Skill UI

Guides engineers through DeepSpeed distributed training and NVMe I/O optimizations, covering ZeRO, pipeline parallelism, mixed precision, and DeepNVMe handles for synchronous and asynchronous tensor storage.

HuggingFace Accelerate adds distributed support to any PyTorch script with just four lines, providing a unified API for DDP, DeepSpeed, FSDP, and Megatron, plus automatic device placement, mixed precision, and simple launch/config commands for fast prototyping.

Guides production-grade miles RL training for massive MoE models, covering FP8/INT4 low precision, train-inference alignment, and speculative rollout workflows to maximize throughput in enterprise deployments.

Flash Attention optimizer delivers 2-4x speedup and 10-20x memory savings for transformer attention, ideal when training or running long-context models that hit GPU memory or latency limits; supports PyTorch native scaled_dot_product_attention, the flash-attn library, H100 FP8 precision, and sliding-window attention.

Guides adding PyTorch FSDP2’s fully_shard integration to training loops, covering correct distributed init, bottom-up sharding, mixed precision/offload policies, DTensor-aware optimizers, and distributed checkpointing so oversized models or DeviceMesh sharding goals work reliably.

This utility ensures the prevention of silent floating-point mismatch bugs across all EVM chains. It covers runtime querying of decimals, chain-aware caching, handling precision drift in bridged tokens, and safe normalization for bots, dashboards, and DeFi tools. It is crucial for accurate cross-chain balance comparison.

A comprehensive tool for updating Feishu cloud documents. It supports seven distinct update modes including appending, full overwrite, targeted range replacement, and multi-instance replacement. The system allows highly precise positioning using content ranges (ellipsis) or chapter titles, ensuring content updates are localized and safe, thus preserving images, comments, and formatting.

This skill operates on two layers: the THINK layer enforces rigorous coding discipline (simplicity, surgical changes, goal-driven execution), and the SPEAK layer compresses prose output (caveman/classical modes) for maximum token efficiency. Use it when brevity is explicitly requested or when disciplined, high-precision code generation is required.

This utility ensures the prevention of silent floating-point mismatch bugs across all EVM chains. It covers runtime querying of decimals, chain-aware caching, handling precision drift in bridged tokens, and safe normalization for bots, dashboards, and DeFi tools. It is crucial for accurate cross-chain balance comparison.

This comprehensive guide provides detailed instructions for revising academic papers, specifically in the field of learning theory, for the COLT conference. It teaches authors how to structure the argument rigorously, ensuring the model is unambiguous, mastering the informal-to-formal theorem pairing, and maintaining mathematical precision (quantifier and constant hygiene).

This guide provides a crucial late-stage polish for economic manuscripts targeting rigorous, broad-audience journals like the International Economic Review (IER). It focuses on elevating prose, structuring the introduction (four-move approach), and ensuring formal precision while maintaining cross-field legibility. Use this when your content is strong, but the narrative structure or academic writing voice needs sharpening to successfully communicate complex results to diverse referees.

This comprehensive guide equips authors with a structured framework for writing the Related Work section for top-tier NLP conferences (e.g., NAACL). It emphasizes tracking methodological, data, and task lineages, ensuring citation precision across Findings, workshops, and preprints, and culminating in a single, falsifiable "delta sentence" to precisely position the novel contribution.