Projects

Deepseek-v3 BF16 3-node training TensorBoard

Distributed training Deepseek-v3 in BF16 on 3 nodes

Multi-node training experiments for Deepseek-v3 in BF16 across three and four A100 Nodes (World size=24/32), covering Torch/DeepSpeed orchestration, Torchrun and Ray Train, NCCL networking, driver/firmware stability, and robust checkpointing.

DistributedBF16NCCL
Qwen-72B 2-step training (SFT/DPO) TensorBoard

Qwen-72B 2-step training

Two-stage fine-tuning of Qwen 2.5-72B for content generation: Supervised Fine-Tuning (SFT) with DeepSpeed ZeRO-3 CPU offload and LoRA, followed by Direct Preference Optimization (DPO) on preference data.

DeepSpeedLoRADPO
LLaMA-3.1 8B Distributed Fine-tuning TensorBoard

LLaMA-3.1 8B Distributed Fine-tuning

Distributed fine-tuning of LLaMA-3.1 8B on 16 A100 GPUs with DeepSpeed ZeRO-3 CPU offload and LoRA. Stable training curves and efficient memory usage across 16 GPUs.

DistributedDeepSpeedLoRA
Try Simple Ops

Try Simple Ops

A Website I built for my consulting company, Try Simple Ops, to showcase our custom workflow automation solutions for businesses of all sizes.

ReactJSRedux
Casas Ilimitadas Real Estate Platform

Casas Ilimitadas

A real estate listing and pricing service I built to help users find and evaluate properties. Built with ReactJS and Redux for efficient state management and seamless user experience.

ReactJSRedux
Analytica Inmobiliario Dashboard

Analytica Inmobiliario

A data visualization project I built to learn fullstack development with Elixir Phoenix. Features both REST and GraphQL backend APIs with a React frontend using Ant Design components.

Elixir PhoenixGraphQLReact
Fizyl Financial Data Platform

Fizyl Financial Data Platform

A financial data analysis and visualization platform I built with vanilla React. Provides comprehensive tools for analyzing financial data and creating interactive visualizations.

ReactData Visualization