SemianiakaVY/openpencil
1
0
Fork 0
Code Issues Pull requests Actions 1 Packages Projects Releases Wiki Activity

chore(ab-corpus): route glm-5.1 + kimi-k2.6 through 方舟 CP (Ark)

Browse source 
Volcengine 方舟 (Ark) added GLM-5.1 and Kimi-K2.6 to its coding
plan on 2026-04-22 — single ARK_CODING_KEY covers both. Harness
now prefers this route over the previous paths:

  - glm-5.1 was routed to clients/glm.ts (GLM official CP via
    open.bigmodel.cn with GLM_OFFICIAL_CODING_KEY). Now routed to
    new clients/ark.ts. The old glm.ts file is kept on disk for
    historical comparison but not wired into the default router —
    callers who want to A/B the old GLM-official path vs. new Ark
    path can import callGlm directly.
  - kimi-k2.6 is new — added as a dedicated router branch above
    the kimi-k2.5 (bailian) branch so the version-specific match
    lands on Ark.

Old kimi-k2.5 continues to route through clients/bailian.ts
(DashScope aggregator) for continuity with earlier A/B runs.

Key management (unchanged from the harness convention):
  - ARK_CODING_KEY — Volcengine 方舟 CP UUID format key. Export
    in shell before running --live; never committed.
  - Existing MINIMAX_API_KEY / GLM_OFFICIAL_CODING_KEY /
    DASHSCOPE_BAILIAN_CODING_KEY all still honored for their
    respective routes.

Throw message updated so missing-key errors surface the correct
env var for each route.
This commit is contained in:
Fini 2026-04-22 11:05:00 +08:00
parent dd5d156160
commit 5538350a2f
2 changed files with 103 additions and 10 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

46
scripts/ab-corpus/clients/ark.ts Normal file
View file

@ -0,0 +1,46 @@
/**
* 方舟 (Volcengine Ark) Coding Plan wrapper. One API key covers
* multiple third-party models hosted on Ark's coding tier —
* currently GLM-5.1 and KIMI-K2.6. Cribbed from builtin-provider-
* presets.ts `ark-coding` preset.
*
* Default baseURL: https://ark.cn-beijing.volces.com/api/coding/v3
* Override via `ARK_BASE_URL` if Volcengine ever ships a regional
* variant (none published as of 2026-04-22).
*
* Key env var: `ARK_CODING_KEY` — Volcengine's UUID-format ARK
* access key. NOT committed to the repo; export in your shell
* before running `bun scripts/ab-corpus/run.ts --live`.
*/
import { callOpenAICompat } from './openai-compat';
const ARK_BASE_URL = process.env.ARK_BASE_URL ?? 'https://ark.cn-beijing.volces.com/api/coding/v3';
const ARK_API_KEY_ENV = 'ARK_CODING_KEY';
export interface CallArkArgs {
model: string;
system: string;
user: string;
temperature?: number;
maxTokens?: number;
}
export async function callArk(args: CallArkArgs): Promise<string> {
const apiKey = process.env[ARK_API_KEY_ENV];
if (!apiKey) {
throw new Error(
`${ARK_API_KEY_ENV} not set — export it (Volcengine 方舟 CP UUID key) before running with --live`,
);
}
return callOpenAICompat({
baseURL: ARK_BASE_URL,
apiKey,
model: args.model,
system: args.system,
user: args.user,
temperature: args.temperature,
maxTokens: args.maxTokens,
label: 'ark',
});
}

67
scripts/ab-corpus/real-model.ts
View file

@ -15,7 +15,7 @@
import { callMinimax } from './clients/minimax';
import { callCodex } from './clients/codex-cli';
import { callBailian } from './clients/bailian';
import { callGlm } from './clients/glm';
import { callArk } from './clients/ark';
import { buildSystemPrompt } from './build-prompt';
import type { ModelCall } from './stub-model';
@ -24,15 +24,31 @@ import type { ModelCall } from './stub-model';
*
* - `minimax*` → clients/minimax.ts (MINIMAX_API_KEY)
* - `gpt-*` / o* → clients/codex-cli.ts (Codex CLI subscription)
* - `glm-5.1` → clients/glm.ts (GLM_OFFICIAL_CODING_KEY,
* GLM-official coding-plan endpoint)
* - `glm-5.1` → clients/ark.ts (ARK_CODING_KEY;
* Volcengine 方舟 now hosts
* GLM-5.1 on its coding plan
* as of 2026-04-22. Replaces
* the earlier GLM-official
* CP route via open.bigmodel.cn
* — clients/glm.ts is kept on
* disk for historical reference
* but no longer auto-routed.)
* - `kimi-k2.6` → clients/ark.ts (ARK_CODING_KEY; Volcengine
* added K2.6 to 方舟 CP on
* 2026-04-22)
* - `glm-*` → clients/bailian.ts (DASHSCOPE_BAILIAN_CODING_KEY,
* earlier GLM versions hosted on
* Bailian's DashScope aggregator)
* - `kimi-*` → clients/bailian.ts (DASHSCOPE_BAILIAN_CODING_KEY,
* Kimi K-series also on Bailian)
* Kimi K2.5 and earlier on Bailian)
*
* Anything else → throw with the supported list.
*
* When benchmarking across routes (e.g. to compare the new Ark GLM-5.1
* vs. the legacy GLM-official route), import `callGlm` from
* `./clients/glm.ts` directly and wire it here under a different model
* alias (say `glm-5.1-legacy`). The default router picks Ark because
* that's the path we're actively promoting.
*/
export async function realModelCall(call: ModelCall): Promise<string> {
const built = buildSystemPrompt(call.variant);
@ -50,8 +66,15 @@ export async function realModelCall(call: ModelCall): Promise<string> {
return callCodex({ model, system: built.system, user });
}
if (/^glm-5\.1/i.test(model)) {
return callGlm({
model: mapGlmOfficialId(model),
return callArk({
model: mapGlmArkId(model),
system: built.system,
user,
});
}
if (/^kimi-k2\.6/i.test(model)) {
return callArk({
model: mapKimiArkId(model),
system: built.system,
user,
});
@ -71,7 +94,7 @@ export async function realModelCall(call: ModelCall): Promise<string> {
});
}
throw new Error(
`No live adapter for model "${model}". Supported: minimax-* (MINIMAX_API_KEY), gpt-*/o1/o3/o4 (Codex CLI), glm-5.1 (GLM_OFFICIAL_CODING_KEY), glm-5 / kimi-* (DASHSCOPE_BAILIAN_CODING_KEY).`,
`No live adapter for model "${model}". Supported: minimax-* (MINIMAX_API_KEY), gpt-*/o1/o3/o4 (Codex CLI), glm-5.1 / kimi-k2.6 (ARK_CODING_KEY), glm-5 / kimi-k2.5 (DASHSCOPE_BAILIAN_CODING_KEY).`,
);
}
@ -84,14 +107,38 @@ function mapMinimaxId(id: string): string {
}
function mapGlmOfficialId(id: string): string {
// GLM official CP ships "glm-4.7" as the current coding model
// (per builtin-provider-presets.ts modelPlaceholder). User's
// preferred short form "glm-5.1" maps to that id here.
// GLM official CP (open.bigmodel.cn) used to ship "glm-4.7" as the
// current coding model. The harness no longer routes `glm-5.1` here
// (Ark took over, see mapGlmArkId) but this mapper is kept so
// callers who manually import `./clients/glm.ts` for A/B comparison
// can still get the canonical id.
const lower = id.toLowerCase();
if (lower === 'glm-5.1' || lower === 'glm-5.1-coding') return 'glm-4.7';
return id;
}
function mapGlmArkId(id: string): string {
// 方舟 CP exposes Zhipu GLM-5.1 on the coding-plan endpoint. The
// on-Ark model name is "glm-5.1" verbatim (no endpoint-ID alias
// required as of 2026-04-22 — earlier Ark models needed `ep-xxx`
// but GLM-5.1 was onboarded with the promoted short name).
const lower = id.toLowerCase();
if (lower === 'glm-5.1' || lower === 'glm-5.1-coding' || lower === 'glm-5.1-ark') {
return 'glm-5.1';
}
return id;
}
function mapKimiArkId(id: string): string {
// 方舟 CP exposes Moonshot Kimi K2.6 on the coding-plan endpoint.
// Volcengine accepts the promoted short name `kimi-k2.6` directly.
const lower = id.toLowerCase();
if (lower === 'kimi-k2.6' || lower === 'kimi-2.6' || lower === 'kimi-k2.6-ark') {
return 'kimi-k2.6';
}
return id;
}
function mapGlmBailianId(id: string): string {
return mapGlmBailianIdInternal(id);
}