README.md

torch-models

Learning by implementing papers

Basics

Generative Models

• Variational Autoencoder, 2013: With ELBO
• Masked Autoregressive Flow, 2018

Transformer Models

• GPT2, 2019
• ColBERT, 2020
• Coconut, sample code
• Combine Whisper with Qwen 7b

Reinforcement Learning

• Reinforce
• On-policy DPO, Tulu 3, code
• Reinforcement on Verifiable Reward (RLVR), Tulu 3, code
• free process rewards without process labels

ICL

• ICLR: In-context Learning of Representations

Training

C/CUDA

• llm.c
• llm.cu

Distributed Computing

• picotron

---

Project Ideas

• Turn an LLM into a Speech to Speech Model
• Test Time Training