Reduce torch.compile cold starts with a content-addressable cache.
npx skills add https://github.com/gar-ai/mallorn --skill mithril-cache-agent
Please help me install this Agent Skill. Skill: mithril-cache-agent Source: https://github.com/gar-ai/mallorn/tree/main/.claude/skills/mithril-cache-agent Command: npx skills add https://github.com/gar-ai/mallorn --skill mithril-cache-agent
Build mithril-cache to reduce torch.compile cold starts by providing a content-addressable storage layer and cache-management primitives.
Configure and run the mithril-cache-agent to initialize the content-addressable cache for torch.compile.
High-intent search queries and answers about installing and using this skill.
Reduce torch.compile cold starts by integrating a content-addressable storage layer that caches artifacts. This approach intercepts cache directory requests to reuse intermediate results, avoiding recompilation delays.
Content-addressable storage for PyTorch caching is a mechanism that stores compiled artifacts using hashes as keys. It ensures deterministic retrieval of intermediate results, preventing redundant compilations across runs.
Set up cache eviction for torch.compile artifacts by applying a least recently used policy. This approach automatically removes outdated cache entries, managing storage limits while keeping frequently accessed data available.
Yes, you can redirect Torch cache directories using environment hooks. This integration intercepts standard framework requests and reroutes them to a custom content-addressable storage backend.
No, the mithril-cache-agent operates with minimal dependencies. It provides an MVP strategy using Rust and Python components without requiring external libraries, ensuring a clear upgrade path for future integration.
cipherstashCreate isolated git worktrees with safety checks and project setup.
zardusai-cyberCache file processing results using SHA-256 content hashes as keys.
Xza85hrfCache expensive inputs and model results using SHA-256 keyed entries.
irizqiImplement cache patterns, atomic locks, and tag-based invalidation in Laravel applications.
