From e3a485bd3209fa3f615199d6afa32d579df4fc9f Mon Sep 17 00:00:00 2001 From: Yichen Jiang Date: Thu, 22 Oct 2026 13:34:14 +0800 Subject: [PATCH] fix(llm-deepseek): sync the built-in context window with the official listing The official `/models` listing declares a 1,048,576-token context window for `deepseek-flash` and `deepseek-v4-pro`; the built-in catalog said 1,000,000, a decimal-versus-binary mismatch. Both built-in entries now carry the listed value, so offline starts and listing failures size the context the same way the listing does. The Models page shows the exact count for these rows, since 1,048,576 has no decimal K/M spelling. `defaultContextWindow`, the fallback for model ids the catalog does not describe, stays 1,000,000. --- .../onboarding-deepseek-config/default-models.expected.md | 2 +- packages/llm/llm-deepseek-account/tests/provider.spec.ts | 2 +- packages/llm/llm-deepseek/README.i18n.yaml | 4 ++-- packages/llm/llm-deepseek/README.md | 2 +- packages/llm/llm-deepseek/README.zh.md | 2 +- packages/llm/llm-deepseek/src/models.ts | 8 +++++--- packages/llm/llm-deepseek/tests/listed-catalog.spec.ts | 4 ++-- packages/llm/llm-deepseek/tests/runtime.spec.ts | 4 ++-- 8 files changed, 15 insertions(+), 13 deletions(-) diff --git a/apps/web/tests/expected/onboarding-deepseek-config/default-models.expected.md b/apps/web/tests/expected/onboarding-deepseek-config/default-models.expected.md index 6ea83d310a..bde82cec8f 100644 --- a/apps/web/tests/expected/onboarding-deepseek-config/default-models.expected.md +++ b/apps/web/tests/expected/onboarding-deepseek-config/default-models.expected.md @@ -37,7 +37,7 @@ - text: 上下文窗口 - textbox "上下文窗口 1": - /placeholder: 1M - - text: 1M + - text: "1048576" - text: 最大输出 token 数 - textbox "最大输出 token 数 1": - /placeholder: 256K diff --git a/packages/llm/llm-deepseek-account/tests/provider.spec.ts b/packages/llm/llm-deepseek-account/tests/provider.spec.ts index d4a75057d2..c3b5fd2f41 100644 --- a/packages/llm/llm-deepseek-account/tests/provider.spec.ts +++ b/packages/llm/llm-deepseek-account/tests/provider.spec.ts @@ -105,7 +105,7 @@ it('falls back to the bundled catalog when the listing fails', async () => { await ctx.plugin(Account, { baseURL: server.baseURL }) expect((await ctx.llm.listModels('deepseek-account')).map(row => row.id)).toEqual(['deepseek-flash', 'deepseek-v4-pro']) expect(warn).toHaveBeenCalledExactlyOnceWith('[llm-deepseek-account]', expect.stringContaining('using the bundled catalog') as string) - await expect(ctx.llm.resolveModelInfo('deepseek-account', 'deepseek-flash')).resolves.toMatchObject({ context: { contextWindow: 1_000_000 } }) + await expect(ctx.llm.resolveModelInfo('deepseek-account', 'deepseek-flash')).resolves.toMatchObject({ context: { contextWindow: 1_048_576 } }) }) it('hides and refuses listed models this build cannot serve before any inference request', async () => { diff --git a/packages/llm/llm-deepseek/README.i18n.yaml b/packages/llm/llm-deepseek/README.i18n.yaml index 83903a3df5..8f1b2c46aa 100644 --- a/packages/llm/llm-deepseek/README.i18n.yaml +++ b/packages/llm/llm-deepseek/README.i18n.yaml @@ -21,8 +21,8 @@ en: 6850d08bc091ad23 zh: 2906e6cf1dfed6ce /deepseek-ai-dsh-llm-deepseek/use-this-package/minimal-configuration: - en: 65feb096f8a7b8ea - zh: 77b9d16ed9aec498 + en: 6a15f2e0efec4c37 + zh: 7fb5f1e4a92b01ad /deepseek-ai-dsh-llm-deepseek/use-this-package/endpoint-and-wire-format: en: 522e74b770e9d4ca zh: a8b261b41677bfe7 diff --git a/packages/llm/llm-deepseek/README.md b/packages/llm/llm-deepseek/README.md index eaaa327eda..aef0563887 100644 --- a/packages/llm/llm-deepseek/README.md +++ b/packages/llm/llm-deepseek/README.md @@ -50,7 +50,7 @@ Choose this adapter for DeepSeek's official API or a Messages-compatible gateway filesApiTimeoutMs: 60000 ``` -A request selects the route with `provider: deepseek-official`; the model id passes through to the wire, so new DeepSeek models need no re-registration. Omitted `models` advertises the text- and image-capable `deepseek-flash` alongside the text-only `deepseek-v4-pro`, each with a 1,000,000-token context window. An explicit list replaces those defaults, and core calls still pass unlisted model ids through as text-only routes. On the official endpoint the API-key route follows the service's model listing while `models` holds those defaults ([listing rules](../llm-deepseek-api-key/README.md#use-this-package)). GUI selection requires a catalog entry; a saved selection can still submit requests after its catalog entry disappears. Clients, including model discovery tools, can read the advisory entries through `ctx.llm.listModels('deepseek-official')`. Image-capable entries may set `imagePixelBudget` to a positive integer or `low`, and may set `imageMaxBytes`. An entry may declare `systemPromptUpdate: in-history` when its endpoint reads the latest `system` message at any position of `messages` as the complete effective system prompt; the adapter reports the mode on the resolved model and the prepared call, and the agent loop then appends a changed prompt after the cached history instead of rewriting the leading system message ([decision rule](../../core/agent-loop/README.md#understand-the-implementation)). The default `deepseek-flash` entry declares this mode; other models require an explicit `models` declaration, and any value other than `in-history` fails at load with `llm-deepseek: catalog model "" systemPromptUpdate must be "in-history" when present`. +A request selects the route with `provider: deepseek-official`; the model id passes through to the wire, so new DeepSeek models need no re-registration. Omitted `models` advertises the text- and image-capable `deepseek-flash` alongside the text-only `deepseek-v4-pro`, each with the 1,048,576-token context window the official model listing declares. An explicit list replaces those defaults, and core calls still pass unlisted model ids through as text-only routes. On the official endpoint the API-key route follows the service's model listing while `models` holds those defaults ([listing rules](../llm-deepseek-api-key/README.md#use-this-package)). GUI selection requires a catalog entry; a saved selection can still submit requests after its catalog entry disappears. Clients, including model discovery tools, can read the advisory entries through `ctx.llm.listModels('deepseek-official')`. Image-capable entries may set `imagePixelBudget` to a positive integer or `low`, and may set `imageMaxBytes`. An entry may declare `systemPromptUpdate: in-history` when its endpoint reads the latest `system` message at any position of `messages` as the complete effective system prompt; the adapter reports the mode on the resolved model and the prepared call, and the agent loop then appends a changed prompt after the cached history instead of rewriting the leading system message ([decision rule](../../core/agent-loop/README.md#understand-the-implementation)). The default `deepseek-flash` entry declares this mode; other models require an explicit `models` declaration, and any value other than `in-history` fails at load with `llm-deepseek: catalog model "" systemPromptUpdate must be "in-history" when present`. | Field | Default | Meaning | |---|---|---| diff --git a/packages/llm/llm-deepseek/README.zh.md b/packages/llm/llm-deepseek/README.zh.md index fcfddc6a73..092e9b804a 100644 --- a/packages/llm/llm-deepseek/README.zh.md +++ b/packages/llm/llm-deepseek/README.zh.md @@ -50,7 +50,7 @@ kind: "package-reference" filesApiTimeoutMs: 60000 ``` -请求用 `provider: deepseek-official` 选择路由;模型 id 原样传到协议,因此新增 DeepSeek 模型无需重新注册。省略 `models` 时公布支持文本和图像的 `deepseek-flash`,以及仅支持文本的 `deepseek-v4-pro`,各自的上下文窗口均为 1,000,000 token。显式列表会替换这些默认值,核心调用中未列出的模型 id 仍作为纯文本路由原样通过。在官方端点上,当 `models` 保持这些默认值时,API key 路由改为使用服务端的模型列表([列表规则](../llm-deepseek-api-key/README.zh.md#use-this-package))。GUI 选择要求模型具有目录条目;条目消失后,已保存的选择仍可提交请求。包括模型发现工具在内的客户端可通过 `ctx.llm.listModels('deepseek-official')` 读取这些建议性条目。支持图片的条目可把 `imagePixelBudget` 设置为正整数或 `low`,也可以设置 `imageMaxBytes`。当端点把 `messages` 中任意位置最新的 `system` 消息读作完整的有效系统提示词时,条目可以声明 `systemPromptUpdate: in-history`;适配器会在已解析模型与已准备调用上报告该模式,agent loop(智能体循环)随后把变化后的提示词追加到已缓存历史之后,而不是改写开头的 system 消息([决策规则](../../core/agent-loop/README.zh.md#understand-the-implementation))。默认的 `deepseek-flash` 条目声明该模式;其他模型需通过 `models` 显式声明,`in-history` 以外的任何值都会在加载时以 `llm-deepseek: catalog model "" systemPromptUpdate must be "in-history" when present` 失败。 +请求用 `provider: deepseek-official` 选择路由;模型 id 原样传到协议,因此新增 DeepSeek 模型无需重新注册。省略 `models` 时公布支持文本和图像的 `deepseek-flash`,以及仅支持文本的 `deepseek-v4-pro`,各自的上下文窗口均为官方模型列表声明的 1,048,576 token。显式列表会替换这些默认值,核心调用中未列出的模型 id 仍作为纯文本路由原样通过。在官方端点上,当 `models` 保持这些默认值时,API key 路由改为使用服务端的模型列表([列表规则](../llm-deepseek-api-key/README.zh.md#use-this-package))。GUI 选择要求模型具有目录条目;条目消失后,已保存的选择仍可提交请求。包括模型发现工具在内的客户端可通过 `ctx.llm.listModels('deepseek-official')` 读取这些建议性条目。支持图片的条目可把 `imagePixelBudget` 设置为正整数或 `low`,也可以设置 `imageMaxBytes`。当端点把 `messages` 中任意位置最新的 `system` 消息读作完整的有效系统提示词时,条目可以声明 `systemPromptUpdate: in-history`;适配器会在已解析模型与已准备调用上报告该模式,agent loop(智能体循环)随后把变化后的提示词追加到已缓存历史之后,而不是改写开头的 system 消息([决策规则](../../core/agent-loop/README.zh.md#understand-the-implementation))。默认的 `deepseek-flash` 条目声明该模式;其他模型需通过 `models` 显式声明,`in-history` 以外的任何值都会在加载时以 `llm-deepseek: catalog model "" systemPromptUpdate must be "in-history" when present` 失败。 | 字段 | 默认值 | 含义 | |---|---|---| diff --git a/packages/llm/llm-deepseek/src/models.ts b/packages/llm/llm-deepseek/src/models.ts index 15d0c78253..c25d5111fd 100644 --- a/packages/llm/llm-deepseek/src/models.ts +++ b/packages/llm/llm-deepseek/src/models.ts @@ -1,13 +1,15 @@ /** Default DeepSeek model catalog. */ -import { DEFAULT_CONTEXT_WINDOW } from './defaults.ts' import type { DeepSeekCatalogModel } from './types.ts' +/** Context window of both official models, as the official `/models` listing declares it. */ +const OFFICIAL_CONTEXT_WINDOW = 1_048_576 + /** Advisory official model entries; deployments may replace the catalog. */ export const DEFAULT_MODELS: DeepSeekCatalogModel[] = [ { id: 'deepseek-flash', name: 'DeepSeek-V4.1-Flash', - contextWindow: DEFAULT_CONTEXT_WINDOW, + contextWindow: OFFICIAL_CONTEXT_WINDOW, inputModalities: ['text', 'image'], systemPromptUpdate: 'in-history', toolUpdate: 'addition-only', @@ -16,6 +18,6 @@ export const DEFAULT_MODELS: DeepSeekCatalogModel[] = [ id: 'deepseek-v4-pro', name: 'DeepSeek-V4-Pro', description: 'Stronger agentic coding, knowledge, and difficult reasoning; suited to complex or quality-critical tasks at higher cost.', - contextWindow: DEFAULT_CONTEXT_WINDOW, + contextWindow: OFFICIAL_CONTEXT_WINDOW, }, ] diff --git a/packages/llm/llm-deepseek/tests/listed-catalog.spec.ts b/packages/llm/llm-deepseek/tests/listed-catalog.spec.ts index 50021e1d4f..dbb4354777 100644 --- a/packages/llm/llm-deepseek/tests/listed-catalog.spec.ts +++ b/packages/llm/llm-deepseek/tests/listed-catalog.spec.ts @@ -29,7 +29,7 @@ describe('without a listing', () => { it('offers the bundled catalog', () => { const catalog = resolve(undefined) expect(catalog.offered.map(model => [model.id, model.name, model.contextWindow])).toEqual([ - ['deepseek-flash', 'DeepSeek-V4.1-Flash', 1_000_000], ['deepseek-v4-pro', 'DeepSeek-V4-Pro', 1_000_000], + ['deepseek-flash', 'DeepSeek-V4.1-Flash', 1_048_576], ['deepseek-v4-pro', 'DeepSeek-V4-Pro', 1_048_576], ]) expect(catalog.resolution).toEqual(catalog.offered) expect(catalog.refusals.size).toBe(0) @@ -57,7 +57,7 @@ describe('with a listing', () => { it('keeps this build\'s values for fields a listing omits', () => { const [pro] = resolve([{ id: 'deepseek-v4-pro', api_capabilities: { anthropic_messages: {} } }]).offered - expect(pro).toMatchObject({ name: 'DeepSeek-V4-Pro', contextWindow: 1_000_000, inputModalities: ['text'] }) + expect(pro).toMatchObject({ name: 'DeepSeek-V4-Pro', contextWindow: 1_048_576, inputModalities: ['text'] }) expect(pro).not.toHaveProperty('maxTokens') }) diff --git a/packages/llm/llm-deepseek/tests/runtime.spec.ts b/packages/llm/llm-deepseek/tests/runtime.spec.ts index 54cec3fe36..3c4676ba0d 100644 --- a/packages/llm/llm-deepseek/tests/runtime.spec.ts +++ b/packages/llm/llm-deepseek/tests/runtime.spec.ts @@ -1810,7 +1810,7 @@ describe('plugin registration and config', () => { name: 'DeepSeek-V4.1-Flash', inputModalities: ['text', 'image'], systemPromptUpdate: 'in-history', - context: { contextWindow: 1_000_000 }, + context: { contextWindow: 1_048_576 }, defaultMaxTokens: 256_000, reasoning: { efforts: [ @@ -1832,7 +1832,7 @@ describe('plugin registration and config', () => { expect(info).toMatchObject({ id: 'deepseek-v4-pro', inputModalities: ['text'], - context: { contextWindow: 1_000_000 }, + context: { contextWindow: 1_048_576 }, defaultMaxTokens: 256_000, }) expect(info?.systemPromptUpdate).toBeUndefined()