Downloads · 30 days
0
gateddeltanet-srm-v3-omega
Lgr54HFi/gateddeltanet-srm-v3-omega
gateddeltanet-srm-v3-omega is a machine learning model from Lgr54HFi. Use it for the machine learning task on the model card, and read the license before you ship it in a product.
Access
Public
Updated Apr 23, 2026
Repo size
—
Likes
0
Public
Repo makeup
Click a slice to open those files.
.py133 KB · 98%
At a glance
- Access
- Public
- Created
- Apr 23, 2026
- Updated
- Apr 23, 2026
- SHA
- 00de6004
Model card
From the Hugging Face model README
GatedDeltaNet SRM v3 OMEGA
A single-file production implementation of the Gated Delta Rule with novel contributions:
- Infinite lossless context: bit-exact streaming equivalence to batch processing, verified with
torch.equal - O(1) VRAM: activation memory fixed regardless of sequence length via Symplectic Reversible Recurrence (SRR)
- No approximations: Magnus-2 exact flow, not first-order Euler
- Fused Triton kernels: forward + reverse-recurrence backward with zero stored tape
- Unified Pageable Memory Manager: transparent state migration across VRAM/RAM/NVMe
- Holographic State Compression: lossless tensor-product decomposition of recurrent state
- Topological Geodesic Scan: exact O(log T) parallel scan with no rank truncation
- Geodesic Reversible Block: fully invertible residual wrapper for O(1) layer activation memory
Self-test suite validates all five axiomatic invariants (A1-A5) and all ten novel contributions (N1-N10).
- Created
- Apr 23, 2026
- Updated
- Apr 23, 2026