Benchmark Dataloader
A benchmarking setup for multimodal dataloaders, built to surface throughput bottlenecks before they become training-time surprises.
San Francisco, CAMember of Technical Staff at Datology
Building multimodal systems where data decisions stay legible.
At Datology, I work across evaluation, curation, distributed training, and launch infrastructure. I try to keep the path from a data choice to a trustworthy comparison short.
Field note
The interesting cases are the ones where the benchmark and the training run finally line up with what the data pipeline is doing.I like research surfaces that stay rigorous without feeling bureaucratic.
Personal anchor
Haoli Yin
Research engineer working across multimodal data, evaluation, and launch systems.
Research tile
DatBench
Public work on VLM evaluation and benchmark design.
Read paperSystems work
Benchmark Dataloader
Open systems work for finding dataloader bottlenecks before they burn training time.
View repoCurrent work
Systems work on the critical path
Three operating lanes keep the loop moving. They cover benchmark design, data handling, and launch.
Benchmark and evaluation systems for VLM research
Built evaluation paths that keep model comparisons useful for curation decisions and model iteration.
Multimodal data curation and export pipelines
Built ingestion and export paths that make large multimodal corpora easier to train on and inspect.
Distributed training and launch infrastructure
Added vLLM eval support and hardened multi-node launch plus checkpoint behavior for faster experimental turnover.
Selected public work
Open projects that show the same systems taste at a smaller scale.
SpecReFlow
Reflection-aware video restoration research translated into a public implementation for medical imaging workflows.