8 skills found
Optimizes Claude Code memory files in 4 interactive steps: removes duplicates, migrates rules to CLAUDE.md/rules files, compresses remaining entries, validates with cleanup. Typical reduction: 30-50% on token count.
Validate and use selective and full activation recompute in Megatron Bridge to reduce GPU memory usage at the cost of extra compute.
Validate and use CPU offloading in Megatron Bridge, including layer-level activation offloading and fractional optimizer state offloading with HybridDeviceOptimizer.
cProfile、メモリプロファイラー、パフォーマンスベストプラクティスを使用してPythonコードをプロファイルおよび最適化。低速なPythonコードのデバッグ、ボトルネックの最適化、アプリケーションパフォーマンスの改善時に使用。
Expert skill for AI model quantization and optimization. Covers 4-bit/8-bit quantization, GGUF conversion, memory optimization, and quality-performance tradeoffs for deploying LLMs in resource-constrained JARVIS environments.
Python performance optimization patterns using profiling, algorithmic improvements, and acceleration techniques. Use when optimizing slow Python code, reducing memory usage, or improving application throughput and latency.
Post-training 4-bit quantization for LLMs with minimal accuracy loss. Use for deploying large models (70B, 405B) on consumer GPUs, when you need 4× memory reduction with <2% perplexity degradation, or for faster inference (3-4× speedup) vs FP16. Integrates with transformers and PEFT for QLoRA fine-tuning.
Activation-aware weight quantization for 4-bit LLM compression with 3x speedup and minimal accuracy loss. Use when deploying large models (7B-70B) on limited GPU memory, when you need faster inference than GPTQ with better accuracy preservation, or for instruction-tuned and multimodal models. MLSys 2024 Best Paper Award winner.
