HyperConv is a TensorFlow layer that combines local convolution, global hypernetwork-based modulation, and associative memory, designed for efficient long-context representation learning without self-attention.
Primary use cases:
- Embedding models
- Retrieval and similarity search
- Long-context sequence modeling
HyperConv is the mainstay of all openlem series since openlem2 of OpenLab-NLP.
First released: 2025-12-15 (UTC)
