AI models
Hybrid H3-2.7B
Training compute
6.5×10²¹ FLOP
Parameters
2.7B
Published
Dec 28, 2022
Hybrid H3-2.7B is an AI model developed by Stanford University and University at Buffalo (United States), first published in December 2022. It works in the language domain, on tasks such as language modeling/generation and question answering.
Training it took an estimated 6.5×10²¹ FLOP of compute (estimation method: operation counting). The model has 2,700,000,000 parameters. It was trained on roughly 400B datapoints. Training ran on 8 NVIDIA A100 SXM4 80 GB.
Access: Open weights (unrestricted). Its weights are openly available. The reference paper has 636 citations. Epoch AI rates the confidence of this record as likely.
Full record
- Organization
- Stanford University, University at Buffalo
- Country of organization
- United States
- Domain
- Language
- Task
- Language modeling/generation, Question answering
- Training compute
- 6.5×10²¹ FLOP
- Compute estimation method
- Operation counting
- Parameters
- 2,700,000,000
- Dataset size
- 400B
- Training hardware
- NVIDIA A100 SXM4 80 GB
- Chips used
- 8
- Training power draw
- 6.4 kW
- Numerical format
- BF16
- Model accessibility
- Open weights (unrestricted)
- Open weights
- Yes
- Citations
- 636
- Epoch confidence
- Likely
More from Stanford University,University at Buffalo
SourceEpoch AI, 'AI Models'. Published online at epoch.ai. Retrieved 2026-07-29 from https://epoch.ai/data/ai-models. Licensed under CC BY 4.0.
← All ai models