developing-flax-models
A comprehensive guide for developing, training, and managing neural networks using Flax NNX. Use when defining models, managing state, or writing training loops.
Find the perfect capability for your agent.
A comprehensive guide for developing, training, and managing neural networks using Flax NNX. Use when defining models, managing state, or writing training loops.
Use when creating an R modeling package that needs standardized preprocessing for formula, data frame, matrix, and recipe interfaces. Covers: mold() for training data preprocessing, forge() for prediction data validation, blueprints, model constructors, spruce functions for output formatting.
深度学习训练体系与实验编排专家。当用户询问“训练”“验证”“测试”“启动命令” “数据集划分”等问题时使用。要求模型结构已被明确定义。
Fix phantom MaxDD values in training by calibrating reward_scale. Trigger when: (1) validation MaxDD shows 35-80% values, (2) MaxDD doesn't correlate with training quality, (3) gating thresholds seem too lenient/strict.
Complete context for TAR UMT Data Science FYP implementing CrossViT for COVID-19 chest X-ray classification. Use when working on Jupyter notebooks, code implementation, data analysis, model training, or any task related to Tan Ming Kai's final year project. This skill provides dataset specs, model architecture details, hardware constraints (NVIDIA RTX 6000 Ada Generation (51GB VRAM) VRAM), preprocessing parameters, baseline models, evaluation metrics, hypotheses, and coding guidelines for reproducible research following TAR UMT academic requirements.
Training models on extended context lengths using optimized RoPE scaling and memory-efficient attention kernels. Triggers: long context, max_seq_length, rope scaling, large context window, flex attention.
Post-training model validation workflow: gating, backtesting, walk-forward validation, deployment decisions. Trigger after GPU training completes.
Run model validation on a dataset. Use when testing model performance, comparing checkpoints, or running final evaluation.
See the main Model Explainability skill for comprehensive XAI coverage.
Эксперт active learning. Используй для ML с участием человека, uncertainty sampling, annotation workflows и labeling optimization.
Follow these patterns when extending the OptAIC Python SDK with new domain operations. Use for adding client methods for datasets, signals, portfolios, backtests, and other resources. Covers async/sync interfaces, uploads, and long-running operations.
GPU-aware PPO training configuration for A100/H100. Trigger when training is slow or GPU utilization is low.
Recognize/Describe clearML architecture and extension points
Creates database migrations, implements RLS policies, and manages data transformations
Create database migrations following Laravel best practices with reversible up/down methods, focused single-purpose changes, proper index management, and clear naming conventions. Use this skill when creating new database tables, adding or modifying columns in existing tables, creating or dropping indexes, implementing foreign key constraints, renaming tables or columns, or managing database schema changes. This skill applies when working on migration files in database/migrations/ directory, generating new migrations via artisan commands, or ensuring zero-downtime deployments with backwards-compatible schema changes.
Comprehensive Supabase database management skill for creating migrations, managing RLS policies, optimizing performance, and maintaining database security. Use when creating/modifying database schema, auditing security, fixing RLS issues, adding indexes, or performing any Supabase database operations. Automatically applies project conventions and best practices.
Эксперт DB replication. Используй для настройки репликации MySQL, PostgreSQL, MongoDB, failover и high availability.
SQL query optimization, indexing strategies, and EXPLAIN analysis for database performance. WHEN: MUST use for slow queries, EXPLAIN analysis, index design, N+1 problems. Invoke with "/sql-optimization-patterns" or "optimize this query", "analyze EXPLAIN", "fix slow query". WHEN NOT: ORM-specific patterns (use framework skills), application-level caching, database migrations.