chore: refresh model provider presets#509
Conversation
There was a problem hiding this comment.
Pull request overview
Refreshes the static model provider preset catalogs to reflect current official model listings/capabilities, ensuring the infrastructure “known models” set stays up to date.
Changes:
- Added new Moonshot (Kimi) presets:
kimi-k3andkimi-k2.7-code-highspeed. - Added Groq preset for
minimaxai/minimax-m2.7. - Updated ERNIE presets: corrected
ernie-5.1reasoning/tool flags and addedernie-5.0. - Added new Baichuan presets:
Baichuan4-Air,Baichuan3-Turbo-128k, andBaichuan3-Turbo.
Reviewed changes
Copilot reviewed 4 out of 4 changed files in this pull request and generated no comments.
| File | Description |
|---|---|
| packages/infrastructure/src/static-data/models/provider/Moonshot/index.ts | Adds Kimi K3 and K2.7 HighSpeed presets to Moonshot’s model list. |
| packages/infrastructure/src/static-data/models/provider/Groq/index.ts | Adds Groq’s minimaxai/minimax-m2.7 preset. |
| packages/infrastructure/src/static-data/models/provider/Ernie/index.ts | Fixes ernie-5.1 capability flags and adds ernie-5.0. |
| packages/infrastructure/src/static-data/models/provider/Baichuan/index.ts | Adds Baichuan 4 Air and Baichuan 3 Turbo variants to presets. |
💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.
![chatgpt-codex-connector[bot]](https://avatars.githubusercontent.com/in/1144995?s=60&v=4){kind=small}
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: 1d3cd0bb54
ℹ️ About Codex in GitHub
Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".
| maxTokens: 4000, | ||
| quoteMaxToken: 100000, | ||
| maxTemperature: 1.2, | ||
| responseFormatList: ['text', 'json_object'], |
There was a problem hiding this comment.
When users choose JSON output with Baichuan3-Turbo-128k or the Baichuan3-Turbo entry below, the preset now advertises json_object, but Baichuan's official API docs list response_format JSON mode support only for Baichuan4-Turbo and Baichuan4-Air. That makes FastGPT offer a response format the API rejects for these newly added Baichuan3 models, so the json_object capability should be removed from both Baichuan3 presets.
Useful? React with 👍 / 👎.
| responseFormatList: ['text', 'json_object'], | ||
| vision: true, | ||
| reasoning: true, | ||
| reasoningEffort: true, |
There was a problem hiding this comment.
For the new kimi-k2.7-code-highspeed preset, setting reasoningEffort advertises the top-level reasoning_effort option even though Kimi's parameter reference says the K2.7 Code series, including the highspeed variant, does not support reasoning_effort (only K3 does). If a user enables that option for this model, Moonshot will reject the request, so this flag should stay false for the highspeed preset.
Useful? React with 👍 / 👎.
Summary
minimaxai/minimax-m2.7.Official sources
Validation
bun run testpassed: 40 test files, 249 tests.git diff --checkpassed.bun tsc --noEmitremains blocked by an existing type error insdk/factory/src/tool-factory.test.ts:285involvingPluginChannelStreamSourceandStreamData; this file is outside the change.