feat: :sparkles: add Zero3Layer class for efficient parameter management in distributed training

feat: add LoRA modules with linear and embedding layers for efficient parameterization

feat: add FeedbackAttention module with support for key-value precomputation and positional embeddings

feat: add GLU variants implementation with Tiny Shakespeare dataset and training framework

feat: :zap: add GPT model implementation with custom optimizer and training configurations

feat: add CompressiveTransformer and related classes for enhanced memory compression in transformer models

feat: add BERTChunkEmbeddings and RetroIndex for enhanced text processing and embedding retrieval

feat: :sparkles: implement Attention with Linear Biases (ALiBi) for input length extrapolation

feat: add Rotary Position Embedding and RotaryPEMultiHeadAttention classes

feat: add Relative Multi-Headed Attention implementation with shift functionality

refactor: :art: refactor the file structe

feat: enhance TransformerXL with improved forward method and layer normalization

docs: add reference link for Transformer XL attention span explanation

partial update transformerxl

Start learning transformer xl