| Name | Last commit | Updated |
|---|
|
..
|
|
__init__.py
|
Gate A/B: scaffold, mandate, inventory tooling, prior art (59 refs),... |
|
|
distributed.py
|
H4: fused up+gate option (bit-identical, no end-to-end gain), D-011, LOG |
|
|
moe.py
|
H4: fused up+gate option (bit-identical, no end-to-end gain), D-011, LOG |
|
|
predictors.py
|
Probe predictor (fixed) results, depth variants, attribution figure/table, H4... |
|
|
prefetch.py
|
Fix: MLX arrays created on the consuming thread only (worker/stacked-layout... |
|
|
registry.py
|
SSD tier: parallel demand reads + expert-contiguous TITAN layout (cold 16-slot... |
|
|
runtime.py
|
H4: fused up+gate option (bit-identical, no end-to-end gain), D-011, LOG |
|
|
slots.py
|
H4: fused up+gate option (bit-identical, no end-to-end gain), D-011, LOG |
|
|
store.py
|
Fix: MLX arrays created on the consuming thread only (worker/stacked-layout... |
|