Joel Hestness
Scholar

Joel Hestness

Google Scholar ID: wkbvCf0AAAAJ
Distinguished Research Scientist, Cerebras Systems
Deep LearningLanguage UnderstandingHeterogeneous SystemsHigh-performance Computing
Google Scholar↗
Citations & Impact
All-time
Citations
5,836
 
H-index
19
 
i10-index
27
 
Publications
20
 
Co-authors
3
list available
Contact
No contact links provided.
Publications
8 items
Don't Drop Dropout: Optimizing Layer Sparsity for Efficient LLM Training and Inference
2026
Cited
0
Unlocking Lossless Speedups in LLMs via Discrete Diffusion
2026
Cited
0
GQA-{\mu}P: The maximal parameterization update for grouped query attention
2026
Cited
0
Predicting Training Re-evaluation Curves Enables Effective Data Curriculums for LLMs
2025
Cited
0
Scaling with Collapse: Efficient and Predictable Training of LLM Families
2025
Cited
0
Power Lines: Scaling Laws for Weight Decay and Batch Size in LLM Pre-training
2025
Cited
0
Don't be lazy: CompleteP enables compute-efficient deep transformers
2025
Cited
0
Straight to Zero: Why Linearly Decaying the Learning Rate to Zero Works Best for LLMs
2025
Cited
1
Resume (English only)
Co-authors
3 total
Co-author 1
Co-author 1
Newsha Ardalani
Newsha Ardalani
Research Scientist, Meta AI Research (FAIR)
Steve Keckler
Steve Keckler
Vice President of Architecture Research, NVIDIA