Results for “training”

GPU-accelerated data curation for LLM training. Supports text/image/video/audio. Features fuzzy deduplication (16× faster), quality filtering (30+ heuristics), semantic deduplication, PII redaction, NSFW detection. Scales across GPUs with RAPIDS. Use for preparing high-quality training datasets, cleaning web data, or deduplicating large corpora.

Trains large language models (2B-462B parameters) using NVIDIA Megatron-Core with advanced parallelism strategies. Use when training models >1B parameters, need maximum GPU efficiency (47% MFU on H100), or require tensor/pipeline/sequence/context/expert parallelism. Production-ready framework used for Nemotron, LLaMA, DeepSeek.

Launch distributed Megatron-LM training jobs on a SLURM cluster with a minimal sbatch skeleton, environment-variable setup for torch.distributed.run, CUDA_DEVICE_MAX_CONNECTIONS rules, container conventions, monitoring, and per-rank failure diagnosis.

Adds PyTorch FSDP2 (fully_shard) to training scripts with correct init, sharding, mixed precision/offload config, and distributed checkpointing. Use when models exceed single-GPU memory or when you need DTensor-based sharding with DeviceMesh.

Conducts Data Protection Impact Assessments for AI and ML systems per EDPB Guidelines 04/2025 on AI processing. Covers training data lawfulness evaluation, model risk assessment, automated decision triggers, and AI-specific DPIA methodology. Keywords: AI DPIA, machine learning impact assessment, EDPB AI guidelines, model risk, training data.

Explain and apply core ML/AI concepts — model types, training pipelines, evaluation metrics, and neural architectures.

Expert guidance for fast fine-tuning with Unsloth - 2-5x faster training, 50-80% less memory, LoRA/QLoRA optimization

Expert guidance for fast fine-tuning with Unsloth - 2-5x faster training, 50-80% less memory, LoRA/QLoRA optimization

Expert guidance for fast fine-tuning with Unsloth - 2-5x faster training, 50-80% less memory, LoRA/QLoRA optimization

Expert guidance for fast fine-tuning with Unsloth - 2-5x faster training, 50-80% less memory, LoRA/QLoRA optimization

Expert guidance for fast fine-tuning with Unsloth - 2-5x faster training, 50-80% less memory, LoRA/QLoRA optimization

Expert guidance for fast fine-tuning with Unsloth - 2-5x faster training, 50-80% less memory, LoRA/QLoRA optimization

Expert guidance for distributed training with DeepSpeed - ZeRO optimization stages, pipeline parallelism, FP16/BF16/FP8, 1-bit Adam, sparse attention

Expert guidance for distributed training with DeepSpeed - ZeRO optimization stages, pipeline parallelism, FP16/BF16/FP8, 1-bit Adam, sparse attention

Train Mixture of Experts (MoE) models using DeepSpeed or HuggingFace. Use when training large-scale models with limited compute (5× cost reduction vs dense models), implementing sparse architectures like Mixtral 8x7B or DeepSeek-V3, or scaling model capacity without proportional compute increase. Covers MoE architectures, routing mechanisms, load balancing, expert parallelism, and inference optimization.

Train Mixture of Experts (MoE) models using DeepSpeed or HuggingFace. Use when training large-scale models with limited compute (5× cost reduction vs dense models), implementing sparse architectures like Mixtral 8x7B or DeepSeek-V3, or scaling model capacity without proportional compute increase. Covers MoE architectures, routing mechanisms, load balancing, expert parallelism, and inference optimization.

Conducts structured post-breach remediation using a lessons learned framework covering root cause remediation, control gap closure, policy updates, training modifications, monitoring enhancements, and regulatory follow-up. Provides a systematic approach to preventing breach recurrence and demonstrating accountability to supervisory authorities. Keywords: post-breach, remediation, lessons learned, root cause, control gap, policy update, training.

Machine learning development agent for end-to-end ML workflows: data preprocessing, model training, evaluation, hyperparameter tuning, and deployment

This skill should be used when users want to train or fine-tune language models using TRL (Transformer Reinforcement Learning) on Hugging Face Jobs infrastructure. Covers SFT, DPO, GRPO and reward modeling training methods, plus GGUF conversion for local deployment. Includes guidance on the TRL Jobs package, UV scripts with PEP 723 format, dataset preparation and validation, hardware selection, cost estimation, Trackio monitoring, Hub authentication, and model persistence. Should be invoked for tasks involving cloud GPU training, GGUF conversion, or when users mention training on Hugging Face Jobs without local GPU setup.

This skill should be used when working with reinforcement learning tasks including high-performance RL training, custom environment development, vectorized parallel simulation, multi-agent systems, or integration with existing RL environments (Gymnasium, PettingZoo, Atari, Procgen, etc.). Use this skill for implementing PPO training, creating PufferEnv environments, optimizing RL performance, or developing policies with CNNs/LSTMs.

Visualize training metrics, debug models with histograms, compare experiments, visualize model graphs, and profile performance with TensorBoard - Google's ML visualization toolkit

Visualize training metrics, debug models with histograms, compare experiments, visualize model graphs, and profile performance with TensorBoard - Google's ML visualization toolkit

PyTorch deep learning patterns and best practices for building robust, efficient, and reproducible training pipelines, model architectures, and data loading.

Use when implementing risk-adjusted rewards, discussing Sharpe ratio in RL training, or tuning reward components for risk awareness

Submit one or more wandb-logged training/finetuning runs to the HPC scheduler. `WANDB_PROJECT` is fixed per repo (snake_case basename); `WANDB_RUN_GROUP` is picked per invocation. The training script must take the experiment/group name as a config key (e.g. Hydra `meta.experiment_name=<group>`); the skill passes it on the command line. The working tree is committed first so each run pins to a real SHA. Delegates SLURM/PBS templating to `cluster-instructions`. Use when the user asks to submit, queue, launch, or kick off a wandb training/finetuning job.

Evaluates the correlation between a zero-cost NAS metric (epsilon) and actual training accuracy across different neural architecture search spaces, testing the metric's ability to rank architectures without training. It probes whether output dispersion from constant weight initializations can serve as a reliable.

Use to measure how long new support agents take to reach proficiency and where their ramp stalls, so training and nesting can be targeted. Trigger for "how long until new hires are productive", "time to proficiency", "how is the new cohort doing", nesting or ramp design, comparing training cohorts, or a new hire being assessed before they have had time to ramp.

Navigate the PhysicsNeMo repository by discovering model families, datapipes, and examples through live file search, without writing training code.

Automate BigML machine learning operations through Composio's toolkit via Rube MCP, including model creation, training, and deployment.

Maintains brand consistency, guidelines, and visual identity across all touchpoints, including audits, training, and compliance monitoring.

Architecture specialist for model and training systems. Use when the user asks to talk to Chamber, requests the ML architect, or needs model architecture decisions.

Track ML experiments with automatic logging, visualize training in real-time, optimize hyperparameters with sweeps, and manage model registry with W&B - collaborative MLOps platform

Track ML experiments with automatic logging, visualize training in real-time, optimize hyperparameters with sweeps, and manage model registry with W&B - collaborative MLOps platform

Track ML experiments with automatic logging, visualize training in real-time, optimize hyperparameters with sweeps, and manage model registry with W&B - collaborative MLOps platform

Manage NVIDIA Brev GPU instances for TAO training, evaluation, and inference using the Brev CLI and Docker.

Production machine-learning engineering workflow for data contracts, reproducible training, model evaluation, deployment, monitoring, and rollback. Use when building, reviewing, or hardening ML systems beyond one-off notebooks.