libre-mlops

Data labeling: annotation guidelines, platforms, inter-annotator agreement, active learning, weak supervision, and label error detection.

1 skill · 1 command · 1 agent

ML data pipelines: Apache Beam, Spark, dbt, orchestration, streaming features, validation, and backfills.

1 skill · 1 command · 1 agent

Data versioning: DVC, Delta Lake, lineage, dataset cards, and reproducible dataset splits.

1 skill · 1 command · 1 agent

Distributed training: DDP, FSDP, DeepSpeed ZeRO, model parallelism, mixed precision, and multi-node debugging.

1 skill · 1 command · 1 agent

Experiment tracking: MLflow, Weights & Biases, Neptune, ClearML, hyperparameter search, and run organization.

1 skill · 1 command · 1 agent

Feature engineering: feature stores, transformations, selection, time-series features, leakage checks, and drift.

1 skill · 1 command · 1 agent

GPU optimization: profiling, memory management, torch.compile, Flash Attention, Triton kernels, and quantization.

1 skill · 1 command · 1 agent

LLM fine-tuning: LoRA and QLoRA, instruction datasets, DPO and PPO alignment, adapter merging, and evaluation.

1 skill · 1 command · 1 agent

MLflow: tracking server, model signatures, Model Registry, projects, custom flavors, and serving.

1 skill · 1 command · 1 agent

ML testing: data quality tests, model unit tests, behavioral tests, and performance regression gates.

1 skill · 1 command · 1 agent

Model deployment: ONNX export, BentoML, TorchServe, Triton, FastAPI serving, dynamic batching, and canary rollouts.

1 skill · 1 command · 1 agent

Model evaluation: metrics with confidence intervals, calibration, sliced and fairness evaluation, and LLM output quality.

1 skill · 1 command · 1 agent

Model monitoring: data and concept drift, performance estimation without labels, and alerting without alert fatigue.

1 skill · 1 command · 1 agent

Model registry: versioning, staging gates, champion and challenger promotion, model cards, and CI/CD webhooks.

1 skill · 1 command · 1 agent

Model training: PyTorch Lightning and Hugging Face Trainer loops, LR schedules, gradient health, checkpoints, and early stopping.

1 skill · 1 command · 1 agent

Prompt engineering: prompting techniques, system prompts, structured output, injection defense, DSPy, and prompt versioning.

1 skill · 1 command · 1 agent

PyTorch: custom datasets and DataLoaders, nn.Module design, model surgery, custom autograd, and profiling.

1 skill · 1 command · 1 agent

Retrieval-augmented generation: chunking, embeddings, vector stores, hybrid retrieval, reranking, grounded generation, and evaluation.

1 skill · 1 command · 1 agent

TensorFlow and Keras: model building, tf.data pipelines, custom layers, SavedModel export, TFLite quantization, and TFX.

1 skill · 1 command · 1 agent

Vector databases: index selection and tuning, pgvector, Pinecone, Qdrant, multi-tenancy, and reindexing after embedding changes.

1 skill · 1 command · 1 agent

Optional LibreMLOps hooks: an ML project summary at session start, a confirmation prompt before Claude touches secrets files or runs commands that destroy data, models, or run history, and post-edit checks for empty writes and matching tests.

hooks