Krishu Kumar Thapa

Krishu Kumar Thapa

PhD candidate in Computer Science, Washington State University

I work on scalable ML, from distributed training and efficient attention to reinforcement learning for data centers and uncertainty-aware climate forecasting. Right now I'm focused on long-sequence efficiency for LLMs.

Previously at HPE Labs and Argonne National Laboratory, and 2.5 years as a software engineer. Looking for ML research scientist and applied ML engineer roles starting 2027.

Selected work

Efficient attention for long sequences

Thesis work in progress: sparse attention that replaces dense self-attention in Qwen2.5 LLMs to cut the quadratic cost of long contexts.

Distributed training at scale

3× throughput scaling ViT pretraining to 64 GPUs, and 20% higher GPU utilization with hybrid full and low-rank training. Argonne National Laboratory.

Reinforcement learning for data center scheduling

23.8% lower job wait time than the strongest baseline, with the lowest SLA violation rate at 5.4%. HPE Labs.

Uncertainty-aware forecasting

Spatiotemporal attention models that forecast snow-water equivalent with calibrated uncertainty. Two first-author AAAI papers.

Publications

ForeSWE: Forecasting Snow-Water Equivalent with an Uncertainty-Aware Attention Model K. K. Thapa, S. Savalkar, B. Singh, T. N. Hoang, K. Rajagopalan, et al. AAAI 2026
PreLoRA: Hybrid Pre-training of Vision Transformers with Full Training and Low-Rank Adapters K. K. Thapa, R. Barik, K. T. Chitty-Venkata, M. Emani, V. Vishwanath. AI on HPC Workshop, ISC High Performance
HPT: Hybrid Acceleration of Spatio-Temporal Attention Model Training on Heterogeneous Manycore Architectures S. Dahal, P. Dhingra, K. K. Thapa, P. P. Pande, A. Kalyanaraman. IEEE TPDS 2025
Attention-Based Models for Snow-Water Equivalent Prediction K. K. Thapa, B. Singh, S. Savalkar, A. Fern, K. Rajagopalan, A. Kalyanaraman. AAAI 2024

Experience

ML Research Intern, HPE Labs2026
ML Research Intern, Argonne National Laboratory2025
Research Assistant, Washington State University2022 to now
Software Engineer, Leapfrog Technology2019 to 2022

Education

PhD, Computer Science, Washington State University2027
BE, Computer Engineering, Kathmandu University2019

Skills

Python, C++, PyTorch, CUDA, multi-GPU distributed training, LLMs and Transformers, reinforcement learning, GNNs, Gaussian processes.

Recognition