Guides scalable AI system design using hardware-software co-design and model distillation.
npx skills add https://github.com/K-Dense-AI/mimeographs --skill jeff-dean
Please help me install this Agent Skill. Skill: jeff-dean Source: https://github.com/K-Dense-AI/mimeographs/tree/main/mimeographs/jeff-dean Command: npx skills add https://github.com/K-Dense-AI/mimeographs --skill jeff-dean
This skill provides a principled, engineering-focused frame inspired by Jeff Dean to help teams design and optimize large-scale AI systems, balancing frontier capabilities with practical latency, energy, and deployment constraints.
Initiate by identifying the top bottlenecks in your current ML system using Back-of-the-Envelope thinking, then draft a 5-10x scaling plan that preserves latency and energy budgets.
High-intent search queries and answers about installing and using this skill.
Design scalable AI systems by applying hardware-software co-design, model distillation, and sparse activation to maintain latency and energy budgets while planning for 5-10x scaling horizons.
Transition from specialized to unified multimodal models when siloed compute becomes unsustainable, using massively multi-task modeling to reduce infrastructure overhead and streamline deployment.
Plan ML scaling using Back-of-the-Envelope thinking to identify bottlenecks, drafting a 5-10x scaling plan that preserves latency and energy budgets to avoid 100x premature scaling.
Hardware-ML co-design improves deployment by aligning model architectures with accelerator constraints, enabling latency-first design and energy-aware decisions for cost-effective scaling.
Use model distillation to reduce latency in production AI systems by transferring knowledge from large models to smaller ones, enabling scalable deployment while preserving frontier capabilities.
Multi-task models suffer high latency during scaling when architectures ignore accelerator constraints; applying sparse activation and hardware-algorithm co-design enforces latency-conscious decisions.
OzeroHAXSelect system-level performance patterns for latency, throughput, or cost constraints.
informatico-madridCompress large language models into smaller student models using knowledge distillation.
Open330Benchmark ML models by measuring latency, throughput, memory, and accuracy.
nickm538Benchmark and optimize claude-flow v3 for Flash Attention, HNSW search, and memory usage.
