dsh-plugin-fusion
Run the following command in DeepSeek Harness:
dsh plugin install alexpadholol/dsh-plugin-fusion
Paste the following prompt into your AI chat to install this plugin:
Run dsh plugin install alexpadholol/dsh-plugin-fusion in your terminal to install it; the source is at https://github.com/alexpadholol/dsh-plugin-fusion, then restart dsh web to make the fusion_generate tool available.
About this plugin
A single model call is at the mercy of sampling temperature and stochasticity; slightly shifting the temperature can yield wildly different answers, leaving users stuck with whatever one draw produces.
fusion-plugin-fusion registers a fusion_generate tool that works like drafting multiple versions and picking the best one: it fires several calls across a temperature range (0.2 to 1.8, default 5, capped at 10) to collect diverse candidates, then runs a judge pass that reviews every candidate and selects the most suitable one. The response includes all candidates alongside the chosen answer and the judge reasoning, keeping the selection transparent and auditable.
It is well suited for developers who need higher-quality single-shot inference and want a generate-then-select workflow, as well as for review or education settings where an explainable selection rationale matters. The plugin is zero-dependency and registers via raw JSON Schema, so once added to dsh it is available in every session immediately.
Use Cases
- Single-shot inference quality is inconsistent and requires best-of-N selection
- Auditable selection rationale and reasoning are needed
- Comparing model outputs at different temperatures to pick the best
Best For
- Developers sensitive to single-shot inference quality
- Teams requiring a generate-then-select workflow
- Review and education scenarios demanding explainable selections
Related Plugins
ModLens is a vision plugin for DeepSeek Harness that gives text-only models sight by reading images pasted directly into chat, with zero-config setup and multiple vision engines.
On-demand vision for text-only DeepSeek Harness agents: built-in free keyless vision chain and 14 vision tools, routing image turns as tool calls to vision models with pixel fidelity, no Python needed, one-command install.
Give text-only models in DeepSeek Harness eyes, enabling image Q&A, long-screenshot OCR, UI restoration, and GUI visual tasks.
An armor-breaking plugin for DeepSeek: rewrites instruction priority, expands refusal-blocked phrases, persists across sessions, and shows a green active indicator.