9 results
Tuning Engines CLI, MCP server, and Python agent runtime adapters for governed model, agent, skill, and MCP workflows. Fine-tune open-source LLMs, run inference, manage datasets/evaluations, and connect LangGraph or Temporal while Tuning Engines handles policy, audit, usage, and token economics.
An eye that grows: accumulated design taste for DSH agents. A taste corpus + codified principles that sharpen with every example you feed — review designs against learned judgment, not a fixed ruler.
驯兽师（驯兽场）：把人类语言翻译给智能体的 DSH 插件——注入「开智」内核（第一性原理 + 需求剖析 + 极简沟通 + 最少 token），左上角悬浮萌宠，设置页可配内核与开关。
基础训练
Math_Agent: a math training/coaching agent preset for DeepSeek Harness, with a programmatic zero-leak output guard that never reveals answers
把技能当可训练参数：像训练神经网络一样训练 agent 技能（epochs/batchsize/learning rate/验证门禁，但不碰模型权重）——rollout→reflect→aggregate→select→update→evaluate 循环、候选编辑仅在严格改善 held-out 验证分时接受、文本学习率预算、零推理时模型调用、部署紧凑 best_skill.md。受 microsoft/SkillOpt（MIT）启发。
自主研究 agent：给 AI agent 一个小而真实的 LLM 训练设置，让它过夜自主实验——改代码→固定 5 分钟训练→检查 val_bpb 是否改善→保留或丢弃→重复（约 12 实验/小时、一晚约 100 个）；单文件修改、program.md 超轻量 skill。受 karpathy/autoresearch（94k★ MIT）启发。
ICPC competitive programming training tools for DeepSeek Harness — sync, analysis, plans, reviews, templates
Bioinformatics skill router plugin for DeepSeek Harness: searchable index over a large skill library, an anysearch web-search provider, remote-exec tooling, and scenario guidance for literature review, bio-data analysis, and remote training