Provision GPU-enabled Modal sandboxes with pre-loaded models for interpretability research.
npx skills add https://github.com/ajobi-uhc/seer-claude-plugin --skill seer
Please help me install this Agent Skill. Skill: seer Source: https://github.com/ajobi-uhc/seer-claude-plugin/tree/main/seer-plugin/skills/seer Command: npx skills add https://github.com/ajobi-uhc/seer-claude-plugin --skill seer
This skill streamlines provisioning of GPU sandbox environments for interpretability research, enabling researchers to quickly spin up pre-loaded notebooks and experiments without manual configuration.
Install or enable the seer skill in your workflow, configure a SandboxConfig with a supported GPU (e.g., A100), include desired models, start the sandbox, create a notebook session, and attach to it to begin experiments.
High-intent search queries and answers about installing and using this skill.
You can provision a GPU sandbox for interpretability research by configuring SandboxConfig with a supported GPU like A100, including desired models, and starting the sandbox to run experiments.
Yes, the workflow supports provisioning GPU sandboxes across diverse GPUs including H100 and A100, allowing you to create and attach to notebook sessions for interpretability experiments.
The best way to preload models into a remote sandbox is by defining a ModelConfig within your setup script, which automates the remote provisioning and resource loading before you attach to the notebook session.
Yes, you need a script-based workflow to create and connect to notebook sessions, utilizing Modal-based tooling like ExecutionMode to manage the remote provisioning and attachment workflows.
Automated GPU sandbox creation uses Modal tooling with SandboxConfig, ModelConfig, and ExecutionMode to provision remote environments, preload models, and establish notebook sessions without manual configuration.
KaiserWhoLearnsRun Python code on cloud GPUs via Modal serverless platform.
skkeoriwRemove refusal behaviors from large language models using mechanistic interpretability techniques.
OpenHandsRead, modify, execute, and convert Jupyter notebooks to HTML, Python, or Markdown.
jacardlDeploy GPU-accelerated ML workloads on demand with Modal.
