Data Science & ML Agent Skills

Cortex-M Backend for ExecuTorch

Optimizes and deploys PyTorch models to Arm Cortex-M processors using the CMSIS-NN backend.

ComfyUI Image Generation

Connects and controls ComfyUI via API to generate and edit high-quality images with token-saving efficiency.

ComfyUI OpenClaw Integration

Connects Claude directly to ComfyUI via API to automate professional image generation and editing workflows with high token efficiency.

ComfyUI Integration for Claude

Integrates ComfyUI with Claude for token-efficient image generation and automated editing workflows.

LLaVA Multimodal Vision Assistant

Integrates LLaVA to enable sophisticated visual instruction following and multi-turn conversational image understanding.

nnsight Remote Interpretability

Interprets and manipulates neural network internals across local and remote models using the nnsight library and NDIF execution.

LangChain Framework Integration

Builds sophisticated LLM applications using agents, chains, and Retrieval-Augmented Generation (RAG) with a unified interface.

Weights & Biases MLOps

Integrates Weights & Biases into your workflow to track machine learning experiments, visualize training metrics, and manage model artifacts in real-time.

OpenRLHF Model Training

Deploys and manages high-performance RLHF training pipelines for large-scale language models using Ray and vLLM acceleration.

BigCode Evaluation Harness

Evaluates AI code generation models using industry-standard benchmarks and pass@k metrics.

Speculative Decoding & Inference Optimization

Accelerates LLM inference speed by up to 3.6x using speculative decoding, Medusa heads, and lookahead techniques without sacrificing model quality.

PyTorch Lightning Training

Standardizes and accelerates PyTorch model training with built-in support for distributed computing, logging, and engineering best practices.

Knowledge Distillation for LLMs

Compresses large language models into efficient student models while retaining performance through advanced teacher-student transfer techniques.

AWQ LLM Quantization & Optimization

Optimizes Large Language Models using 4-bit activation-aware weight quantization to achieve 3x faster inference with minimal accuracy loss.

Outlines Structured Generation

Ensures guaranteed valid JSON, XML, and type-safe code generation from LLMs using constrained token sampling and Pydantic models.

Guidance: Structured LLM Control

Enforces structured outputs and grammatical constraints on Large Language Models to guarantee valid JSON, code, and regex-compliant text.

TorchTitan LLM Pretraining

Accelerates large-scale LLM pretraining using PyTorch-native 4D parallelism and Float8 optimization.

Flash Attention Optimization

Optimizes Transformer attention mechanisms using Flash Attention for significant speedups and 10-20x memory reduction.

Segment Anything Model (SAM)

Implements zero-shot image segmentation using Meta AI's SAM to identify and extract objects via points, boxes, or automatic mask generation.

LLaMA-Factory Fine-Tuning

Streamlines the fine-tuning of 100+ large language models using LLaMA-Factory with support for QLoRA and multimodal architectures.

LitGPT LLM Implementation & Training

Implements, fine-tunes, and deploys high-performance Large Language Models using Lightning AI's LitGPT framework.

LLM Quantization with bitsandbytes

Quantizes Large Language Models to 8-bit or 4-bit formats to reduce memory usage by up to 75% with minimal accuracy loss.

Enterprise RL Training (miles)

Optimizes large-scale Reinforcement Learning training for Mixture-of-Experts models using high-performance quantization and speculative rollout techniques.

Llama.cpp Inference Serving

Deploys high-performance LLM inference on CPU, Apple Silicon, and non-NVIDIA GPUs using GGUF quantization.

Model Pruning & LLM Compression

Reduces Large Language Model size and accelerates inference using advanced pruning techniques like Wanda and SparseGPT.

verl LLM Reinforcement Learning

Trains large language models using advanced reinforcement learning algorithms like GRPO and PPO with the production-ready verl framework.

Unsloth Fine-Tuning Expert

Optimizes LLM fine-tuning workflows by providing expert guidance on Unsloth implementation for faster training and reduced memory usage.

NanoGPT Model Architecture

Implements a minimalist, educational GPT-2 architecture in PyTorch for learning and training transformer models from scratch.

GPTQ Model Quantization

Compresses large language models to 4-bit precision to enable high-speed inference and deployment on consumer-grade hardware.

HuggingFace Tokenizers

Provides high-performance, Rust-based tokenization tools for building and training NLP models with support for BPE, WordPiece, and Unigram algorithms.