Practical-Decoder
Implemented a decoder-only GPT model from scratch with attention and MoE architecture experimentation.
Selected implementation-focused ML/GenAI projects.
Implemented a decoder-only GPT model from scratch with attention and MoE architecture experimentation.
Built DiT implementation with end-to-end training, sampling, and checkpoint evaluation workflows.
Implemented post-training methods (GRPO/PPO/DPO + LoRA) on Gemma 2B with practical evaluation loops.