llm-atlas/experiments/k3/attnres_local_path/README.md

Attention Residuals local mixer-path diagnostics

This directory implements preregistered protocol llm-atlas-k3-attnres-local-path-v1.

It is a targeted follow-up to Round 06. It exact-replays the same depth-32 Block training and keeps the learned forward unchanged while switching source value-gradient coefficients only at frozen mixer scopes. It is not a Kimi K3 checkpoint run, a trainable variant, an additive attribution, or a reproduction of unpublished Figure 5 telemetry.

Frozen environment

Python /home/wuyang/.pyenv/versions/3.10.14/envs/navi-router-cu128/bin/python
PyTorch 2.11.0+cu128
GPU NVIDIA GeForce RTX 5090
CUBLAS_WORKSPACE_CONFIG=:4096:8

The runner and exact commands will be added after this preregistration manifest is committed. No result file may precede that commit.