Alejandro Armas
AI Systems Performance Engineer @ HPE | Ex-(Astranis, MIT, UC Davis, NSF) intern
- Role
- Machine Learning Performance Engineer at Hewlett Packard Enterprise
- Location
- San Francisco, CA, US
- LinkedIn followers
- 500 followers
About Alejandro Armas
I am a recent graduate at the University of California, Davis interested in building ML platforms (anything related to training, deploying, monitoring ML models or collecting/transforming big-data).Please check out my Blog!@ blog.alejandroarmas.dev
Experience
Machine Learning Performance Engineer
Nov 2024 — Present
Benchmarked and improved inference throughput by 30.7% for Llama70B and Mixtral8x7B models for efficient execution in single-node GPU environments; by evaluating quantization methods, tuned batch and kv-cache sizes, and configuring tensor/pipeline parallelism strategies• Built a custom Byte-Pair Encoding (BPE) tokenizer, profiled code and made algorithmic improvements to merge and pre-tokenization portions of the algorithm, delivering 6x, 150x, and 7.5x latency speedups, respectively.• Achieved #1 top result to MLPerf® Inference v5.1 benchmark in Offline Datacenter category for Whisper model (34,450.7 samples per second) for Cray XD670 system (features 8x H200 SXM GPUs)• Profiled and optimized multi-node Llama 3.1 8B distributed training throughput on GB200 NVL72 systems, tuning parallelism configurations (context/tensor/pipeline) within NeMo Megatron + Transformer Engine, achieving measurable efficiency gains
Education
Los Medanos College
Electrical Engineering
2017 — 2019
Diablo Valley College
Associate's degree, Computer Science
2018 — 2019
University of California, Davis
Bachelor's degree, Computer Science and Electrical Engineering
2019
Find verified contacts for anyone on LinkedIn
Unifers gives sales teams verified emails and direct dials, enriched profiles, and outreach that lands in the inbox.
Free plan included · No credit card required
This profile is compiled from publicly available professional sources. Unifers is not affiliated with or endorsed by LinkedIn. Request removal of this profile.