starefinger/dsh-llm-qwen-local
LLM adapter for a locally served Qwen model behind vLLM's OpenAI-compatible endpoint, with per-request reasoning effort.
Adapts a locally deployed Qwen model — the README's example is a Qwen3.8-27B behind vLLM — into DSH as a selectable provider over the OpenAI-compatible /v1/chat/completions endpoint. Configuration happens on a settings page that also exposes reasoning levels per request and can discover models from the endpoint, and the adapter maps DSH's request shape onto the local server's dialect.
Install
dsh plugin --profile web add dsh-llm-qwen-localThe npm line, which is the prebuilt install with no build step. Compatibility is exact rather than a range: v0.4.1 targets dsh 0.1.2-rc.1, and the README states that 0.1.1-rc.2 and older make the web app fail to boot. The repository also publishes immutable per-version tags for pinning a harness pair (for example the git+https form ending #dsh-0.1.2-rc.1-plugin-0.4.0), and a fix for an existing tag ships as a new tag rather than by moving it.
Compatibility
Supported from dsh 0.1.2-rc.1 upward; older 0.1.1-rc.2 installs are not supported and the web app fails to boot. Since 0.4.1 the published plugin has no runtime dependency on any @deepseek-ai package. Saving settings applies live, because the adapter re-resolves per request, and the model list may be left empty, which keeps the route mounted but dormant until it is repopulated. Community-maintained, and explicitly not a DeepSeek or Qwen product.
Details
- Repo: starefinger/dsh-llm-qwen-local
- Category: Models & Providers
- Stars: 4
- Version: 0.4.1 (npm), exact target dsh 0.1.2-rc.1
Alternatives
slaren/dsh-llm-llamacpp · tianxia--/dsh-llm-local-token · shengbinxu/dsh-llm-latency