ML Ops
Machine learning operations
scvi-tools
by anthropics
Deep learning for single-cell analysis using scvi-tools. This skill should be used when users need (1) data integration and batch correction with scVI/scANVI, (2) ATAC-seq analysis with PeakVI, (3) CITE-seq multi-modal analysis with totalVI, (4) multiome RNA+ATAC analysis with MultiVI, (5) spatial transcriptomics deconvolution with DestVI, (6) label transfer and reference mapping with scANVI/scArches, (7) RNA velocity with veloVI, or (8) any deep learning-based single-cell method. Triggers include mentions of scVI, scANVI, totalVI, PeakVI, MultiVI, DestVI, veloVI, sysVI, scArches, variational autoencoder, VAE, batch correction, data integration, multi-modal, CITE-seq, multiome, reference mapping, latent space.
jupyter-notebook
by openai
"Use when the user asks to create, scaffold, or edit Jupyter notebooks (.ipynb) for experiments, explorations, or tutorials; prefer the bundled templates and run the helper script new_notebook.py to generate a clean starting notebook."
azure-ai-contentunderstanding-py
by microsoft
Azure AI Content Understanding SDK for Python. Use for multimodal content extraction from documents, images, audio, and video. Triggers: "azure-ai-contentunderstanding", "ContentUnderstandingClient", "multimodal analysis", "document extraction", "video analysis", "audio transcription".
azure-ai-projects-py
by microsoft
Build AI applications using the Azure AI Projects Python SDK (azure-ai-projects). Use when working with Foundry project clients, creating versioned agents with PromptAgentDefinition, running evaluations, managing connections/deployments/datasets/indexes, or using OpenAI-compatible clients. This is the high-level Foundry SDK - for low-level agent operations, use azure-ai-agents-python skill.
azure-ai-ml-py
by microsoft
Azure Machine Learning SDK v2 for Python. Use for ML workspaces, jobs, models, datasets, compute, and pipelines. Triggers: "azure-ai-ml", "MLClient", "workspace", "model registry", "training jobs", "datasets".
implementing-llms-litgpt
by Orchestra-Research
Implements and trains LLMs using Lightning AI's LitGPT with 20+ pretrained architectures (Llama, Gemma, Phi, Qwen, Mistral). Use when need clean model implementations, educational understanding of architectures, or production fine-tuning with LoRA/QLoRA. Single-file implementations, no abstraction layers.
llama-factory
by Orchestra-Research
Expert guidance for fine-tuning LLMs with LLaMA-Factory - WebUI no-code, 100+ models, 2/3/4/5/6/8-bit QLoRA, multimodal support
pytorch-fsdp2
by Orchestra-Research
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.
peft-fine-tuning
by Orchestra-Research
Parameter-efficient fine-tuning for LLMs using LoRA, QLoRA, and 25+ methods. Use when fine-tuning large models (7B-70B) with limited GPU memory, when you need to train <1% of parameters with minimal accuracy loss, or for multi-adapter serving. HuggingFace's official library integrated with transformers ecosystem.
quantizing-models-bitsandbytes
by Orchestra-Research
Quantizes LLMs to 8-bit or 4-bit for 50-75% memory reduction with minimal accuracy loss. Use when GPU memory is limited, need to fit larger models, or want faster inference. Supports INT8, NF4, FP4 formats, QLoRA training, and 8-bit optimizers. Works with HuggingFace Transformers.
miles-rl-training
by Orchestra-Research
Provides guidance for enterprise-grade RL training using miles, a production-ready fork of slime. Use when training large MoE models with FP8/INT4, needing train-inference alignment, or requiring speculative RL for maximum throughput.
mamba-architecture
by Orchestra-Research
State-space model with O(n) complexity vs Transformers' O(n²). 5× faster inference, million-token sequences, no KV cache. Selective SSM with hardware-aware design. Mamba-1 (d_state=16) and Mamba-2 (d_state=128, multi-head). Models 130M-2.8B on HuggingFace.
deepspeed
by Orchestra-Research
Expert guidance for distributed training with DeepSpeed - ZeRO optimization stages, pipeline parallelism, FP16/BF16/FP8, 1-bit Adam, sparse attention
gptq
by Orchestra-Research
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.
verl-rl-training
by Orchestra-Research
Provides guidance for training LLMs with reinforcement learning using verl (Volcano Engine RL). Use when implementing RLHF, GRPO, PPO, or other RL algorithms for LLM post-training at scale with flexible infrastructure backends.
simpo-training
by Orchestra-Research
Simple Preference Optimization for LLM alignment. Reference-free alternative to DPO with better performance (+6.4 points on AlpacaEval 2.0). No reference model needed, more efficient than DPO. Use for preference alignment when want simpler, faster training than DPO/PPO.
modal-serverless-gpu
by Orchestra-Research
Serverless GPU cloud platform for running ML workloads. Use when you need on-demand GPU access without infrastructure management, deploying ML models as APIs, or running batch jobs with automatic scaling.
constitutional-ai
by Orchestra-Research
Anthropic's method for training harmless AI through self-improvement. Two-phase approach - supervised learning with self-critique/revision, then RLAIF (RL from AI Feedback). Use for safety alignment, reducing harmful outputs without human labels. Powers Claude's safety system.
slime-rl-training
by Orchestra-Research
Provides guidance for LLM post-training with RL using slime, a Megatron+SGLang framework. Use when training GLM models, implementing custom data generation workflows, or needing tight Megatron-LM integration for RL scaling.
openrlhf-training
by Orchestra-Research
High-performance RLHF framework with Ray+vLLM acceleration. Use for PPO, GRPO, RLOO, DPO training of large models (7B-70B+). Built on Ray, vLLM, ZeRO-3. 2× faster than DeepSpeedChat with distributed architecture and GPU resource sharing.
axolotl
by Orchestra-Research
Expert guidance for fine-tuning LLMs with Axolotl - YAML configs, 100+ models, LoRA/QLoRA, DPO/KTO/ORPO/GRPO, multimodal support
torchforge-rl-training
by Orchestra-Research
Provides guidance for PyTorch-native agentic RL using torchforge, Meta's library separating infra from algorithms. Use when you want clean RL abstractions, easy algorithm experimentation, or scalable training with Monarch and TorchTitan.
ray-train
by Orchestra-Research
Distributed training orchestration across clusters. Scales PyTorch/TensorFlow/HuggingFace from laptop to 1000s of nodes. Built-in hyperparameter tuning with Ray Tune, fault tolerance, elastic scaling. Use when training massive models across multiple machines or running distributed hyperparameter sweeps.
huggingface-accelerate
by Orchestra-Research
Simplest distributed training API. 4 lines to add distributed support to any PyTorch script. Unified API for DeepSpeed/FSDP/Megatron/DDP. Automatic device placement, mixed precision (FP16/BF16/FP8). Interactive config, single launch command. HuggingFace ecosystem standard.