| license: apache-2.0 | |
| 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) |