
|
||||
}
|
||||
const events = processToolSieveChunk(toolSieveState, p.text, toolNames);
|
||||
for (const evt of events) {
|
||||
if (evt.type === 'tool_call_deltas' && Array.isArray(evt.deltas) && evt.deltas.length > 0) {
|
||||
if (!emitEarlyToolDeltas) {
|
||||
continue;
|
||||
}
|
||||
toolCallsEmitted = true;
|
||||
sendDeltaFrame({ tool_calls: formatIncrementalToolCallDeltas(evt.deltas, streamToolCallIDs) });
|
||||
continue;
|
||||
}
|
||||
if (evt.type === 'tool_calls') {
|
||||
toolCallsEmitted = true;
|
||||
sendDeltaFrame({ tool_calls: formatOpenAIStreamToolCalls(evt.calls) });
|
||||
continue;
|
||||
}
|
||||
if (evt.text) {
|
||||
sendDeltaFrame({ content: evt.text });
|
||||
}
|
||||
}
|
||||
}
|
||||
}
|
||||
}
|
||||
}
|
||||
await finish('stop');
|
||||
} catch (err) {
|
||||
if (clientClosed || isAbortError(err)) {
|
||||
await finish('stop');
|
||||
return;
|
||||
}
|
||||
await finish('stop');
|
||||
}
|
||||
} finally {
|
||||
req.removeListener('aborted', onReqAborted);
|
||||
res.removeListener('close', onResClose);
|
||||
await releaseLease();
|
||||
}
|
||||
}
|
||||
|
||||
function toBool(v) {
|
||||
return v === true;
|
||||
}
|
||||
|
||||
module.exports = {
|
||||
handleVercelStream,
|
||||
};
|
||||
Reference in New Issue
Block a user