Vitaliy Chiley

United States

Join Prog.AI to see contacts

Summary

🤩

Rockstar

🎓

Top School

Vitaliy Chiley is an ML engineer with eight years of experience advancing deep learning research and large-scale model training, currently developing algorithms for Cerebras Systems' wafer-scale AI hardware. He holds BS and MS degrees in Electrical Engineering from UC San Diego with a focus on machine learning, controls, and DSP, and brings a practical systems-oriented perspective to model engineering. His open-source contributions to MosaicML projects show hands-on expertise in scalable training infrastructure—improving FSDP support, mixed precision, memory monitoring, and attention kernel integrations to boost efficiency and numerical fidelity. Vitaliy’s GitHub descriptor "Token Predictor" belies a deeper specialty in optimizing training pipelines and attention mechanisms across Torch, Flash, and Triton backends. Colleagues value his blend of low-level performance tuning and high-level model correctness, particularly when adapting cutting-edge libraries to new hardware and PyTorch releases. Based in the United States, he pairs strong academic foundations with production-driven engineering for large language model training at scale.

8 years of coding experience

University of California, San Diego

Associate of Science (AS) with honors Physics Math Natural Sciences, Associate of Science (AS) with honors Physics Math Natural Sciences at Sierra College

English, Ukrainian, Russian

Stackoverflow

Stats

1reputation

0reached

0answers

0questions

Github Skills (15)

attention-mechanism10

transformer-models10

pytorch10

machine-learning10

triton10

distributed-training10

deep-learning10

python10

parallelization10

ml10

testing9

mlops9

nlp9

cuda9

gpt8

Programming languages (3)

C++Jupyter NotebookPython

Github contributions (5)

mosaicml/llm-foundry

Apr 2023 - Jul 2024

LLM training code for Databricks foundation models

Role in this project:

ML Engineer

Contributions:2 releases, 359 reviews, 114 PRs in 1 year 2 months

Contributions summary:Vitaliy's commits primarily focused on modifications related to the training and comparison of language models, including the integration of different attention mechanisms like Torch, Flash, and Triton. They implemented and tested model configurations for different model sizes. Code changes reveal work on ensuring the numerical equivalence between the Hugging Face GPT2 model and the MosaicML implementation. The user also implemented integration tests.

deep-learningllmneural-networksnlppytorch

mosaicml/composer

Jan 2023 - Jan 2023

Supercharge Your Model Training

Role in this project:

ML Engineer

Contributions:56 reviews, 1 commit, 15 PRs in 1 day

Contributions summary:Vitaliy contributed to the implementation and refinement of Fully Sharded Data Parallel (FSDP) features within the `composer` library. Their work involved adding functionalities such as `ignore_modules`, custom process group support, and mixed precision improvements to enhance training efficiency and flexibility. The user also addressed device naming conventions for specific GPUs and updated the FSDP implementation for compatibility with newer PyTorch versions. Additionally, they contributed to memory monitoring improvements, including enabling aggregate memory monitoring.

pytorchml-systemsdeep-learningneural-networksmachine-learning

Find and Hire Top DevelopersWe’ve analyzed the programming source code of over 60 million software developers on GitHub and scored them by 50,000 skills. Sign-up on Prog,AI to search for software developers.

Request Free Trial