mslxdff 0.1.67 → 0.1.68

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "mslxdff",
-  "version": "0.1.67",
+  "version": "0.1.68",
   "description": "测试项目，请勿使用。",
   "type": "module",
   "bin": {

package/src/bench/probe.js

@@ -0,0 +1,68 @@
+import { joinUrl } from "../providers/base.js";
+
+function normalizeModelsPayload(json) {
+  if (!json) return [];
+  if (Array.isArray(json)) return json;
+  if (Array.isArray(json.data)) return json.data;
+  if (Array.isArray(json.models)) return json.models;
+  if (json.data && typeof json.data === "object" && Array.isArray(json.data.data)) return json.data.data;
+  return [];
+}
+
+function toModelId(m) {
+  if (!m) return "";
+  if (typeof m === "string") return m;
+  if (typeof m.id === "string") return m.id;
+  if (typeof m.model === "string") return m.model;
+  if (typeof m.name === "string") return m.name;
+  return "";
+}
+
+export async function probeModels({
+  baseUrl,
+  modelsPath,
+  chatPath,
+  headers = {},
+  fetchImpl = globalThis.fetch,
+  timeoutMs = 8000,
+} = {}) {
+  const base = String(baseUrl || "").replace(/\/+$/, "");
+  if (!base) return { ok: false, error: "missing baseUrl", tried: [], data: [] };
+  const candidates = [];
+  const seen = new Set();
+  const push = (p) => {
+    const s = String(p || "").trim();
+    if (!s) return;
+    const norm = s.startsWith("/") ? s : `/${s}`;
+    if (!seen.has(norm)) { seen.add(norm); candidates.push(norm); }
+  };
+  if (modelsPath) push(modelsPath);
+  // 通用回退：/v1/models 与 /models
+  push("/v1/models");
+  push("/models");
+  // workbuddy 异形已在 defaultModelsPath 注入，若仍未命中则 candidates 已含
+  const tried = [];
+  let lastError = "";
+  for (const p of candidates) {
+    const url = joinUrl(base, p);
+    tried.push(url);
+    try {
+      const controller = new AbortController();
+      const t = setTimeout(() => controller.abort(new Error(`probe timeout ${timeoutMs}ms`)), timeoutMs);
+      let res;
+      try {
+        res = await fetchImpl(url, { headers, signal: controller.signal });
+      } finally { clearTimeout(t); }
+      if (res instanceof Error) { lastError = res.message || String(res); continue; }
+      if (!res.ok) { lastError = `HTTP ${res.status}`; continue; }
+      let json = {};
+      try { json = await res.json(); } catch { json = {}; }
+      const raw = normalizeModelsPayload(json);
+      const data = raw.map((m) => ({ id: toModelId(m), raw: m })).filter((x) => x.id).map((x) => ({ id: x.id, ...x.raw }));
+      return { ok: true, data, tried, url, rawCount: raw.length };
+    } catch (e) {
+      lastError = e?.message || String(e);
+    }
+  }
+  return { ok: false, error: lastError || "all probes failed", tried, data: [] };
+}

package/src/bench/report.js

@@ -0,0 +1,56 @@
+export function sortResults(list) {
+  const arr = [...(list || [])];
+  arr.sort((a, b) => {
+    if (a.ok !== b.ok) return a.ok ? -1 : 1;
+    if (a.ok && b.ok) {
+      const at = a.ttfbMs ?? a.totalMs ?? 1e9;
+      const bt = b.ttfbMs ?? b.totalMs ?? 1e9;
+      if (at !== bt) return at - bt;
+      const ap = a.tps ?? a.charsPerSec ?? -1;
+      const bp = b.tps ?? b.charsPerSec ?? -1;
+      return bp - ap;
+    }
+    return 0;
+  });
+  return arr;
+}
+
+function pad(s, n, align = "left") {
+  const str = String(s ?? "");
+  if (str.length >= n) return str.slice(0, n);
+  const d = n - str.length;
+  if (align === "right") return " ".repeat(d) + str;
+  return str + " ".repeat(d);
+}
+
+export function formatReport(results, { json = false } = {}) {
+  const sorted = sortResults(results);
+  const winner = sorted.find((r) => r.ok) || null;
+  if (json) {
+    return { text: JSON.stringify(sorted, null, 2), json: sorted, winner, sorted };
+  }
+  const lines = [];
+  lines.push("模型                          状态     TTFB   总耗时  速度        tokens  备注");
+  lines.push("─".repeat(84));
+  for (const r of sorted) {
+    const isWin = winner && r.id === winner.id;
+    const mark = isWin ? "*" : " ";
+    const id = pad((isWin ? "* " : "  ") + r.id, 28);
+    const label = pad(r.label || (r.ok ? "成功" : "失败"), 8);
+    const ttfb = pad(r.ttfbMs != null ? `${r.ttfbMs}ms` : "—", 6, "right");
+    const total = pad(r.totalMs != null ? `${r.totalMs}ms` : "—", 7, "right");
+    let speed = "—";
+    if (r.tps != null) speed = `${r.tps} t/s`;
+    else if (r.charsPerSec != null) speed = `${r.charsPerSec} 字/秒`;
+    speed = pad(speed, 11, "right");
+    const tok = r.tokens?.completion != null ? String(r.tokens.completion) : r.chars != null ? String(r.chars) : "—";
+    const note = r.ok ? "" : (r.error || "").slice(0, 28);
+    lines.push(`${mark}${id} ${label} ${ttfb} ${total} ${speed} ${pad(tok, 6, "right")}  ${note}`);
+  }
+  lines.push("─".repeat(84));
+  if (winner) lines.push(`最快：${winner.id}  TTFB ${winner.ttfbMs}ms  总 ${winner.totalMs}ms  ${winner.tps != null ? `${winner.tps} t/s` : `${winner.charsPerSec ?? "—"} 字/秒`}`);
+  else lines.push("无可用模型（均失败）");
+  const failed = sorted.filter((r) => !r.ok).length;
+  lines.push(`完成：${sorted.length} 个，成功 ${sorted.length - failed}，失败 ${failed}`);
+  return { text: lines.join("\n"), json: sorted, winner, sorted };
+}

package/src/bench/runner.js

@@ -0,0 +1,74 @@
+import { joinUrl } from "../providers/base.js";
+import { computeMetrics, extractUsageFromJson } from "../metrics.js";
+
+function classifyError(status, bodyText) {
+  const t = String(bodyText || "").slice(0, 300);
+  if (status === 401) return { label: "鉴权失败", retryable: false };
+  if (status === 402 || /insufficient balance/i.test(t)) return { label: "余额不足", retryable: false };
+  if (status === 403) return { label: /insufficient/i.test(t) ? "余额不足" : "鉴权失败", retryable: false };
+  if (status === 429) return { label: "限流", retryable: true };
+  if (status >= 500) return { label: `上游错误 ${status}`, retryable: true };
+  if (status === 404) return { label: "模型不存在", retryable: false };
+  return { label: `HTTP ${status}`, retryable: false };
+}
+
+export async function runOne({
+  baseUrl,
+  chatPath = "/v1/chat/completions",
+  model,
+  providerId,
+  apiKey,
+  headers = {},
+  prompt = "hi",
+  maxTokens = 32,
+  timeoutMs = 30000,
+  fetchImpl = globalThis.fetch,
+  clock = Date.now,
+} = {}) {
+  const started = clock();
+  const t0 = typeof performance !== "undefined" && performance.now ? performance.now() : started;
+  if (!baseUrl) return { id: model, ok: false, error: "missing baseUrl", label: "配置错误", ttfbMs: null, totalMs: 0, tps: null, charsPerSec: null, tokens: null };
+  if (!model) return { id: model, ok: false, error: "missing model", label: "配置错误", ttfbMs: null, totalMs: 0, tps: null, charsPerSec: null, tokens: null };
+  if (!apiKey) return { id: model, ok: false, error: "missing apiKey", label: "未配置 Key", ttfbMs: null, totalMs: 0, tps: null, charsPerSec: null, tokens: null };
+  const url = joinUrl(String(baseUrl).replace(/\/+$/, ""), chatPath);
+  const body = { model: String(model).split("/").pop(), stream: false, messages: [{ role: "user", content: prompt }], max_tokens: maxTokens };
+  // workbuddy 需额外 workbuddy 透传头由调用方 headers 注入；此处直接透传
+  const finalHeaders = { "Content-Type": "application/json", Accept: "application/json", ...headers };
+  if (apiKey && !finalHeaders.Authorization) finalHeaders.Authorization = `Bearer ${apiKey}`;
+
+  let ttfbMs = null;
+  let res;
+  try {
+    const controller = new AbortController();
+    const timer = setTimeout(() => controller.abort(new Error(`timeout ${timeoutMs}ms`)), timeoutMs);
+    const fetchStart = typeof performance !== "undefined" && performance.now ? performance.now() : clock();
+    try {
+      res = await fetchImpl(url, { method: "POST", headers: finalHeaders, body: JSON.stringify(body), signal: controller.signal });
+    } finally { clearTimeout(timer); }
+    ttfbMs = Math.round((typeof performance !== "undefined" && performance.now ? performance.now() : clock()) - fetchStart);
+    if (res instanceof Error) throw res;
+    const totalMs = Math.round((typeof performance !== "undefined" && performance.now ? performance.now() : clock()) - t0);
+    if (!res.ok) {
+      let txt = "";
+      try { txt = await res.text(); } catch {}
+      const cls = classifyError(res.status, txt);
+      return { id: model, providerId, ok: false, status: res.status, error: txt.slice(0, 300) || `HTTP ${res.status}`, label: cls.label, ttfbMs, totalMs, tps: null, charsPerSec: null, tokens: null };
+    }
+    let json = {};
+    let txt = "";
+    try { txt = await res.text(); json = JSON.parse(txt); } catch { json = {}; }
+    const usage = extractUsageFromJson(json);
+    const content = json?.choices?.[0]?.message?.content || json?.choices?.[0]?.text || txt || "";
+    const chars = typeof content === "string" ? content.length : 0;
+    const promptTokens = usage?.prompt_tokens ?? null;
+    const completionTokens = usage?.completion_tokens ?? null;
+    const { tps, charsPerSec } = computeMetrics({ ttfbMs, totalMs, promptTokens, completionTokens, chars });
+    const totalTokens = usage?.total_tokens ?? (promptTokens !== null && completionTokens !== null ? promptTokens + completionTokens : null);
+    return { id: model, providerId, ok: true, status: res.status, ttfbMs, totalMs, tps, charsPerSec, tokens: { prompt: promptTokens, completion: completionTokens, total: totalTokens }, chars, label: "成功", raw: json };
+  } catch (e) {
+    const totalMs = Math.round((typeof performance !== "undefined" && performance.now ? performance.now() : clock()) - t0);
+    const msg = e?.message || String(e);
+    const isTimeout = /timeout|abort/i.test(msg);
+    return { id: model, providerId, ok: false, error: msg.slice(0, 300), label: isTimeout ? "超时" : "网络错误", ttfbMs, totalMs, tps: null, charsPerSec: null, tokens: null };
+  }
+}

package/src/cli/commands/provider/bench.js

@@ -0,0 +1,140 @@
+import { probeModels } from "../../../bench/probe.js";
+import { runOne } from "../../../bench/runner.js";
+import { formatReport } from "../../../bench/report.js";
+import { defaultModelsPath, defaultChatPath } from "../../../state/provider-config.js";
+
+function parseBenchArgs(rest) {
+  const opts = { json: false, prompt: "hi", maxTokens: 32, timeoutMs: 30000 };
+  for (let i = 0; i < rest.length; i++) {
+    const a = rest[i];
+    if (a === "--json" || a === "-json") opts.json = true;
+    else if (a === "--prompt" && rest[i + 1]) opts.prompt = rest[++i];
+    else if (a.startsWith("--prompt=")) opts.prompt = a.slice(9);
+    else if (a === "--max-tokens" && rest[i + 1]) opts.maxTokens = Number(rest[++i]) || 32;
+    else if (a.startsWith("--max-tokens=")) opts.maxTokens = Number(a.split("=")[1]) || 32;
+    else if (a === "--timeout" && rest[i + 1]) opts.timeoutMs = Number(rest[++i]) || 30000;
+    else if (a.startsWith("--timeout=")) opts.timeoutMs = Number(a.split("=")[1]) || 30000;
+  }
+  return opts;
+}
+
+function buildHeadersForProvider(providerId, apiKey, auth) {
+  const h = {};
+  if (String(providerId).toLowerCase() === "workbuddy") {
+    h["Content-Type"] = "application/json";
+    h["Accept"] = "text/event-stream";
+    h["User-Agent"] = "CLI/2.115.0 WorkBuddy/2.115.0";
+    h["Origin"] = "https://www.codebuddy.cn";
+    h["Referer"] = "https://www.codebuddy.cn/";
+    h["X-Product"] = "SaaS";
+    if (apiKey) h["Authorization"] = `Bearer ${apiKey}`;
+    if (auth?.uid) h["X-User-Id"] = auth.uid;
+    h["X-Domain"] = auth?.domain || "www.codebuddy.cn";
+    if (auth?.enterpriseId) {
+      h["X-Enterprise-Id"] = auth.enterpriseId;
+      h["X-Tenant-Id"] = auth.enterpriseId;
+    }
+    return h;
+  }
+  if (apiKey) h["Authorization"] = `Bearer ${apiKey}`;
+  return h;
+}
+
+export async function handleProviderBench(id, sub, rest, args, deps = {}) {
+  const isBench = sub === "bench" || sub === "benchmark" || sub === "eval" || sub === "test";
+  if (!isBench) return false;
+  const fetchImpl = deps.fetchImpl || globalThis.fetch;
+  const loadConfigs = deps.loadProviderConfigs || (await import("../../../state.js")).loadProviderConfigs;
+  const loadKeys = deps.loadProviderKeys || (await import("../../../state.js")).loadProviderKeys;
+  const loadAllowed = deps.loadProviderAllowedModels || (await import("../../../state.js")).loadProviderAllowedModels;
+  const loadAllowAny = deps.loadProviderAllowAnyModels || (await import("../../../state.js")).loadProviderAllowAnyModels;
+  const loadBaseUrl = deps.loadProviderBaseUrl || (await import("../../../state.js")).loadProviderBaseUrl;
+  const loadAuths = deps.loadProviderAuths || (async () => { try { const m = await import("../../../state.js"); return m.loadProviderAuths ? m.loadProviderAuths(id) : []; } catch { return []; } }) && (await import("../../../state.js")).loadProviderAuths;
+
+  const providerId = String(id || "").trim();
+  if (!providerId) {
+    console.error("usage: mslxdff -provider <id> bench [--json] [--prompt hi] [--max-tokens 32]");
+    process.exit(1);
+  }
+  const opts = parseBenchArgs(rest || []);
+  const configs = loadConfigs();
+  const cfg = configs[providerId] || {};
+  const keys = loadKeys(providerId) || [];
+  const allowed = loadAllowed(providerId) || [];
+  const allowAny = loadAllowAny(providerId);
+  const baseUrl = (loadBaseUrl(providerId) || cfg.baseUrl || "").trim();
+  let auths = [];
+  try { const m = await import("../../../state.js"); auths = m.loadProviderAuths ? m.loadProviderAuths(providerId) : []; } catch {}
+  if (!baseUrl) {
+    console.error(`provider ${providerId}: missing baseUrl — 先设置: mslxdff -provider ${providerId} set-url https://api.example.com/v1`);
+    process.exit(1);
+  }
+  if (!keys.length) {
+    console.error(`provider ${providerId}: 未配置 Key — 先设置: mslxdff -provider ${providerId} <key>`);
+    process.exit(1);
+  }
+  // 空勾选 → 不发 chat，仅探活模型列表并提示
+  if (!allowed.length) {
+    const modelsPath = cfg.modelsPath || defaultModelsPath(providerId);
+    console.log(`provider ${providerId}: 未设置 allowlist（allowAny=${allowAny ? "ON" : "OFF"}），不发起测速，仅探活模型列表...`);
+    console.log(`尝试：GET ${baseUrl}${modelsPath} → GET ${baseUrl}/v1/models → GET ${baseUrl}/models`);
+    const headers = buildHeadersForProvider(providerId, keys[0], auths[0]);
+    const probed = await probeModels({ baseUrl, modelsPath, headers, fetchImpl, timeoutMs: 8000 });
+    if (!probed.ok) {
+      console.error(`探活失败：${probed.error}`);
+      console.error(`已尝试：${probed.tried.join(", ")}`);
+      console.error(`请手动设置 allowlist：mslxdff -provider ${providerId} allowlist set <model>`);
+      if (opts.json) console.log(JSON.stringify({ ok: false, error: probed.error, tried: probed.tried }, null, 2));
+      process.exit(1);
+    }
+    const list = probed.data || [];
+    if (!list.length) {
+      console.log("探活成功但返回空列表，请确认上游是否暴露 /v1/models");
+      if (opts.json) console.log(JSON.stringify({ ok: true, data: [] }, null, 2));
+      process.exit(0);
+    }
+    console.log(`\n发现 ${list.length} 个模型：`);
+    for (const m of list.slice(0, 30)) {
+      console.log(`  - ${m.id || m.model || m.name || JSON.stringify(m).slice(0, 80)}`);
+    }
+    if (list.length > 30) console.log(`  ... 还有 ${list.length - 30} 个未展示`);
+    console.log(`\n下一步：勾选后再测（只测勾选，避免扣费）`);
+    console.log(`  mslxdff -provider ${providerId} allowlist set ${list.slice(0, 2).map((m) => m.id).join(" ")}`);
+    console.log(`  mslxdff -provider ${providerId} bench${opts.json ? " --json" : ""}`);
+    if (opts.json) console.log(JSON.stringify({ ok: true, data: list, hint: `pick then bench` }, null, 2));
+    process.exit(0);
+  }
+
+  // 有勾选 → 逐个测
+  console.log(`bench ${providerId}: 共 ${allowed.length} 个已勾选模型，逐个测速（串行，${opts.timeoutMs}ms 超时）...`);
+  if (!opts.json) console.log(`prompt="${opts.prompt}" maxTokens=${opts.maxTokens}\n`);
+  const chatPath = cfg.chatPath || defaultChatPath(providerId);
+  const results = [];
+  for (let i = 0; i < allowed.length; i++) {
+    const raw = allowed[i];
+    const model = String(raw || "").trim();
+    if (!opts.json) process.stdout.write(`  [${i + 1}/${allowed.length}] ${model} ... `);
+    // 轮询 key/auth：按索引取，超长循环
+    const kIdx = i % keys.length;
+    const aIdx = Math.min(kIdx, auths.length - 1);
+    const key = keys[kIdx];
+    const auth = auths[aIdx] || auths[0] || null;
+    const headers = buildHeadersForProvider(providerId, key, auth);
+    const r = await runOne({ baseUrl, chatPath, model, providerId, apiKey: key, headers, prompt: opts.prompt, maxTokens: opts.maxTokens, timeoutMs: opts.timeoutMs, fetchImpl });
+    results.push(r);
+    if (!opts.json) {
+      if (r.ok) console.log(`OK  TTFB ${r.ttfbMs}ms  总 ${r.totalMs}ms  ${r.tps != null ? `${r.tps} t/s` : r.charsPerSec != null ? `${r.charsPerSec} 字/秒` : "—"}`);
+      else console.log(`FAIL ${r.label} ${r.error ? `(${r.error.slice(0, 60)})` : ""}`);
+    }
+  }
+  const report = formatReport(results, { json: opts.json });
+  console.log("\n" + report.text);
+  if (opts.json) {
+    // json 已在 text 中输出一次（report.text 是 JSON），无需重复
+  }
+  const failed = results.filter((r) => !r.ok).length;
+  if (failed && !opts.json) {
+    console.log(`\n提示：失败 ${failed} 个多为 402余额不足/429限流/超时，可清冷却或换 Key 后重试`);
+  }
+  process.exit(failed ? 2 : 0);
+}

package/src/cli/commands/provider/index.js

@@ -187,6 +187,8 @@ export async function handleProvider(args) {
   if (await handleProviderAllowlist(id, sub, rest)) return true;
   const { handleProviderModels } = await import("./models.js");
   if (await handleProviderModels(id, sub, args, rest)) return true;
+  const { handleProviderBench } = await import("./bench.js");
+  if (await handleProviderBench(id, sub, rest, args)) return true;
   const { handleProviderKeys } = await import("./keys.js");
   await handleProviderKeys(id, sub, rest, args);
   return true;

package/src/metrics.js

@@ -0,0 +1,63 @@
+import { normalizeFullId, toFullId } from "./providers/model-id.js";
+
+export { normalizeFullId, toFullId };
+
+// 从上游 usage 或文本兜底算 tps / charsPerSec
+export function computeMetrics({ ttfbMs, totalMs, promptTokens, completionTokens, chars }) {
+  const ttfb = Number(ttfbMs);
+  const total = Number(totalMs);
+  const hasTtfb = Number.isFinite(ttfb) && ttfb >= 0;
+  const hasTotal = Number.isFinite(total) && total >= 0;
+  const completionMs = hasTtfb && hasTotal ? Math.max(0, total - ttfb) : hasTotal ? total : null;
+  const comp = Number(completionTokens);
+  const hasComp = Number.isFinite(comp) && comp > 0 && Number.isFinite(completionMs) && completionMs > 0;
+  const tps = hasComp ? Number((comp / (completionMs / 1000)).toFixed(1)) : null;
+  const c = Number(chars);
+  const hasChars = Number.isFinite(c) && c > 0 && Number.isFinite(completionMs) && completionMs > 0;
+  const charsPerSec = !hasComp && hasChars ? Math.round(c / (completionMs / 1000)) : null;
+  return { completionMs, tps, charsPerSec };
+}
+
+// 从 Response 文本或 SSE 末帧提取 usage
+export function extractUsageFromJson(parsed) {
+  if (!parsed || typeof parsed !== "object") return null;
+  const u = parsed.usage;
+  if (!u || typeof u !== "object") return null;
+  const prompt = Number(u.prompt_tokens ?? u.promptTokens ?? u.input_tokens);
+  const comp = Number(u.completion_tokens ?? u.completionTokens ?? u.output_tokens);
+  const total = Number(u.total_tokens ?? u.totalTokens);
+  const out = {};
+  if (Number.isFinite(prompt)) out.prompt_tokens = prompt;
+  if (Number.isFinite(comp)) out.completion_tokens = comp;
+  if (Number.isFinite(total)) out.total_tokens = total;
+  // reasoning_tokens 透传
+  const rt = u.completion_tokens_details?.reasoning_tokens ?? u.reasoning_tokens;
+  if (Number.isFinite(Number(rt))) out.reasoning_tokens = Number(rt);
+  return Object.keys(out).length ? out : null;
+}
+
+export function extractUsageFromSseText(sseText) {
+  if (!sseText) return null;
+  const lines = String(sseText).split("\n");
+  for (let i = lines.length - 1; i >= 0; i--) {
+    const line = lines[i].trim();
+    if (!line.startsWith("data:")) continue;
+    const data = line.slice(5).trim();
+    if (data === "[DONE]") continue;
+    try {
+      const j = JSON.parse(data);
+      if (j.usage) return extractUsageFromJson(j);
+      if (j.choices?.[0]?.finish_reason && j.usage) return extractUsageFromJson(j);
+    } catch {}
+  }
+  return null;
+}
+
+// 归一模型全称：provider + raw -> opencode/xxx
+export function resolveFullId(rawModel, providerHint) {
+  const raw = String(rawModel || "").trim();
+  if (!raw) return "";
+  if (raw.includes("/")) return normalizeFullId(raw);
+  const prov = String(providerHint || "opencode").trim() || "opencode";
+  return toFullId(prov, raw);
+}