dsh-qwen-reasoning
Manifest validAuto-enhances manually-declared Qwen models with the full reasoning-effort levels (off/minimal/low/medium/high/xhigh/max) and defaults to max. Its HTTP routes have no DSH access check, so a web page o
dsh-qwen-reasoning
A DeepSeek Harness (DSH) plugin that gives your Qwen models the full set of reasoning-effort levels — off / minimal / low / medium / high / xhigh / max — and defaults the provider to max, so the DSH reasoning bar exposes every level and reasoning_effort is actually sent to the Qwen gateway.
为你的 qwen 模型启用完整推理档位(off/minimal/low/medium/high/xhigh/max),并把提供商默认设为 max,让 DSH 推理条显示全部档位、并真正向千问网关发送 reasoning_effort。
Why this exists / 为什么需要它
Qwen models served through an OpenAI-compatible gateway (e.g. the Qwen AI Platform Token Plan endpoint https://token-plan.maas.qianwenaiapi.com/compatible-mode/v1) think by default, but a hand-declared model in DSH's llm-pi-ai provider has no reasoningEfforts map, so:
- the reasoning bar shows no selectable levels, and
- DSH never sends
reasoning_effort, leaving you stuck at the gateway default.
This plugin fixes that declaratively. The Qwen gateway accepts reasoning_effort ∈ {none, off, minimal, low, medium, high, xhigh, max} (verified live: an invalid value returns 400 listing exactly these). max is the highest; none turns thinking off.
通过 OpenAI 兼容网关接入的 qwen 模型默认开启思考,但在 DSH llm-pi-ai 里手工声明的模型没有 reasoningEfforts 映射,导致推理条无档位可选、DSH 也不发送 reasoning_effort。本插件以声明方式补上。千问网关接受 reasoning_effort ∈ {none, off, minimal, low, medium, high, xhigh, max}(已实测:非法值返回 400 并列出这八项);max 最高,none 关闭思考。
offmaps to wire valuenone(not "omit"): because Qwen thinks by default, omitting the field would keep thinking on.off → noneis what actually disables it.
off映射到none(而非"省略"):qwen 默认开思考,省略字段会保持思考,off → none才能真正关闭。
What it does / 它做什么
On install (and whenever your llm-pi-ai config changes), the plugin:
- reads your live providers via
settings.describe(); - for every provider with a matching model (default: model id contains
qwen), adds the seven-levelreasoningEffortsmap to models that lack one; - sets the provider's default
reasoningtomaxandcompat.supportsReasoningEffort: truewhen missing; - writes the changes back with
settings.mutate()(theprovidersfield is volatile), persisting to your profile patch.
安装后(以及你的 llm-pi-ai 配置变化时),插件读取实时提供商,为含匹配模型(默认模型 id 含 qwen)的提供商补上七档 reasoningEfforts,在缺省时设默认 reasoning: max 与 compat.supportsReasoningEffort: true,再经 settings.mutate() 写回你的 profile 补丁。
Additive + idempotent / 加法且幂等 — it never overwrites an existing reasoningEfforts map (including an explicit false for non-reasoning models), an existing provider reasoning, or existing compat choices. A second run writes nothing. Non-matching models and other providers are left untouched.
绝不覆盖已有的 reasoningEfforts(含显式 false 的非推理模型)、已有的 provider reasoning 或 compat。再次运行不写入任何内容;不匹配的模型与其他提供商原样保留。
Why a runtime plugin, not a static config patch / 为什么是运行时插件而非静态配置补丁
A DSH bundle patch replaces the target plugin's config wholesale (never deep-merged). A static patch carrying llm-pi-ai providers would wipe your own providers. So this plugin does a surgical runtime read‑modify‑write of just the reasoning fields, preserving everything else.
DSH bundle 补丁对目标插件配置是整段覆盖(从不深合并)。携带 llm-pi-ai 提供商的静态补丁会冲掉你自己的提供商。因此本插件在运行时只针对推理字段做精确的读‑改‑写,其余配置全部保留。
Install / 安装
# DSH plugin CLI (recommended)
dsh plugin --profile web add dsh-qwen-reasoning
Or install by spec in Settings → Plugins → Install:
- npm name:
dsh-qwen-reasoning - git:
git:jasocheung/dsh-qwen-reasoning - local path: the cloned directory
或在 设置 → 插件 → 安装 按规格安装:npm 名 dsh-qwen-reasoning、git git:jasocheung/dsh-qwen-reasoning、或本地克隆目录。
After installing, reload the DSH window (Ctrl+R) so the browser bundle mounts. Linked local installs are outside DSH's HMR scope, so source edits require a host restart to take effect.
安装后刷新 DSH 窗口(Ctrl+R)以挂载浏览器 bundle。链接式本地安装不在 DSH HMR 范围内,改源码需重启 host 才生效。
Configuration / 配置
Open Settings → Plugins → dsh-qwen-reasoning (the configuration card):
- Matched models — each matching
provider/modelwith a badge (levels set/no levels) and its default effort. - Apply now — force the enhancement immediately.
- Auto-apply — re-enhance whenever
llm-pi-aiconfig changes (default on). - Model match — case-insensitive regex on the model id (default
qwen). - Provider match — case-insensitive regex on the provider id (default empty = all providers).
- Default reasoning effort — one of the seven levels (default
max).
打开 设置 → 插件 → dsh-qwen-reasoning 配置卡:匹配模型列表(已有档位/无档位徽章 + 默认档位)、立即应用、自动应用、模型匹配正则、提供商匹配正则、默认推理档位。
The wire-level efforts map is also configurable via the plugin config (efforts), defaulting to:
{ "off": "none", "minimal": "minimal", "low": "low", "medium": "medium", "high": "high", "xhigh": "xhigh", "max": "max" }
Chat command / 聊天命令
/qwen-reasoning
Applies the enhancement and prints the matched models, their level state, and how many settings were written.
应用增强并打印匹配模型、档位状态与本次写入项数。
Routes / 路由(同源)
| Method | Path | Purpose |
|---|---|---|
| GET | /dsh-qwen-reasoning/status | matched models + level state (read-only) |
| POST | /dsh-qwen-reasoning/apply | enhance now (force) |
| GET/POST | /dsh-qwen-reasoning/config | plugin configuration |
| GET | /dsh-qwen-reasoning/client.js | self-hosted browser bundle |
Compatibility / 兼容性
- DSH
0.1.7-rc.2(desktop), Node>=18, web + desktop client. - Pure ESM, prebuilt
lib/— no build step. - Requires the
llm-pi-aiprovider plugin (ships with DSH) and a Qwen model declared under it. engines.dsh: ">=0.1.0-rc.1 <0.3.0-0"(prerelease-aware range).
Development / 开发
node --test # 12 unit tests for the additive/idempotent op builder
node --check lib/index.js
node --check lib/client.js
The op builder (lib/reasoning.js) is pure and dependency-free; the host (lib/index.js) wires it to the DSH settings service. lib/client.js is the hand-authored browser bundle (module-loader factory format, theme tokens only).
Marketplace listing / 市场收录
- Official discoverability: the GitHub topic
dsh-plugin(the only official plugin discovery channel; DSH has no central registry). - awesome-dsh-plugin: listed via a PR adding
data/plugins/jasocheung__dsh-qwen-reasoning.yml. - dsh-plugin-hub / shop / feed: index npm packages carrying the
dsh-pluginkeyword +dshbundle metadata.
License / 许可
Comments
Loading…
Similar plugins
by AkotaP
DSH web plugin: a Model Capabilities page in Settings — manage per-model Context Window, Max Output, Input Modalities and Reasoning Levels (Default = drop the field), plus global preset lists.
★ 0
MIT
JavaScript
Nov 10, 2026
dsh plugin --profile web add dsh-model-tweaksby lovezi0
Models+ settings page for the DSH web UI: per-model reasoning-effort tables, input modalities and compat flags saved through the official settings pipeline, with models.dev metadata prefill and API ke
★ 7
↓ 491/wk
MIT
TypeScript
Oct 17, 2026
dsh plugin --profile web add dsh-model-extensionby imkingjh999
DSH plugin: auto reasoning_effort (low/high/max) per turn via MiniMax complexity scorer + token ledger + per-reply metadata label
★ 0
MIT
TypeScript
Aug 26, 2026
dsh plugin --profile web add dsh-adaptive-effortby TikaFlow
Auto-fills reasoning-efforts, context windows, output limits and image modality for models from unofficial (custom) providers, with data from models.dev. Also provides setting cards to control plugin'
★ 25
MIT
TypeScript
Nov 17, 2026
dsh plugin --profile web add dsh-model-fixby DC1024
DSH plugin: sync model-catalog reasoning levels (reasoningEfforts) from what the provider advertises on /v1/models — sync_reasoning_levels tool plus a one-click button in the Models settings page.
★ 0
MIT
JavaScript
Oct 17, 2026
dsh plugin --profile web add dsh-effort-syncby Triple3h
DSH Web plugin: per-model reasoning-effort configuration and selection — settings page + slider embedded in the model picker. DSH Web 插件:逐模型推理力度配置与选择(设置页 + 模型选择器内嵌滑块)。
★ 0
MIT
JavaScript
Aug 23, 2026
dsh plugin --profile web add dsh-reasoning-effort