Skill UI

Enables Claude to configure neural network architectures through the neural-network-builder plugin, creating or modifying models and defining layers, activation functions, and training parameters whenever network design or tuning is requested.

A comprehensive guide for selecting and optimizing embedding models for advanced vector search and retrieval-augmented generation (RAG) applications. This guide covers model comparison, chunking strategies, dimension reduction techniques, and provides runnable Python templates for both OpenAI and local Sentence Transformer implementations.

Fine-tune LLMs with TRL’s RLHF toolkit, covering SFT, DPO, PPO/GRPO and reward model training so you can align HuggingFace Transformers with human preferences or feedback in post-training workflows.

GPTQ enables post-training 4-bit quantization for large LLMs, delivering up to 4× memory reduction and 3-4× faster inference with under 2% perplexity loss on consumer GPUs through AutoGPTQ, PEFT, and Transformer integrations.

Extends transformer context windows using RoPE, YaRN, ALiBi, and position interpolation so long documents (32k-128k+ tokens) can be processed, pre-trained models lengthened, and efficient positional encodings or extrapolation strategies deployed.

Mamba is a selective state-space model architecture delivering O(n) inference for million-token sequences, offering hardware-aware Mamba-1 and Mamba-2 variants that beat Transformers on speed without KV cache overhead.

nanoGPT is a simplified, minimalist implementation designed specifically for educational purposes. It reproduces the core architecture of GPT-2 (124M) using clean, hackable code, allowing users to understand the entire transformer pipeline from scratch. The framework supports full workflows, including data preparation on various datasets (e.g., Shakespeare, OpenWebText), model training, and text generation, making it ideal for students and researchers learning NLP and deep learning principles.

Flash Attention optimizer delivers 2-4x speedup and 10-20x memory savings for transformer attention, ideal when training or running long-context models that hit GPU memory or latency limits; supports PyTorch native scaled_dot_product_attention, the flash-attn library, H100 FP8 precision, and sliding-window attention.

Outlines guarantees valid JSON/XML/code outputs by applying CFG/FSM-constrained token filtering, Pydantic schemas, and grammar-level sampling while supporting local Transformers, llama.cpp, and vLLM backends for fast inference.

Automated assistance for pyspark transformer work within data pipelines, activating when you request transformer patterns or best practices and covering ETL, data transformation, orchestration, and streaming scenarios.

Auto-activating API Integration skill that guides response transformer work, covering common patterns, third-party APIs, webhooks, SDK generation, and delivers best-practice code plus validation advice.

RWKV is an RNN+Transformer hybrid architecture with linear-time inference, infinite context, and GPT-style training. It delivers constant-memory, streaming generation on CUDA and CPU, ideal for long-context, memory-constrained deployments.