Data analysis, machine learning, LLM apps, and prompt design (9K skills)
Use SAELens to train and analyze sparse autoencoders, which separate neural network activity into features that may represent recognizable concepts. Explore pretrained features, evaluate reconstruction quality, and investigate how individual features influence language model outputs.
Guides experiments that inspect and change internal model values using nnsight. Covers local PyTorch models and remote experiments through NDIF, including replacing intermediate values to test how particular layers affect a model’s predictions.
Evaluate LLM application quality using automated scores, human reviews, and another language model as a judge. Compare models or prompts, establish baseline results, and detect quality drops through repeatable tests and statistical comparisons.
Set up LLM serving with structured responses, including JSON and outputs restricted to specified patterns. Covers reusing shared prompt calculations, tool calling, and multi-turn conversations through SGLang, with examples for server setup and application integration.
Investigate why a PyTorch model behaves a certain way by changing its internal values with pyvene. The skill guides experiments that swap or remove activations, train interventions, compare outputs, and save or share intervention configurations.
Integrate the Gemini API on Agent Platform using the Google Gen AI SDK. Guidance covers authentication, text and multimedia tasks, structured JSON responses, live voice interactions, context caching, and batch prediction across five programming languages.
Run language models locally with llama.cpp on CPUs, Apple Silicon, or AMD and Intel GPUs. The skill guides model setup, memory-saving GGUF format choices, interactive chat, and serving models through an OpenAI-compatible API.
Fine-tune large language models with LoRA and QLoRA while limiting GPU memory use. Guides training settings and adapter management, using small sets of task-specific weights that can be saved, switched, or merged into a base model.
Get guidance on adapting LLMs with Unsloth, including implementation, API usage, and debugging. Covers LoRA and QLoRA, methods for adjusting models with fewer trainable parameters, with a focus on training speed and memory use.
Compress large language models to 4-bit weights to reduce GPU memory use and accelerate inference. Covers loading compressed models, configuring quantization, and using Transformers or vLLM, with guidance on hardware requirements and backend choices.
Get guidance on distributed training with PyTorch FSDP, which splits model parameters across devices. The skill covers implementation, mixed precision, moving data to CPU memory, communication setup, and debugging distributed training code.
Adapt PyTorch training scripts for multiple GPUs or machines with Accelerate. This skill guides code changes, configuration, and launch commands, including mixed precision, which combines numerical formats, and DeepSpeed or FSDP integration for distributed training.
Adapt machine learning training to run across multiple GPUs or machines with Ray Train. Covers changes to existing code, comparing training settings with Ray Tune, and saving checkpoints so training can resume after failures.
Build and train Mixture of Experts models, which activate selected specialist networks for each input. Use DeepSpeed or HuggingFace workflows to configure routing, balance workloads, distribute experts across GPUs, and optimize inference.
Design and implement systems that collect, transform, validate, and store data for analysis. Covers scheduled and real-time processing, data warehouses, quality checks, monitoring, and cost tuning using tools such as Apache Spark, dbt, and Airflow.
Build and operate AutoGPT agents that carry out multi-step workflows. Covers a visual editor for connecting processing blocks, scheduled and webhook triggers, deployment, execution monitoring, and benchmarks for evaluating agent performance.
Design collaborative AI teams with CrewAI by defining each agent’s role, goals, tasks, and expected results. Covers passing work between agents, manager-led coordination, memory settings, and execution planning for workflows such as research and writing.
Design autonomous AI agents that break goals into tasks, use tools, and review their own work. Covers reasoning and action loops, separate planning and execution, and reliability measures such as cost limits, validation, and logging.
Helps configure and troubleshoot DeepSpeed for training large models across multiple devices. Covers memory optimization, parallel training, and DeepNVMe storage transfers, with examples for tuning data movement and configuring models that divide work among specialized components.
Design AI agents that plan tasks, call tools, and retain useful information. Covers coordination between agents, evaluation, and debugging, with practical patterns for execution limits, error handling, and tracing what an agent does.
Organize PyTorch training code with LightningModule and Trainer. Follow examples for validation, logging, model saving, and distributed training across multiple GPUs, with callbacks that add monitoring or stop training when progress stalls.
Prepare training datasets for large language models with GPU-assisted cleaning. Covers duplicate removal, quality filtering, and personal information redaction, with workflows for text, images, video, and audio using NeMo Curator.
Build data processing pipelines for machine learning, from preprocessing to batch predictions. Uses Ray Data to handle datasets larger than memory, distribute work across CPUs and GPUs, and feed data into training frameworks.
Evaluate code generation models with HumanEval, MBPP, and other benchmarks. Test generated programs, compare models across programming languages, and record pass@k scores, which estimate the chance that at least one of k solutions passes the tests.