Devlog 08 — Aug 31: The Crossover Spectrum & v1.0 Ship

Finished multi-model crossover validation across three scales (0.5B, 3B, 27B):

// Adaptive Heterogeneous Scheduling
if (N <= 256) {
    // Tiny matrices (<60 KB) run on CPU SIMD to avoid driver dispatch overhead
    cpu_engine->run(weights, weight_bytes, N, K, act, out);
} else {
    // Medium/Large matrices stream to 16 AIE2 compute tiles
    xdna_engine->run(weights, weight_bytes, N, K, act, out);
}

Measured Crossover Scoreboard

• Qwen2.5-0.5B (403 MB): CPU favored (106.5 t/s vs 72.6 t/s XDNA hybrid) due to small 60 KB payloads.
• Qwen2.5-3B (1.86 GB): XDNA wins decisively: 16.5 tok/s vs 9.8 tok/s CPU (+68.8% / 1.69× speedup).
• Qwen3.8-27B (15.2 GB): XDNA sustains 1.72 tok/s streaming 12.6 GiB smoothly on a 32 GB laptop.
• Numerical Quality: Verified 100% Top-1 token match across all 3 scales.

Shipped v1.0.0 with standalone binaries and clean architecture docs!