-
Remove tests/.pytest_cache
-
Card: standardized form with hero
-
Standardize card: usage-first
-
batched decode
-
Uploaded using `kernel-builder`.
-
batched decode
-
batched decode; batch-invariance gates
-
batched decode; batch-invariance gates
-
gemma-4 coverage and measured numbers
-
model-as-a-kernel v1 source
-
card
-
mechanism description current
-
model-as-a-kernel v1 source
-
model-as-a-kernel v1 source
-
model-as-a-kernel v1 source
-
model-as-a-kernel v1 source
-
measured and certified numbers
-
model-as-a-kernel v1 source
-
model-as-a-kernel v1 source
-
measured numbers
-
model-as-a-kernel v1 source
-
measured numbers
-
model-as-a-kernel v1 source
-
measured and certified numbers
-
model-as-a-kernel v1 source
-
model-as-a-kernel v1 source
-
model-as-a-kernel v1 source
-
model-as-a-kernel v1 source
-
model-as-a-kernel v1 source
-
certified numbers
-
coverage gate calibration
-
tests: phase index, ulp gate
-
tests: chunk-sized buffers
-
tests: op signatures
-
model-as-a-kernel v1 source
-
model-as-a-kernel v1 source
-
model-as-a-kernel v1 source
-
model-as-a-kernel v1 source
-
model-as-a-kernel v1 source
-
model-as-a-kernel v1 source
-
card
-
model-as-a-kernel v1 source
-
certification gates: bulk statistics vs HF pairing spread, layer-0 KV sequence check
-
model-as-a-kernel v1 source
-
model-as-a-kernel v1 source
-
initial commit