Skip to content

Conversation

@yyhhyyyyyy
Copy link
Collaborator

@yyhhyyyyyy yyhhyyyyyy commented Sep 4, 2025

add qwen3 thinking_budget support
Implement the thinking_budget portion of #836.

Summary by CodeRabbit

  • New Features

    • Per-model "Thinking Budget" controls added for Qwen3 (alongside Gemini), with validation, defaults, and ability to disable per-model.
    • Enabled thinking and function-calling for multiple Qwen3 variants and added two new Qwen3 thinking-specific models; thinkingBudget is now communicated to supported providers.
  • Chores

    • Settings UI reorganized to separate Gemini and Qwen3 thinking-budget sections and initialize per-model values.
    • Updated translations across locales to reflect the new structure and labels.

@coderabbitai
Copy link
Contributor

coderabbitai bot commented Sep 4, 2025

Walkthrough

Adds per-model thinkingBudget support for Qwen3 and extends Gemini handling: new Qwen3 thinking variants and thinkingBudget values in defaults and provider settings, propagation of thinking_budget to DashScope requests, UI additions/validation for Qwen3/Gemini thinking budgets, and i18n restructuring to separate gemini and qwen3 keys.

Changes

Cohort / File(s) Summary
Default model settings
src/main/presenter/configPresenter/modelDefaultSettings.ts
Adds two new Qwen3 thinking variants and enables functionCall / sets thinkingBudget on multiple Qwen3 entries; DefaultModelSetting now accepts optional thinkingBudget.
Provider model settings & config plumbing
src/main/presenter/configPresenter/providerModelSettings.ts
Adds thinkingBudget?: number to ProviderModelSetting; adds Qwen3 provider entries with thinkingBudget; getProviderSpecificModelConfig now propagates thinkingBudget into returned ModelConfig.
DashScope provider
src/main/presenter/llmProviderPresenter/providers/dashscopeProvider.ts
When enabling thinking for supported models, conditionally sets thinking_budget to modelConfig.thinkingBudget in request params (in addition to enable_thinking: true).
Renderer: Model config UI
src/renderer/src/components/settings/ModelConfigDialog.vue
Adds Qwen3-specific thinking-budget UI, validation, initialization, and separate Gemini/Qwen3 flows, labels, and error handling; integrates per-model min/max/defaults and disable hints.
i18n: restructure thinkingBudget (Gemini/Qwen3)
src/renderer/src/i18n/en-US/settings.json, .../fa-IR/settings.json, .../fr-FR/settings.json, .../ja-JP/settings.json, .../ko-KR/settings.json, .../ru-RU/settings.json, .../zh-CN/settings.json, .../zh-HK/settings.json, .../zh-TW/settings.json
Moves Gemini-related keys under thinkingBudget.gemini; introduces thinkingBudget.qwen3 with labels/validation; removes or repositions former flat keys; minor punctuation/content adjustments per locale.

Sequence Diagram(s)

sequenceDiagram
  autonumber
  actor User
  participant UI as Renderer (ModelConfigDialog)
  participant Presenter as Presenter (Config/Provider)
  participant Provider as DashScopeProvider
  participant API as DashScope API

  User->>UI: Select model and set thinkingBudget
  UI->>Presenter: Save model config (includes thinkingBudget)
  Presenter->>Presenter: getProviderSpecificModelConfig() merges thinkingBudget
  UI->>Provider: Start stream with ModelConfig
  Provider->>Provider: if reasoning supported -> enable_thinking=true
  alt thinkingBudget present
    Provider->>API: create(..., enable_thinking=true, thinking_budget=<value>)
  else
    Provider->>API: create(..., enable_thinking=true)
  end
  API-->>Provider: Stream responses
  Provider-->>UI: Forward tokens/events
  UI-->>User: Display output
Loading

Estimated code review effort

🎯 4 (Complex) | ⏱️ ~75 minutes

Possibly related PRs

Suggested reviewers

  • zerob13

Poem

A rabbit taps keys with a thoughtful grin,
Qwen3 now counts tokens where thoughts begin.
Gemini hums, DashScope sails the stream,
Budgets set, and reasoning lights the scheme.
Hop, hop, we test — small tweaks, big dream! 🐇✨

✨ Finishing Touches
🧪 Generate unit tests
  • Create PR with unit tests
  • Post copyable unit tests in a comment
  • Commit unit tests in branch feat/qwen3-thinking-budget

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share
🪧 Tips

Chat

There are 3 ways to chat with CodeRabbit:

  • Review comments: Directly reply to a review comment made by CodeRabbit. Example:
    • I pushed a fix in commit <commit_id>, please review it.
    • Open a follow-up GitHub issue for this discussion.
  • Files and specific lines of code (under the "Files changed" tab): Tag @coderabbitai in a new review comment at the desired location with your query.
  • PR comments: Tag @coderabbitai in a new PR comment to ask questions about the PR branch. For the best results, please provide a very specific query, as very limited context is provided in this mode. Examples:
    • @coderabbitai gather interesting stats about this repository and render them as a table. Additionally, render a pie chart showing the language distribution in the codebase.
    • @coderabbitai read the files in the src/scheduler package and generate a class diagram using mermaid and a README in the markdown format.

Support

Need help? Create a ticket on our support page for assistance with any issues or questions.

CodeRabbit Commands (Invoked using PR/Issue comments)

Type @coderabbitai help to get the list of available commands.

Other keywords and placeholders

  • Add @coderabbitai ignore or @coderabbit ignore anywhere in the PR description to prevent this PR from being reviewed.
  • Add @coderabbitai summary to generate the high-level summary at a specific location in the PR description.
  • Add @coderabbitai anywhere in the PR title to generate the title automatically.

CodeRabbit Configuration File (.coderabbit.yaml)

  • You can programmatically configure CodeRabbit by adding a .coderabbit.yaml file to the root of your repository.
  • Please see the configuration documentation for more information.
  • If your editor has YAML language server enabled, you can add the path at the top of this file to enable auto-completion and validation: # yaml-language-server: $schema=https://coderabbit.ai/integrations/schema.v2.json

Status, Documentation and Community

  • Visit our Status Page to check the current availability of CodeRabbit.
  • Visit our Documentation for detailed information on how to use CodeRabbit.
  • Join our Discord Community to get help, request features, and share feedback.
  • Follow us on X/Twitter for updates and announcements.

Copy link
Contributor

@coderabbitai coderabbitai bot left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Actionable comments posted: 3

🧹 Nitpick comments (6)
src/main/presenter/llmProviderPresenter/providers/dashscopeProvider.ts (3)

67-69: Sanitize and always pass numeric thinking_budget

Use a null-check (not truthy) and clamp to a positive integer to avoid passing NaN/float/0 accidentally.

-        if (modelConfig?.thinkingBudget) {
-          modifiedParams.thinking_budget = modelConfig.thinkingBudget
-        }
+        if (modelConfig?.thinkingBudget != null) {
+          const budget = Math.max(1, Math.floor(Number(modelConfig.thinkingBudget)))
+          modifiedParams.thinking_budget = budget
+        }

35-40: Avoid drift between supported-models list and provider settings

ENABLE_THINKING_MODELS duplicates model IDs already enumerated in providerModelSettings (dashscope/qwen3). Consider deriving this from a shared source or matching via provider settings to keep them in sync.


45-85: Monkey-patching create is service-wide; consider per-call injection

Overriding this.openai.chat.completions.create affects concurrent calls on the same instance. If feasible, refactor super.coreStream to accept request overrides or wrap the request locally to avoid global mutation.

src/renderer/src/components/settings/ModelConfigDialog.vue (2)

263-305: Qwen3 thinking-budget UI: adjust step granularity

Qwen3 max can be 20000 (not divisible by 128). Use step=1 for finer control.

-                  :step="128"
+                  :step="1"

566-640: Qwen3 config: align min/canDisable with validation/UI

Your UI enforces min=1 for Qwen3, but getThinkingBudgetConfig returns min: 0 and canDisable: true. Make them consistent.

-    return {
-      min: 0,
+    return {
+      min: 1,
       max: 81920,
       defaultValue: 81920,
-      canDisable: true
+      canDisable: false
     }

Apply similarly to 30B A3B, 32B/14B/8B/4B, and 1.7B/0.6B branches.

src/main/presenter/configPresenter/modelDefaultSettings.ts (1)

1271-1274: Behavior change: functionCall + reasoning enabled on qwen3-235b-a22b

Turning on functionCall/reasoning alters defaults for existing users. Confirm provider capability and consider a short CHANGELOG/notes entry so users aren’t surprised by new tool-calls on this model.

If this is intentional, add a brief inline comment to document the change:

   match: ['qwen3-235b-a22b', 'qwen/qwen3-235b-a22b'],
   vision: false,
-  functionCall: true,
-  reasoning: true,
+  functionCall: true, // Enabled: native tool-calls supported by DashScope for Qwen3 235B
+  reasoning: true,    // Enables thinking mode; see dashscopeProvider for thinking_budget
   thinkingBudget: 81920
📜 Review details

Configuration used: CodeRabbit UI

Review profile: CHILL

Plan: Pro

💡 Knowledge Base configuration:

  • MCP integration is disabled by default for public repositories
  • Jira integration is disabled by default for public repositories
  • Linear integration is disabled by default for public repositories

You can enable these sources in your CodeRabbit configuration.

📥 Commits

Reviewing files that changed from the base of the PR and between f8824f2 and 5fb6e77.

📒 Files selected for processing (13)
  • src/main/presenter/configPresenter/modelDefaultSettings.ts (9 hunks)
  • src/main/presenter/configPresenter/providerModelSettings.ts (3 hunks)
  • src/main/presenter/llmProviderPresenter/providers/dashscopeProvider.ts (1 hunks)
  • src/renderer/src/components/settings/ModelConfigDialog.vue (11 hunks)
  • src/renderer/src/i18n/en-US/settings.json (1 hunks)
  • src/renderer/src/i18n/fa-IR/settings.json (1 hunks)
  • src/renderer/src/i18n/fr-FR/settings.json (1 hunks)
  • src/renderer/src/i18n/ja-JP/settings.json (1 hunks)
  • src/renderer/src/i18n/ko-KR/settings.json (1 hunks)
  • src/renderer/src/i18n/ru-RU/settings.json (1 hunks)
  • src/renderer/src/i18n/zh-CN/settings.json (1 hunks)
  • src/renderer/src/i18n/zh-HK/settings.json (1 hunks)
  • src/renderer/src/i18n/zh-TW/settings.json (1 hunks)
🧰 Additional context used
📓 Path-based instructions (18)
**/*.{js,jsx,ts,tsx}

📄 CodeRabbit inference engine (.cursor/rules/development-setup.mdc)

**/*.{js,jsx,ts,tsx}: 使用 OxLint 进行代码检查
Log和注释使用英文书写

Files:

  • src/main/presenter/llmProviderPresenter/providers/dashscopeProvider.ts
  • src/main/presenter/configPresenter/providerModelSettings.ts
  • src/main/presenter/configPresenter/modelDefaultSettings.ts
src/{main,renderer}/**/*.ts

📄 CodeRabbit inference engine (.cursor/rules/electron-best-practices.mdc)

src/{main,renderer}/**/*.ts: Use context isolation for improved security
Implement proper inter-process communication (IPC) patterns
Optimize application startup time with lazy loading
Implement proper error handling and logging for debugging

Files:

  • src/main/presenter/llmProviderPresenter/providers/dashscopeProvider.ts
  • src/main/presenter/configPresenter/providerModelSettings.ts
  • src/main/presenter/configPresenter/modelDefaultSettings.ts
src/main/**/*.ts

📄 CodeRabbit inference engine (.cursor/rules/electron-best-practices.mdc)

Use Electron's built-in APIs for file system and native dialogs

Files:

  • src/main/presenter/llmProviderPresenter/providers/dashscopeProvider.ts
  • src/main/presenter/configPresenter/providerModelSettings.ts
  • src/main/presenter/configPresenter/modelDefaultSettings.ts
**/*.{ts,tsx}

📄 CodeRabbit inference engine (.cursor/rules/error-logging.mdc)

**/*.{ts,tsx}: 始终使用 try-catch 处理可能的错误
提供有意义的错误信息
记录详细的错误日志
优雅降级处理
日志应包含时间戳、日志级别、错误代码、错误描述、堆栈跟踪(如适用)、相关上下文信息
日志级别应包括 ERROR、WARN、INFO、DEBUG
不要吞掉错误
提供用户友好的错误信息
实现错误重试机制
避免记录敏感信息
使用结构化日志
设置适当的日志级别

Enable and adhere to strict TypeScript type checking across the codebase

Files:

  • src/main/presenter/llmProviderPresenter/providers/dashscopeProvider.ts
  • src/main/presenter/configPresenter/providerModelSettings.ts
  • src/main/presenter/configPresenter/modelDefaultSettings.ts
src/main/presenter/llmProviderPresenter/providers/*.ts

📄 CodeRabbit inference engine (.cursor/rules/llm-agent-loop.mdc)

src/main/presenter/llmProviderPresenter/providers/*.ts: Each file in src/main/presenter/llmProviderPresenter/providers/*.ts should handle interaction with a specific LLM API, including request/response formatting, tool definition conversion, native/non-native tool call management, and standardizing output streams to a common event format.
Provider implementations must use a coreStream method that yields standardized stream events to decouple the main loop from provider-specific details.
The coreStream method in each Provider must perform a single streaming API request per conversation round and must not contain multi-round tool call loop logic.
Provider files should implement helper methods such as formatMessages, convertToProviderTools, parseFunctionCalls, and prepareFunctionCallPrompt as needed for provider-specific logic.
All provider implementations must parse provider-specific data chunks and yield standardized events for text, reasoning, tool calls, usage, errors, stop reasons, and image data.
When a provider does not support native function calling, it must prepare messages using prompt wrapping (e.g., prepareFunctionCallPrompt) before making the API call.
When a provider supports native function calling, MCP tools must be converted to the provider's format (e.g., using convertToProviderTools) and included in the API request.
Provider implementations should aggregate and yield usage events as part of the standardized stream.
Provider implementations should yield image data events in the standardized format when applicable.
Provider implementations should yield reasoning events in the standardized format when applicable.
Provider implementations should yield tool call events (tool_call_start, tool_call_chunk, tool_call_end) in the standardized format.
Provider implementations should yield stop events with appropriate stop_reason in the standardized format.
Provider implementations should yield error events in the standardized format...

Files:

  • src/main/presenter/llmProviderPresenter/providers/dashscopeProvider.ts
src/main/**/*.{ts,js,tsx,jsx}

📄 CodeRabbit inference engine (.cursor/rules/project-structure.mdc)

主进程代码放在 src/main

Files:

  • src/main/presenter/llmProviderPresenter/providers/dashscopeProvider.ts
  • src/main/presenter/configPresenter/providerModelSettings.ts
  • src/main/presenter/configPresenter/modelDefaultSettings.ts
**/*.{ts,tsx,js,jsx,vue}

📄 CodeRabbit inference engine (CLAUDE.md)

Write logs and comments in English

Files:

  • src/main/presenter/llmProviderPresenter/providers/dashscopeProvider.ts
  • src/renderer/src/components/settings/ModelConfigDialog.vue
  • src/main/presenter/configPresenter/providerModelSettings.ts
  • src/main/presenter/configPresenter/modelDefaultSettings.ts
src/main/presenter/**/*.ts

📄 CodeRabbit inference engine (CLAUDE.md)

Maintain one Presenter per functional domain under src/main/presenter/

Files:

  • src/main/presenter/llmProviderPresenter/providers/dashscopeProvider.ts
  • src/main/presenter/configPresenter/providerModelSettings.ts
  • src/main/presenter/configPresenter/modelDefaultSettings.ts
src/renderer/src/**/*

📄 CodeRabbit inference engine (.cursor/rules/i18n.mdc)

src/renderer/src/**/*: All user-facing strings must use i18n keys (avoid hardcoded user-visible text in code)
Use the 'vue-i18n' framework for all internationalization in the renderer
Ensure all user-visible text in the renderer uses the translation system

Files:

  • src/renderer/src/components/settings/ModelConfigDialog.vue
  • src/renderer/src/i18n/zh-HK/settings.json
  • src/renderer/src/i18n/fr-FR/settings.json
  • src/renderer/src/i18n/fa-IR/settings.json
  • src/renderer/src/i18n/ja-JP/settings.json
  • src/renderer/src/i18n/zh-CN/settings.json
  • src/renderer/src/i18n/en-US/settings.json
  • src/renderer/src/i18n/ko-KR/settings.json
  • src/renderer/src/i18n/ru-RU/settings.json
  • src/renderer/src/i18n/zh-TW/settings.json
src/renderer/**/*.{vue,ts,js,tsx,jsx}

📄 CodeRabbit inference engine (.cursor/rules/project-structure.mdc)

渲染进程代码放在 src/renderer

Files:

  • src/renderer/src/components/settings/ModelConfigDialog.vue
src/renderer/src/**/*.{vue,ts,tsx,js,jsx}

📄 CodeRabbit inference engine (.cursor/rules/vue-best-practices.mdc)

src/renderer/src/**/*.{vue,ts,tsx,js,jsx}: Use the Composition API for better code organization and reusability
Implement proper state management with Pinia
Utilize Vue Router for navigation and route management
Leverage Vue's built-in reactivity system for efficient data handling

Files:

  • src/renderer/src/components/settings/ModelConfigDialog.vue
src/renderer/src/**/*.vue

📄 CodeRabbit inference engine (.cursor/rules/vue-best-practices.mdc)

Use scoped styles to prevent CSS conflicts between components

Files:

  • src/renderer/src/components/settings/ModelConfigDialog.vue
src/renderer/**/*.{ts,tsx,vue}

📄 CodeRabbit inference engine (.cursor/rules/vue-shadcn.mdc)

src/renderer/**/*.{ts,tsx,vue}: Use descriptive variable names with auxiliary verbs (e.g., isLoading, hasError).
Use TypeScript for all code; prefer types over interfaces.
Avoid enums; use const objects instead.
Use arrow functions for methods and computed properties.
Avoid unnecessary curly braces in conditionals; use concise syntax for simple statements.

Files:

  • src/renderer/src/components/settings/ModelConfigDialog.vue
src/renderer/**/*.{vue,ts}

📄 CodeRabbit inference engine (.cursor/rules/vue-shadcn.mdc)

Implement lazy loading for routes and components.

Files:

  • src/renderer/src/components/settings/ModelConfigDialog.vue
src/renderer/**/*.{ts,vue}

📄 CodeRabbit inference engine (.cursor/rules/vue-shadcn.mdc)

src/renderer/**/*.{ts,vue}: Use useFetch and useAsyncData for data fetching.
Implement SEO best practices using Nuxt's useHead and useSeoMeta.

Use Pinia for frontend state management

Files:

  • src/renderer/src/components/settings/ModelConfigDialog.vue
src/renderer/**/*.vue

📄 CodeRabbit inference engine (CLAUDE.md)

src/renderer/**/*.vue: Use Vue 3 Composition API for all Vue components
Use Tailwind CSS with scoped styles for component styling
When creating new UI components, use Composition API with proper TypeScript typing, ensure responsive design (Tailwind), and include error/loading states

Files:

  • src/renderer/src/components/settings/ModelConfigDialog.vue
src/renderer/src/**

📄 CodeRabbit inference engine (CLAUDE.md)

Place full application UI functionality (tab content) under src/renderer/src/ organized by feature

Files:

  • src/renderer/src/components/settings/ModelConfigDialog.vue
  • src/renderer/src/i18n/zh-HK/settings.json
  • src/renderer/src/i18n/fr-FR/settings.json
  • src/renderer/src/i18n/fa-IR/settings.json
  • src/renderer/src/i18n/ja-JP/settings.json
  • src/renderer/src/i18n/zh-CN/settings.json
  • src/renderer/src/i18n/en-US/settings.json
  • src/renderer/src/i18n/ko-KR/settings.json
  • src/renderer/src/i18n/ru-RU/settings.json
  • src/renderer/src/i18n/zh-TW/settings.json
src/main/presenter/configPresenter/**

📄 CodeRabbit inference engine (CLAUDE.md)

Centralize configuration logic under configPresenter/

Files:

  • src/main/presenter/configPresenter/providerModelSettings.ts
  • src/main/presenter/configPresenter/modelDefaultSettings.ts
🧠 Learnings (3)
📚 Learning: 2025-09-01T09:16:58.043Z
Learnt from: CR
PR: ThinkInAIXYZ/deepchat#0
File: CLAUDE.md:0-0
Timestamp: 2025-09-01T09:16:58.043Z
Learning: Applies to src/main/presenter/configPresenter/providers.ts : Add new provider configuration entries in configPresenter/providers.ts

Applied to files:

  • src/main/presenter/configPresenter/providerModelSettings.ts
📚 Learning: 2025-07-21T01:46:30.354Z
Learnt from: CR
PR: ThinkInAIXYZ/deepchat#0
File: .cursor/rules/i18n.mdc:0-0
Timestamp: 2025-07-21T01:46:30.354Z
Learning: Applies to src/renderer/src/i18n/*.json : Maintain consistent structure across all translation files

Applied to files:

  • src/renderer/src/i18n/fr-FR/settings.json
  • src/renderer/src/i18n/fa-IR/settings.json
  • src/renderer/src/i18n/zh-CN/settings.json
  • src/renderer/src/i18n/ko-KR/settings.json
📚 Learning: 2025-07-21T01:46:30.354Z
Learnt from: CR
PR: ThinkInAIXYZ/deepchat#0
File: .cursor/rules/i18n.mdc:0-0
Timestamp: 2025-07-21T01:46:30.354Z
Learning: Applies to src/renderer/src/i18n/common.json : Shared translation keys must be placed in 'src/renderer/src/i18n/common.json'

Applied to files:

  • src/renderer/src/i18n/fa-IR/settings.json
⏰ Context from checks skipped due to timeout of 90000ms. You can increase the timeout in your CodeRabbit configuration to a maximum of 15 minutes (900000ms). (1)
  • GitHub Check: build-check (x64)
🔇 Additional comments (30)
src/renderer/src/i18n/en-US/settings.json (3)

144-168: Gemini thinkingBudget block looks good; structure matches new schema.

No issues spotted with keys or placeholders.


169-178: Qwen3 thinkingBudget strings present and consistent.

Validation messages and placeholders look correct.


141-179: Manual parity check required for thinkingBudget keys across locales
Automated verification script failed in this environment; manually confirm that every locale’s thinkingBudget.gemini and thinkingBudget.qwen3 sections contain exactly the same keys as in en-US/settings.json to avoid runtime fallbacks.

src/renderer/src/i18n/ja-JP/settings.json (2)

144-168: Gemini 思考予算: OK.

Keys mirror en-US; placeholders and warnings/hints read well.


169-178: Qwen3 思考予算: OK.

Validation block and copy are consistent.

src/renderer/src/i18n/zh-CN/settings.json (1)

176-185: Qwen3 思考预算: OK.

Validation messages and placeholders are correct.

src/renderer/src/i18n/fa-IR/settings.json (2)

144-167: بلوک جمینی: خوب است.

ساختار و کلیدها مطابق طرح جدید است.


169-178: بلوک Qwen3: خوب است.

اعتبارسنجی و متن‌ها درست‌اند.

src/renderer/src/i18n/ko-KR/settings.json (2)

144-167: Gemini 사고 예산: OK.

키/자리표시자 일치합니다.


169-178: Qwen3 사고 예산: OK.

검증 메시지/플레이스홀더 이상 없음.

src/renderer/src/i18n/zh-HK/settings.json (2)

144-167: Gemini thinkingBudget i18n restructure — LGTM

Keys and wording align with ModelConfigDialog.vue usage (gemini.*).


169-177: All locales define Qwen3 thinkingBudget translation keys

src/renderer/src/i18n/zh-TW/settings.json (2)

144-167: Gemini thinkingBudget i18n restructure — LGTM

Matches component expectations (gemini.* keys, warnings/hints present).


169-177: Confirmed parity for qwen3 translation keys (including validation messages) across all locales.

src/main/presenter/configPresenter/providerModelSettings.ts (3)

2476-2597: DashScope Qwen3 entries and defaults — LGTM

Ordering (specific “thinking-2507” first), match keys, and default thinkingBudget values align with UI ranges.


2960-2972: Propagate thinkingBudget to ModelConfig — LGTM

Mapping includes maxCompletionTokens and thinkingBudget correctly.


20-21: thinkingBudget field present in ModelConfig and handled in persistence
The ModelConfig interface in src/shared/presenter.d.ts declares thinkingBudget?: number, and the SQLite presenter tables map it to/from the thinking_budget column — no further action required.

src/renderer/src/components/settings/ModelConfigDialog.vue (8)

199-262: Gemini thinking-budget UI path — LGTM

Uses gemini.* keys, dynamic toggle, and validations as expected.


475-484: Default Qwen3 thinkingBudget initialization — verify store precedence

You only set when undefined, which is good. Confirm settingsStore.getModelConfig doesn’t already return a provider default to avoid double-defaulting.


515-523: Form validity wiring — LGTM

Includes both Gemini and Qwen3 budget errors in isValid.


655-662: Gemini visibility condition — LGTM


664-672: Qwen3 visibility condition — LGTM

Scoped to DashScope + reasoning + supported Qwen3 IDs.


680-729: Gemini budget validation — LGTM

Covers dynamic (-1), zero-disable rules, and min/max per model.


730-751: Qwen3 budget validation — LGTM

Matches i18n keys and enforces 1..max.


765-769: Disable-hint helper bound to Gemini — LGTM

src/main/presenter/configPresenter/modelDefaultSettings.ts (1)

1276-1286: Add Qwen3 30B thinking variant: double-check budget/value

Same notes as the 235B thinking SKU: confirm 81920 is accepted by provider and aligns with UI max for this model.

src/renderer/src/i18n/ru-RU/settings.json (2)

144-168: Gemini thinkingBudget moved under thinkingBudget.gemini: structure LGTM

Keys (description/dynamic/warnings/hints, etc.) match the new UI usage. Nice.


169-178: Approve changes: Qwen3 thinkingBudget entries (description, valueLabel, placeholder, validation.required/minValue/maxValue) are present and consistent across all locales and align with ModelConfigDialog.vue expectations.

src/renderer/src/i18n/fr-FR/settings.json (2)

145-167: Gemini thinkingBudget nested under gemini: OK

All required subkeys are present and consistent with other locales.


170-177: Qwen3 thinkingBudget block with validation: OK

Matches the new UI logic for Qwen3 budget input and errors.

Same cross-locale key check as RU to ensure consistency (see script in RU comment).

Comment on lines +1251 to +1262
{
id: 'qwen3-235b-a22b-thinking-2507',
name: 'Qwen3 235B A22B Thinking 2507',
temperature: 0.6,
maxTokens: 8192,
contextLength: 40960,
match: ['qwen3-235b-a22b-thinking-2507', 'qwen/qwen3-235b-a22b-thinking-2507'],
vision: false,
functionCall: true,
reasoning: true,
thinkingBudget: 81920
},
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

💡 Verification agent

🧩 Analysis chain

Add Qwen3 235B thinking variant: verify provider limits and FC support

thinkingBudget 81920 may exceed provider/model limits; also toggling functionCall to true assumes native tool-calls are available. Please verify DashScope accepts this budget and function calling for this SKU.

Run this to cross-check budgets and provider wiring:


🏁 Script executed:

#!/bin/bash
set -euo pipefail

# Ensure these IDs exist in provider model settings and expose budgets
rg -n "qwen3-235b-a22b-thinking-2507" -C3 src/main/presenter/configPresenter/providerModelSettings.ts || true

# Confirm DashScope provider forwards "thinking_budget"
rg -n "thinking_budget" src/main/presenter/llmProviderPresenter/providers/dashscopeProvider.ts -C3 || true

Length of output: 872


🏁 Script executed:

#!/bin/bash
set -euo pipefail

# Confirm DashScope provider forwards function calling flags
rg -n "function_call" -C3 src/main/presenter/llmProviderPresenter/providers/dashscopeProvider.ts || true
rg -n "functionCall" -C3 src/main/presenter/llmProviderPresenter/providers/dashscopeProvider.ts || true

Length of output: 226


Propagate functionCall flag in DashScope provider and confirm thinkingBudget limit

  • In dashscopeProvider.ts, you already map modelConfig.thinkingBudgetthinking_budget, but there’s no mapping for modelConfig.functionCall. Add something like:
    if (modelConfig?.functionCall) {
      modifiedParams.function_call = true
    }
  • Confirm that a thinkingBudget of 81920 is supported for qwen3-235b-a22b-thinking-2507 by DashScope (provider/model limits).
🤖 Prompt for AI Agents
In src/main/presenter/configPresenter/modelDefaultSettings.ts around lines 1251
to 1262, add mapping and validation so DashScope receives the model's
functionCall and a supported thinkingBudget: when building modifiedParams from
modelConfig, if modelConfig?.functionCall is truthy set
modifiedParams.function_call = true; ensure you already map
modelConfig.thinkingBudget → modifiedParams.thinking_budget and then validate
that the value (81920) is within DashScope/provider/model limits—if unsupported,
clamp to the provider max or replace with a safe default and/or surface an error
so the provider isn't sent an out-of-range thinking_budget.

@zerob13 zerob13 merged commit 89ebbed into dev Sep 4, 2025
1 of 2 checks passed
Copy link
Contributor

@coderabbitai coderabbitai bot left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Actionable comments posted: 0

♻️ Duplicate comments (1)
src/renderer/src/i18n/zh-CN/settings.json (1)

152-175: Gemini thinkingBudget 文案与结构对齐且补齐 onlySupported,LGTM

已包含 onlySupported,键位与其他语言版本保持一致,警告与提示枚举完整。

🧹 Nitpick comments (1)
src/renderer/src/i18n/zh-CN/settings.json (1)

177-186: Qwen3:占位语与校验提示微调为参数化与一致口吻

为与同文件其他占位语风格统一,建议将占位语改为“请输入…”。同时将最小值提示改为可参数化的“{min}”,避免与实际最小值配置不一致。

         "qwen3": {
           "description": "设置思考过程的最大token长度,用于控制模型推理时的思考深度",
           "valueLabel": "思考预算值",
-          "placeholder": "输入思考预算值",
+          "placeholder": "请输入思考预算值",
           "validation": {
             "required": "请输入思考预算值",
-            "minValue": "思考预算值必须大于等于1",
+            "minValue": "思考预算值必须大于等于 {min}",
             "maxValue": "思考预算值不能超过 {max}"
           }
         }
📜 Review details

Configuration used: CodeRabbit UI

Review profile: CHILL

Plan: Pro

💡 Knowledge Base configuration:

  • MCP integration is disabled by default for public repositories
  • Jira integration is disabled by default for public repositories
  • Linear integration is disabled by default for public repositories

You can enable these sources in your CodeRabbit configuration.

📥 Commits

Reviewing files that changed from the base of the PR and between 5fb6e77 and 2b4c633.

📒 Files selected for processing (1)
  • src/renderer/src/i18n/zh-CN/settings.json (1 hunks)
🧰 Additional context used
📓 Path-based instructions (2)
src/renderer/src/**/*

📄 CodeRabbit inference engine (.cursor/rules/i18n.mdc)

src/renderer/src/**/*: All user-facing strings must use i18n keys (avoid hardcoded user-visible text in code)
Use the 'vue-i18n' framework for all internationalization in the renderer
Ensure all user-visible text in the renderer uses the translation system

Files:

  • src/renderer/src/i18n/zh-CN/settings.json
src/renderer/src/**

📄 CodeRabbit inference engine (CLAUDE.md)

Place full application UI functionality (tab content) under src/renderer/src/ organized by feature

Files:

  • src/renderer/src/i18n/zh-CN/settings.json
⏰ Context from checks skipped due to timeout of 90000ms. You can increase the timeout in your CodeRabbit configuration to a maximum of 15 minutes (900000ms). (1)
  • GitHub Check: build-check (x64)

zerob13 added a commit that referenced this pull request Sep 9, 2025
* fix: chat confg need sync to new value when change model in chat (#823)

* fix: gemini nano banana not read image from chatinput

* fix: remove file-type ,this will mark html as audio (#824)

* fix: Solve the problem of the window becoming larger when dragging floating button under Windows (#826)

* fix: improve OpenAI compatible provider compatibility with third-party services

* chore: update vue-renderer-markdown to v0.0.35 (#827)

* refactor: remove custom-prompts-server and decouple prompts from MCP lifecycle (#829)

- Remove custom-prompts-server service entirely including configuration
- Implement data source merging in MCP store to load prompts from both config and MCP
- Add upgrade migration logic for versions < 0.3.5 to clean up old configurations
- Ensure @ operations work independently of MCP state through config data source
- Update CLAUDE.md with prompt management guidelines

The @ prompt functionality now works completely independently of MCP,
loading custom prompts directly from config storage while maintaining
full compatibility with existing MCP prompt sources.

* chore: add better log for mcp tool name

* feat: ux update (#831)

* feat: ux update

* chore: format

* feat: setting provider ux update (#832)

* feat: add current datetime to system prompt

- Add current date and time information to user's system prompt when not empty
- Include complete datetime with timezone, year, month, day, hour, minute, second
- Apply to both preparePromptContent and buildContinueToolCallContext methods
- Update token calculation to use processed system prompt for accuracy
- Skip datetime addition for image generation models and empty prompts

* refactor: extract system prompt datetime enhancement to common method

- Add enhanceSystemPromptWithDateTime private method to reduce code duplication
- Update both preparePromptContent and buildContinueToolCallContext to use common method
- Improve code maintainability and ensure consistency across all system prompt processing
- Add comprehensive JSDoc documentation for the new method

* fix(markdown): auto-wrap hide scroll (#833)

* feat: add enable_thinking parameter support for siliconcloud (#835)

* chore: bump deps (#834)

* chore: bump up deps

* fix: change win arm to wasm32 sharp

* chore: revert sharp config

* feat: drop windows arm support

* fix(coderabbitai): remove action for windows arm64

* refactor: adjust scroll-to-bottom button glow effect (#837)

* feat: add mutual exclusive confirmation dialogs for DeepSeek-V3.1 (#838)

* feat: add sanitizeText utility for clipboard data handling (#843)

* feat: support canary upgrade (#840)

* feat: support canary upgrade

* feat: add update channel selection for stable/canary versions

- Add update channel configuration to config presenter
- Implement dynamic URL switching for version checks and downloads
- Add UI selector in AboutUsSettings for channel selection
- Support stable and canary update channels with different endpoints
- Add internationalization support for channel selection

* chore: change contributors charts to openomy

* refactor: improve update channel handling and network resilience

- Sanitize update channel input to prevent invalid values
- Add 10-second timeout to version check requests to prevent hanging
- Treat unknown channels as 'upgrade' (stable) for robustness
- Improve error handling for network timeouts and invalid channels

* feat: implement thinking parameter support for doubao models (#842)

* feat: implement dedicated DashScope provider with enable_thinking support (#844)

* feat: implement dedicated DashScope provider with enable_thinking support

* refactor: remove unnecessary API key status check methods from DashscopeProvider

* fix: prioritize provider.id over apiType in createProviderInstance (#846)

* feat: add qwen3 thinking budget support (#848)

* feat: add qwen3 thinking budget support

* fix: add missing gemini.onlySupported key in zh-CN locale

* refactor: merge duplicate silicon case statements in provider creation

* feat: add qwen3 thinking budget support in ChatConfig (#849)

* refactor(types): 🚀split monolithic presenter.d.ts into strict per-domain *.presenter.d.ts + typed core layer  (#847)

* docs: Add strong-typed message architecture and implementation guidelines

- Update message-architecture.md with strong-typed design, remove compatibility compromises
- Add event-to-UI mapping table and rendering checklist for contract compliance
- Create presenter-split-plan.md for type system refactoring
- Add implementation-tasks.md with phased rollout plan
- Create .cursor/rules/provider-guidelines.mdc for provider implementation guidance

This establishes a foundation for strong-typed, maintainable message architecture without legacy compatibility debt.

* types(core): add strong-typed core types and barrel exports\n\n- Add usage.ts (UsageStats, RateLimitInfo)\n- Add llm-events.ts (discriminated union + factories + guards)\n- Add agent-events.ts (LLMAgentEvent*, shared types)\n- Add chat.ts (Message/AssistantMessageBlock/UserMessageContent)\n- Add mcp.ts (MCP content/response/definition)\n- Add types/index.d.ts barrel exports\n\nNo compatibility shims included by design.

* refactor(types): move legacy presenters and add @shared/presenter stub; keep legacy exports in types/index to maintain build\n\n- Move legacy presenters to src/shared/types/presenters/legacy.presenters.d.ts\n- Add src/shared/presenter.d.ts re-export stub\n- Temporarily export only legacy presenters from types/index.d.ts to avoid type mismatches\n- Update implementation-tasks.md (Stage 2 done)\n\nNext: progressive import updates to new cores, then flip barrel to new types and delete legacy.

* refactor(types): alias legacy core message types to strong-typed core (B-plan)\n\n- legacy.presenters.d.ts now re-exports ChatMessage/ChatMessageContent/LLMAgentEvent/LLMAgentEventData/LLMCoreStreamEvent from core\n- Prepares for flipping interfaces without changing import sites

* docs(tasks): mark Phase 3 as completed\n\n- Successfully implemented B-plan approach with type aliasing\n- Unified core message types (ChatMessage, LLMAgentEvent, LLMCoreStreamEvent)\n- Created core model types and maintained build compatibility\n- All type checks passing with no breaking changes

* fix(types): revert to legacy-only exports and fix MESSAGE_ROLE\n\n- Revert types/index.d.ts to only export legacy presenters\n- Remove 'function' from MESSAGE_ROLE to match core definition\n- Maintain build stability while preserving type unification work

* feat(provider): implement factory functions for LLMCoreStreamEvent\n\n- Replace all manual event creation with createStreamEvent factory functions\n- Ensures type safety and consistent event structure\n- Updated OpenAICompatibleProvider with strong-typed events:\n  - text, reasoning, toolCallStart, toolCallChunk, toolCallEnd\n  - error, usage, stop, imageData events\n- All type checks passing\n- Phase 4.1 completed: Provider now outputs LLMCoreStreamEvent with factory construction

* feat(provider): update OllamaProvider with factory functions\n\n- Replace all manual event creation with createStreamEvent factory functions\n- Ensures consistent tool_call_start → tool_call_chunk → tool_call_end sequence\n- Updated all event types: text, reasoning, toolCall*, usage, stop, error\n- Maintains proper tool call ID aggregation and sequencing\n- Phase 4.2 completed: Tool call sequences now strictly follow start/chunk/end pattern

* docs(tasks): update Phase 4 progress\n\n- Completed Provider strong-typed event output with factory functions\n- Completed tool_call_* strict start/chunk/end sequences\n- Completed usage event sent before stop in all providers\n- Updated implementation tasks document with current progress

* feat(phase4): complete Provider strong-typed event integration\n\n- Added factory functions import to AwsBedrockProvider\n- Updated error handling to use createStreamEvent.error() + createStreamEvent.stop('error')\n- Created comprehensive unit tests for LLMCoreStreamEvent factory functions\n- Tests validate: event creation, tool call sequences, error+stop patterns, type guards\n- All 12 core event tests passing ✅\n\n## Phase 4 Completed:\n- ✅ Provider strong-typed event output (factory construction)\n- ✅ tool_call_* strict start/chunk/end sequences with ID aggregation\n- ✅ Usage events sent before stop in all providers\n- ✅ Rate limit events (handled at Presenter layer)\n- ✅ Unified error + stop event patterns\n- ✅ Comprehensive unit tests for sequences/fields/boundaries

* feat(phase5): complete Agent & UI alignment

## Phase 5 Completed: Agent & UI Alignment ✅

### 1. Agent Layer Enhancement
- ✅ Agent consumes only CoreEvent and outputs LLMAgentEvent
- ✅ Strict separation of response/error/end event types
- ✅ Added rate_limit event handling in Agent layer
- ✅ Maintained proper event transformation pipeline

### 2. UI Layer Unification
- ✅ Removed standalone tool_call_permission type
- ✅ Unified to action + action_type pattern
- ✅ Updated all UI components and thread presenter logic
- ✅ Preserved backward compatibility with conditional rendering

### 3. Renderer Mapping & Testing
- ✅ Created comprehensive renderer contract tests (17 tests)
- ✅ Added message block data structure snapshot tests (14 tests)
- ✅ Validated event-to-UI block mapping compliance
- ✅ Ensured consistent status transitions and type safety

### Key Technical Achievements:
- **Type Safety**: All event transformations now type-safe
- **Consistency**: Unified action_type handling across codebase
- **Testing**: 31 new tests covering contracts and snapshots
- **Architecture**: Clean separation between Agent/UI concerns
- **Quality**: All lint, format, and type checks passing ✅

### Files Modified:
- Agent layer: llmProviderPresenter/index.ts
- UI components: MessageItemAssistant.vue
- Thread logic: threadPresenter/index.ts
- Type definitions: shared/chat.d.ts
- Tests: rendererContract.test.ts, messageBlockSnapshot.test.ts

Ready for Phase 6: Quality & Tools 🚀

* feat(phase6): complete Quality & Tools phase

## Phase 6 Completed: Quality & Tools ✅

### 1. Table-Driven Contract Tests
- ✅ Created comprehensive event mapping contract tests (25 test cases)
- ✅ Validated all event types from architecture mapping table
- ✅ Verified status transitions and type constraints
- ✅ Added tool call ID aggregation and lifecycle validation
- ✅ Implemented error recovery pattern testing

### 2. Performance Evaluation & Stress Testing
- ✅ Large text content processing (50KB < 10ms)
- ✅ Large reasoning content handling (100KB < 15ms)
- ✅ Large image data processing (1MB < 20ms)
- ✅ Batch image processing (100 images < 50ms)
- ✅ High-frequency event processing (1000 events < 100ms)
- ✅ Mixed event type handling (500 events < 80ms)
- ✅ Memory leak prevention (5000 events < 500ms)
- ✅ Extreme parameter handling (10KB params < 5ms)
- ✅ Concurrent processing simulation (10 workers < 100ms)

### Key Technical Achievements:
- **Comprehensive Testing**: 67 tests passing across all scenarios
- **Performance Validation**: All benchmarks meet performance targets
- **Type Safety**: Full TypeScript compliance (0 errors)
- **Code Quality**: Lint and format checks passing ✅
- **Architecture Compliance**: All mapping table rules verified
- **Stress Testing**: System handles extreme loads efficiently

### Test Coverage Summary:
- Event mapping contract tests: 25 tests ✅
- Renderer contract tests: 17 tests ✅
- Performance evaluation tests: 9 tests ✅
- Core event factory tests: 12 tests ✅
- Message block snapshot tests: 14 tests ✅
- Shell integration tests: 8 tests ✅

### Files Added:
- test/renderer/message/eventMappingTable.test.ts (comprehensive mapping validation)
- test/renderer/message/performanceEvaluation.test.ts (stress & performance testing)

Ready for production deployment with full quality assurance! 🚀

* fix(providers): complete strong-typed event integration across all providers

* fix(vitest): modify test case

* fix: default settings

* chore: update doc

* fix(ci): remove duplicate check in pr ci

* feat: add pnpm cache for pr check

* fix(ci): pr check with pnpm cache

* fix(ci): change cache key to package.json

* ci: remove pnpm cache

* feat: add glow breathing effect to scroll-to-bottom button (#850)

* feat: add glow breathing effect to scroll-to-bottom button

* fix: ensure exclusive display between MessageList and ArtifactDialog

* fix: refine MessageList–ArtifactDialog interaction logic; correct z-order between dialog and ArtifactDialog

* chore: prettier .vue

* feat: add web search support with configurable options for dashscope (#851)

* feat: add web search support with configurable options for dashscope

* fix: correct qwen model parameters to match official documentation

* feat: add web search support with configurable options for dashscope (#852)

* feat: add web search support with configurable options for dashscope

* fix: correct qwen model parameters to match official documentation

* feat: add search configuration support to ChatConfig components

* fix: fix enableSearch state sync and parameter passing issues

* fix: preserve search settings during data import

* feat: add dashscope commercial models to enable_thinking support (#853)

* feat: add search capability icon for model list (#854)

* feat: add search capability icon for model list

* fix: clear search settings when creating new conversation

* feat(markdown): Thinking panel now supports LaTeX compilation for mathematical formulas & markdown performance optimization (#857)

* feat(markdown): 思考栏支持数学公式latex编译显示 & markdown 性能优化
close: #845

* chore: lint

* chore(ai): update claude code rules and agents

* fix(ui): revert Dialog z-index to z-50 to fix dropdown visibility

Reverts DialogContent z-index from z-[100] back to z-50 to resolve issue where Select and EmojiPicker dropdowns were not appearing. This maintains proper layering hierarchy without breaking other UI components.

* feat: upgrade vue-renderer-markdown & vue-use-monaco (#862)

1. ignore math-block warning
2. Compatible with the syntax issues of mermaid produced by AI, greatly reducing the probability of mermaid rendering errors

* feat(dashscope): add qwen3-max-preview model (#865)

* fix: mcp params support more types (#861)

* feat(mcp): enhance tool parameter display with enum type support

- Add enum parameter type detection and enhanced display
- Show enum parameters with distinct blue badge styling (enum(string), array[enum(string)])
- Display allowed values for both direct enum and array item enum parameters
- Add i18n support for "allowedValues" and "arrayItemValues" labels
- Maintain consistent UI design with existing parameter display patterns
- Improve developer experience when debugging MCP tools with constrained parameters

* fix: enum params support

* fix(context-menu): handle local file paths in image save functionality

- Fix URL parsing error when saving images from local file paths
- Add proper handling for http/https URLs, file:// URLs, and direct file paths
- Use fs.promises for reading local files instead of net.fetch for invalid URLs
- Prevent "Failed to parse URL from" error when saving local images

* fix(context-menu): improve URL handling robustness in image save

- Add try-catch around net.fetch to handle invalid URLs gracefully
- Implement fallback methods for file:// URLs and local file paths
- Add debug logging to track source URL values for troubleshooting
- Prevent "Failed to parse URL from" errors with comprehensive URL validation

* fix(context-menu): handle empty srcURL in image save functionality

- Add comprehensive URL detection when srcURL is empty
- Implement fallback URL sources (linkURL, pageURL) for better compatibility
- Add debug logging to track all available context menu parameters
- Prevent "Failed to parse URL from" errors caused by empty URLs
- Provide clear error message when no valid URL can be found

* chore: format code

* fix: ai review

* fix: prevent @ symbol remaining when deleting mentions (#867)

* Merge commit from fork

* feat: implement separated system and custom prompt management (#868)

* feat: implement separated system and custom prompt management

* style: code fmt

* fix: add migration for legacy default_system_prompt to system_prompts

* feat: add Moonshot model configurations (#869)

* refactor: translate all cn comments and log to en (#871)

* refactor: translate all cn comments and log to en

* fix: revert translate in params

* feat: add reasoning support for Grok thinking models (#873)

* feat: add reasoning support for Grok thinking models

* fix: code lint

* fix: escaping character issue

---------

Co-authored-by: zerob13 <zerob13@gmail.com>

---------

Co-authored-by: hllshiro <40970081+hllshiro@users.noreply.github.com>
Co-authored-by: yyhhyyyyyy <yyhhyyyyyy8@gmail.com>
Co-authored-by: xiaomo <wegi866@gmail.com>
Co-authored-by: Simon He <57086651+Simon-He95@users.noreply.github.com>
Co-authored-by: luy <12696648@qq.com>
@zerob13 zerob13 deleted the feat/qwen3-thinking-budget branch January 6, 2026 12:17
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

3 participants