LLM

Complete implementations of large language models including all sub-components.

Repository Structure

torch/
├── gpt/               # GPT-1 style implementation
└── llama/             # LLaMA-1/2 implementation

What's Implemented

GPT (torch/gpt/):

• Multi-head self-attention with causal masking
• Learned positional embeddings
• LayerNorm, feedforward blocks
• Training loop with loss estimation

LLaMA (torch/llama/):

• Multi-head attention with Rotary Position Embeddings (RoPE)
• RMSNorm (instead of LayerNorm)
• SwiGLU feedforward network
• Top-p sampling for generation
• SentencePiece tokenizer

Usage

GPT:

cd torch/gpt
python train.py

LLaMA:

cd torch/llama
python generate.py

Default Configurations

Parameter	GPT	LLaMA
Embedding dim	384	4096
Hidden dim	-	11008
Heads	6	32
Layers	6	32
Context length	256	2048
Dropout	0.2	0.0

References

• Attention Is All You Need - Vaswani et al., 2017
• Layer Normalization - Ba et al., 2016
• Root Mean Square Layer Normalization - Zhang & Sennrich, 2019
• RoFormer: Enhanced Transformer with Rotary Position Embedding - Su et al., 2021
• LLaMA: Open and Efficient Foundation Language Models - Touvron et al., 2023
• LLaMA 2: Open Foundation and Fine-Tuned Chat Models - Touvron et al., 2023