Data Science & ML Claude 技能

Llama.cpp Inference

384

Deploys and optimizes LLM inference on CPU, Apple Silicon, and consumer hardware using GGUF quantization.

Flash Attention Optimization

384

Optimizes Transformer models using Flash Attention to achieve significant speedups and memory reductions during training and inference.

PEFT Fine-Tuning

384

Fine-tunes large language models using LoRA, QLoRA, and other parameter-efficient methods to drastically reduce memory and compute requirements.

Long Context Transformer Extension

384

Extends transformer context windows using RoPE, YaRN, and ALiBi techniques to process documents exceeding 128k tokens.

LLaMA Factory Fine-Tuning

384

Streamlines the fine-tuning process for over 100 large language models using the LLaMA-Factory framework and QLoRA techniques.

DSPy Declarative AI Programming

384

Builds complex AI systems using Stanford's declarative programming framework to optimize prompts and create modular RAG systems automatically.

Whisper Speech Recognition

384

Transcribes audio, translates speech to English, and automates multilingual audio processing using OpenAI's Whisper models.

LangChain Framework

384

Builds LLM-powered applications using agents, retrieval-augmented generation (RAG), and modular chains.

CLIP Vision-Language Model

384

Enables zero-shot image classification and semantic image search by connecting visual concepts with natural language.

OpenRLHF Training Framework

384

Deploys high-performance Reinforcement Learning from Human Feedback (RLHF) workflows using Ray and vLLM acceleration for large-scale model alignment.

vLLM High-Performance Inference Serving

384

Serves Large Language Models with maximum throughput and efficiency using vLLM's PagedAttention and continuous batching.

GGUF Quantization & Model Optimization

384

Optimizes AI models for efficient local inference using the GGUF format and llama.cpp quantization techniques.

Speculative Decoding LLM Accelerator

384

Accelerates LLM inference speeds by up to 3.6x using advanced decoding techniques like Medusa heads and lookahead decoding.

BLIP-2 Multimodal Vision

384

Integrates Salesforce's BLIP-2 framework to enable advanced image captioning, visual question answering, and multimodal reasoning within AI workflows.

LLaVA Multimodal Assistant

384

Enables advanced vision-language capabilities for image understanding, multi-turn visual conversations, and document analysis.

PyTorch FSDP Expert

384

Optimizes large-scale AI model training using PyTorch Fully Sharded Data Parallelism for efficient memory management and scaling.

LlamaIndex RAG Framework

384

Connects LLMs to private data sources through advanced document ingestion, vector indexing, and retrieval-augmented generation (RAG) pipelines.

Model Merging & Fusion

384

Merges multiple fine-tuned AI models using mergekit to combine specialized capabilities like math and coding without expensive retraining.

NeMo Curator - GPU Data Curation

384

Curates high-quality datasets for LLM training using GPU-accelerated deduplication, filtering, and PII redaction.

nnsight Model Interpretability

384

Interprets and manipulates neural network internals for any PyTorch model, including massive foundation models via remote execution.

AudioCraft AI Generation

384

Generates high-fidelity music and sound effects from text descriptions using Meta's AudioCraft framework.

Outlines Structured Generation

384

Guarantees valid, type-safe JSON and structured outputs from Large Language Models using grammar-based constraints.

Ray Train Distributed Computing

384

Orchestrates distributed machine learning training across clusters to scale PyTorch, TensorFlow, and Hugging Face models.

GRPO RL Fine-Tuning

384

Implements Group Relative Policy Optimization (GRPO) using the TRL library to enhance model reasoning and structured output capabilities.

TensorBoard ML Visualization

384

Visualizes machine learning training metrics and model performance to streamline experiment tracking and model debugging.

Pinecone Vector Database Integration

384

Manages high-performance vector search and storage for production RAG and AI applications using Pinecone's serverless infrastructure.

LLM Guidance & Constrained Generation

384

Enforces structured LLM outputs using regex and grammars to guarantee valid JSON, XML, and code generation.

Constitutional AI Safety Alignment

384

Implements Anthropic's Constitutional AI method to train harmless, helpful models through self-critique and automated AI feedback.

RWKV Model Architecture

384

Implements and optimizes RWKV architectures, a hybrid RNN-Transformer model offering linear-time inference and infinite context windows.

SAELens: Mechanistic Interpretability

384

Decomposes complex neural network activations into sparse, interpretable features to understand and steer model behavior.