Create and manage dstack presets for agent-driven model inference optimization.
npx skills add https://github.com/dstackai/dstack --skill dstack-presets
Please help me install this Agent Skill. Skill: dstack-presets Source: https://github.com/dstackai/dstack/tree/main/skills/dstack-presets Command: npx skills add https://github.com/dstackai/dstack --skill dstack-presets
Finding an optimized inference configuration for a model requires manual benchmarking across hardware and serving settings. This Skill guides the creation and management of dstack presets, which capture the serving configuration, benchmark results, and verified hardware in a portable format.
dstack preset CLI commands.Ask the assistant to create a dstack preset to optimize inference for your model on your target hardware.
High-intent search queries and answers about installing and using this skill.
Create a preset by defining the serving configuration and letting agents iterate on it to reach an optimized benchmark. The preset records the final configuration, the benchmark achieved, and the exact hardware it was verified on.
Use the dstack preset CLI commands to watch active optimization sessions, list existing presets, export them for deployment, and delete presets you no longer need.
Yes, presets use a portable format that deploys the verified serving configuration to any cloud, Kubernetes cluster, or bare-metal fleet supported by dstack backends.
Use a preset only when you want to create or manage an inference optimization configuration. For deploying or serving a model directly, use a dstack service instead, as presets are not meant for serving.
A preset stores the serving configuration that produced the optimization result, the benchmark it reached, and the exact hardware the result was verified on, making it reproducible across environments.
dstackaiProvision and orchestrate GPU workloads across clouds using YAML configurations.
peterschmidt85Orchestrate GPU provisioning across cloud providers, Kubernetes, and on-premises clusters.
agentic-inSelect inference paths, serving stacks, and benchmarking plans for ML models.
QingHuan-6Benchmark SGLang, vLLM, and TensorRT-LLM serving to produce Markdown/CSV summaries.