Valerois AI Logo
ValeroisAI
Bifrost CSL
Scientific Disclosures & Mathematics

Bifrost CSL Architecture Specification

Comprehensive mathematical derivations, continuous receptive field formulations, and zero KV-cache memory proofs behind the Valkir model family.

Technical Paper • Valerois Whitepaper v1.4

Bifrost CSL: Sub-quadratic Linear Foundation Models

Abstract

Transformer self-attention incurs $O(N^2)$ time and $O(N)$ KV-cache spatial complexity during inference, creating massive hardware bottlenecks for long context and local edge deployment. We introduce Bifrost CSL (Convolutional Sub-quadratic Linear), which synthesizes hierarchical grouped dilated convolutions with input-dependent SiLU gating and SwiGLU feedforward networks. Evaluated on 4.74 billion pretraining tokens on a single consumer AMD Radeon RX 9070 XT GPU, our 116.8M reference model (Valkir 16L) matches or exceeds Pythia-160M and OPT-125M across OpenAI HumanEval (54.0% AST pass), WinoGrande (51.62%), and ARC-Easy (33.08%) while maintaining a strictly immutable $O(1)$ 8.8 MB state buffer.

1. Dual-Scale Temporal ConvolutionsSyntax & Context
Parallel grouped kernels capturing local code syntax and long-range narrative continuity.

Eliminates quadratic attention complexity by utilizing concurrent short and dilated temporal convolutions, establishing an expansive 1,233-token receptive field across 16 layers.

2. Dynamic Token GatingO(1) Attention Replacement
Content-aware gating that selectively focuses on critical tokens in linear time.

Dynamically modulates contextual information flow based on token importance without storing past key-value vectors in VRAM, guaranteeing sub-2ms time-to-first-token.

3. Adaptive LayerScale StabilizationDeep Numerical Stability
Zero gradient explosion and rock-solid mixed-precision training up to 44 layers.

Calibrated scaling factors applied to residual connections completely eliminate NaN loss anomalies during marathon pretraining runs on consumer AMD hardware.

4. Fixed Circular State Buffer8.86 MB Constant VRAM
Strictly constant memory usage that never scales with input prompt length.

While traditional Transformers consume gigabytes of VRAM at 16K or 32K context, Bifrost CSL maintains an immutable 8.86 MB buffer, enabling infinite context streaming on edge devices.

Empirical Pretraining Curriculum (4.74 Billion Cumulative Tokens)

Phase 1: Base Pretrain
10.5 hours • 1.11B tokens
General text + Python AST seeds
Phase 2: Continual Alignment
9.24 hours • 1.25B tokens
60% Code + 40% FineWeb-Edu
Phase 3: 18h Marathon
17.75 hours • 2.38B tokens
100% pure unseen FineWeb-Edu + GSM8K

Cite Bifrost CSL

@article{valerois2026bifrost,
  title={Bifrost CSL: Convolutional Sub-quadratic Linear Foundation Models with Zero KV-Cache},
  author={Valerois AI Research Team},
  journal={Valerois Technical Report},
  volume={1},
  number={1},
  year={2026},
  url={https://valerois.com/research}
}