Paper complete · Empirical validation of GTC proposal from Paper IV. Batched-Jacobi 97× speedup at B=10. Cache hit rate 90.4–91.5% across 3 model scales. 15.5× faster than disk-based RAG. Compressed record store at 5.96 KB/record.
Paper VIII · May 2026 · v1.0

GTC Runtime: Measured Cache Coverage and Batch Jacobi

Empirical validation of the Geodesic Trajectory Cache. 97× batched speedup. 15.5× faster than RAG.

By William Ken Ohara Stewart (NagusameCS) · Repository · TeX source

Abstract

Paper IV proposed the Geodesic Trajectory Cache (GTC) as a manifold-native alternative to retrieval-augmented generation. This paper provides the empirical validation. On three model scales (135M, 360M, 1.5B parameters), GTC achieves cache hit rates of 90.4–91.5% at a 25%-fraction coverage budget — scale-invariant within ±0.5% across a 33× parameter range. Batch Jacobi evaluation reaches 97× speedup at batch size B=10. GTC is 15.5× faster than disk-based RAG. Compressed trajectory records occupy 5.96 KB each (rank-5 exact reconstruction). End-to-end speculative decode with GTC drafts achieves 76.5 tok/s (1.53× baseline).

1. Method

GTC stores trajectory embeddings $t_i \in \mathbb{R}^k$ in a k-d tree with cosine-similarity indexing. Query resolution: (1) exact match via hash, (2) cosine-similarity threshold, (3) jury vote over top-N neighbors. Cache coverage is measured as the fraction of held-out queries whose nearest cached neighbor falls within similarity threshold $\varepsilon^*$.

2. Measured Results

2.1 Cache Coverage (3-model scale invariance)

ModelParametersCoverage at 25% budget
SmolLM2-135M135M91.0%
Qwen2.5-360M360M90.4%
Qwen2.5-1.5B1.5B91.5%

Coverage is scale-invariant within ±0.5% — the manifold structure is independent of model size.

2.2 Batch Jacobi Speedup

Batch Size BSpeedup vs. SerialError
214.6×float64 floor
544.5×float64 floor
1097.9×float64 floor
2060.0×float64 floor

Jacobi propagation error remains at float64 floor precision across all batch sizes. Peak at B=10; diminishing returns beyond due to memory bandwidth.

2.3 Compressed Record Store

MetricValue
Per-record size5.96 KB
Total cache (24 records)143 KB
Query latency30.9 μs
ReconstructionRank-5 exact

2.4 GTC vs. RAG

MethodQuery TimeSpeedup vs. RAG
Disk-based RAG (FAISS)475 μs1.0×
GTC (in-memory k-d tree)30.9 μs15.5×

3. Verification

All claims verified on RTX 4070 Laptop (8GB VRAM). Cache coverage measured via 5-fold cross-validation on held-out trajectory sets. Batch Jacobi validated against serial evaluation to float64 precision. Record compression verified via round-trip reconstruction. Scripts: benchmarks/gtc_* directories.

References

  1. Stewart, W.K.O. Organic Training Theory and the Manifold Hypothesis. HyperTensor Paper IV, 2026.
  2. Stewart, W.K.O. GRC Attention Compression. HyperTensor Paper I, 2026.