- Tokyo/JAPAN
- info@metaclan.jp
- @RecyncAI
Popular repositories Loading
-
Recync_framework
Recync_framework PublicRuntime detection and control of LLM coherence failures (looping, hallucination, context loss). No fine-tuning. Zero iatrogenic harm. 69 experiments across 5 architectures.
Python
-
geometric_phase_extraction
geometric_phase_extraction PublicGeometric phase structure in Transformer hidden states. LayerNorm placement predicts manifold geometry — 6x difference in PCA concentration. 9 models, 13 experiments.
Python
-
norm-separated-quantization
norm-separated-quantization PublicTraining-free fix for KV cache INT4 failures. Norm separation + per-channel quantization. Qwen2-7B: 744× improvement (ΔPPL +238 → +0.32). 12 models, 124M–40B. 4 lines of PyTorch.
Python 1
-
manifold_topology_experiment
manifold_topology_experiment PublicTopology and geometry of Transformer hidden states via persistent homology. The manifold is a 1D arc (PC1=norm, r=0.999), not a torus or sphere. PC2=position, PC3=surprisal+POS. Validated up to 40B.
Python
If the problem persists, check the GitHub status page or contact support.
