feat(alibaba-token-plan): add qwen-deepseek profile preset - #3827
Conversation
![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: 24f6b0cb06
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".
| }), | ||
| profile("alibaba-token-plan-qwen-deepseek", ["alibaba-token-plan"], { | ||
| default: "alibaba-token-plan/qwen3.8-max-preview:high", | ||
| executor: "alibaba-token-plan/deepseek-v4-flash-0731:medium", |
There was a problem hiding this comment.
When this new profile is activated against the real Alibaba catalog, deepseek-v4-flash-0731 only supports low, high, and max efforts (packages/ai/src/models.json:27-35). Since profile activation clamps unsupported efforts down to the nearest supported level (packages/ai/src/model-thinking.ts:277-295), the requested executor :medium is silently materialized as :low, so this preset will not run the executor at the advertised/intended effort. Use one of the supported DeepSeek levels here, likely :high if this role should be above low.
Useful? React with 👍 / 👎.
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: 6ea45c784a
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".
| architect: "alibaba-token-plan/qwen3.8-max-preview:xhigh", | ||
| critic: "alibaba-token-plan/qwen3.8-max-preview:xhigh", | ||
| }), | ||
| profile("alibaba-token-plan-qwen-deepseek", ["alibaba-token-plan"], { |
There was a problem hiding this comment.
This adds a bundled packages/coding-agent model preset, but the commit does not add a packages/coding-agent/CHANGELOG.md entry under ## [Unreleased]; release notes for the package will omit the new --mpreset surface even though the repo contract requires package changes to be recorded there. Please add an Unreleased Added entry for the new profile.
AGENTS.md reference: AGENTS.md:L178-L178
Useful? React with 👍 / 👎.
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: 6e5e5ea777
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".
| critic: "alibaba-token-plan/qwen3.8-max-preview:xhigh", | ||
| }), | ||
| profile("alibaba-token-plan-qwen-deepseek", ["alibaba-token-plan"], { | ||
| default: "alibaba-token-plan/qwen-3.8-max:high", |
There was a problem hiding this comment.
When this preset is used against Alibaba Token Plan, the GA Qwen3.8-Max model id is qwen3.8-max (see Alibaba's Model Studio page), not qwen-3.8-max. Because the new qwen-deepseek profile makes this selector the default and the bundled catalog/provider preset register the same typo, activation succeeds locally but the first default/architect/critic request is sent with a model name Alibaba does not serve. Please use the upstream id consistently.
Useful? React with 👍 / 👎.
6e5e5ea to
4a80cd4
Compare
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: 4a80cd423e
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".
| if (existing) { | ||
| Object.assign(existing, metadata); | ||
| return; | ||
| const qwen: Model<"openai-responses"> = { |
There was a problem hiding this comment.
This adds a new bundled packages/ai Alibaba Token Plan model fallback and generator behavior, but the commit does not add an ## [Unreleased] entry to packages/ai/CHANGELOG.md. When the ai package is released, the new catalog/model behavior will be omitted from package release notes; please add the missing ai changelog entry separately from the coding-agent preset entry.
AGENTS.md reference: AGENTS.md:L178-L178
Useful? React with 👍 / 👎.
Mirror the local qwen-deepseek preset as an Alibaba Token Plan built-in profile so the Qwen-plus-DeepSeek role split works without a custom models.yml. Alibaba vendors DeepSeek V4 Flash as deepseek-v4-flash-0731, which only exposes low/high/max efforts, so the layofflabs planner xhigh/executor medium map to the nearest valid high/low. Lore-id: 8f2e1c9a Confidence: high Scope-risk: low Reversibility: revert-profile Tested: model-profiles-catalog/model-profile-activation/model-profiles-redteam Not-tested: live activation against an Alibaba Token Plan credential
Use the literal layofflabs reasoning split on deepseek-v4-flash-0731. Lore-id: 8f2e1c9a Confidence: high Scope-risk: low Reversibility: revert-profile Tested: model-profiles-catalog/model-profile-activation/model-profiles-redteam
Raise the qwen-deepseek profile's default role to high reasoning and the planner role to max, matching the operator's preferred split. Lore-id: 8f2e1c9a Confidence: high Scope-risk: low Reversibility: revert-profile Tested: model-profiles-catalog/model-profile-activation/model-profiles-redteam
Lore-id: 8f2e1c9a Confidence: high Scope-risk: low Reversibility: revert-profile Tested: model-profiles-catalog/model-profile-activation/model-profiles-redteam
Register qwen-3.8-max as the non-preview Qwen model in the Alibaba catalog and provider preset, and add the glm-deepseek profile mirroring qwen-deepseek's role split with GLM 5.2 as the expensive model. - qwen-deepseek: Qwen 3.8 Max (default/architect/critic) + DeepSeek V4 Flash 0731 (planner max / executor high) - glm-deepseek: GLM 5.2 (default high/architect xhigh/critic xhigh) + DeepSeek V4 Flash 0731 (planner max / executor high) Lore-id: 8f2e1c9a Confidence: high Scope-risk: low Reversibility: revert-profile Tested: model-profiles-catalog/model-profile-activation/model-profiles-redteam/provider-onboarding; ai generate-models Not-tested: live activation against an Alibaba Token Plan credential
… preset The docs-index.generated.ts embeds a hash of models.md; adding the new qwen-deepseek profile preset changed the models.md length, so the embedded hash went stale and failed the check:public-sync gate. Lore-id: 8f2e1c9a Confidence: high Scope-risk: low Reversibility: revert-profile Tested: check:public-sync Not-tested: full test suite (worktree has no node_modules)
The affected-path check:@gajae-code/ai gate failed on a formatting violation: Biome prefers the models.find(...) call on a single line. Collapse it to satisfy the formatter. Lore-id: 8f2e1c9a Confidence: high Scope-risk: low Reversibility: revert-profile Tested: biome check packages/ai/scripts/generate-models.ts
07755c7 to
b962e5a
Compare
1 participant
Add built-in Alibaba Token Plan profile
alibaba-token-plan-qwen-deepseekmirroring the local qwen-deepseek preset: Qwen 3.8 Max carries default (high), architect (xhigh), critic (xhigh), while DeepSeek V4 Flash 0731 covers planner (max) and executor (high).Alibaba vendors the model as
deepseek-v4-flash-0731, which ships with a 1M context / 384K output envelope in packages/ai/src/models.json.Changes: model-profiles.ts (profile + presentation), model-profiles-catalog.test.ts (36-profile matrix, presentation, routing), model-profile-activation.test.ts (five-role activation), docs/models.md.
Verified: 75 profile tests pass; biome + tsc clean. Not tested: live activation against an Alibaba Token Plan credential.