Data Science & ML Claude 技能

Sentence Transformers

Generates state-of-the-art text and image embeddings for RAG, semantic search, and clustering tasks.

NeMo Curator - GPU Data Curation

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

SentencePiece Tokenizer

Implements language-independent subword tokenization using BPE and Unigram algorithms for advanced AI model development.

Llama.cpp Inference

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

Outlines Structured Generation

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

Whisper Speech Recognition

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

BLIP-2 Multimodal Vision

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

CrewAI Multi-Agent Orchestration

Orchestrates teams of autonomous AI agents to collaborate on complex tasks through role-based delegation and memory.

Ray Data Processing

Processes large-scale datasets for machine learning workloads using distributed streaming execution across CPU and GPU clusters.

Long Context Transformer Extension

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

Axolotl LLM Fine-Tuning

Streamlines the fine-tuning of large language models using Axolotl through expert YAML configuration guidance and advanced training techniques.

SAELens: Mechanistic Interpretability

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

LLM Model Pruning

Compresses Large Language Models using advanced techniques like Wanda and SparseGPT to reduce memory footprint and accelerate inference speeds.

HuggingFace Tokenizers

Provides high-performance, Rust-optimized text tokenization for NLP research and production-grade machine learning pipelines.

DeepSpeed Distributed Training

Optimizes large-scale AI model training using DeepSpeed's ZeRO, pipeline parallelism, and high-performance DeepNVMe I/O handling.

bitsandbytes LLM Quantization

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

Mamba Architecture & SSM Implementation

Implements and optimizes Mamba-based Selective State Space Models for high-efficiency sequence modeling and long-context AI research.

Weights & Biases MLOps

Tracks machine learning experiments and manages model lifecycles with real-time visualization and collaborative tools.

LLM Reinforcement Learning & Fine-Tuning

Aligns Large Language Models with human preferences using advanced reinforcement learning techniques including SFT, DPO, PPO, and GRPO.

Knowledge Distillation for LLMs

Compresses large language models using teacher-student learning techniques to reduce inference costs while maintaining high performance.

AWQ LLM Quantization

Optimizes Large Language Models using activation-aware 4-bit quantization to achieve 3x inference speedups and significant memory reduction with minimal accuracy loss.

BigCode Evaluation Harness

Evaluates AI code generation models across multiple programming languages and benchmarks using standardized pass@k metrics.

HuggingFace Accelerate Distributed Training

Simplifies PyTorch distributed training by providing a unified API for DDP, DeepSpeed, and FSDP with minimal code changes.

Flash Attention Optimization

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

AudioCraft AI Generation

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

LLaVA Multimodal Assistant

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

vLLM High-Performance Inference Serving

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

DSPy Declarative AI Programming

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

CLIP Vision-Language Model

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

GGUF Quantization & Model Optimization

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