Skip to content

Actions: yhirose/cpp-tensorlib

Actions

All workflows

Actions

Loading...
Loading

Showing runs from all workflows
146 workflow runs
146 workflow runs

Filter by Workflow

⌥ + click/return to exclude

Filter by Event

⌥ + click/return to exclude

Filter by Status

⌥ + click/return to exclude

Filter by Branch

⌥ + click/return to exclude

Filter by Actor

⌥ + click/return to exclude
broadcast_to says where elementwise work goes; softmax checks against… CI #121: Commit 14cb1ca pushed by yhirose
7m 56s master
master
7m 56s
clone's device arm goes through a runtime hook CI #120: Commit 8b4ed50 pushed by yhirose
6m 8s master
master
6m 8s
Drop the widened-view push-down CI #119: Commit 1ad0ec8 pushed by yhirose
7m 33s master
master
7m 33s
profile: say which declines leave no row CI #118: Commit 0de6823 pushed by yhirose
7m 56s master
master
7m 56s
eager ops name themselves only once they run; layer_norm_bwd tidied CI #117: Commit 3facd71 pushed by yhirose
6m 55s master
master
6m 55s
layer_norm_bwd: the pullback in three launches CI #116: Commit 09fac95 pushed by yhirose
7m 40s master
master
7m 40s
profile: rows carry their kind, metal binds through one place CI #115: Commit 46e7566 pushed by yhirose
6m 16s master
master
6m 16s
profile: where evaluation goes, by scope, op and kernel CI #114: Commit 80f1e0f pushed by yhirose
7m 47s master
master
7m 47s
metal copy_nd: the test is matches_gpu_oracle, the enum sits by where_nd CI #113: Commit e1be994 pushed by yhirose
7m 16s master
master
7m 16s
metal: copy_nd, so a strided view clones on the device CI #112: Commit 8f5529e pushed by yhirose
7m 53s master
master
7m 53s
adam_step: device_rmw_, and the comments say what the fallback does CI #111: Commit 181ea18 pushed by yhirose
6m 14s master
master
6m 14s
adam_step reads its state up before writing it; one metal stub, not two CI #110: Commit 4398c09 pushed by yhirose
7m 27s master
master
7m 27s
adam_step is total where the layout allows it CI #109: Commit 4ad2978 pushed by yhirose
7m 21s master
master
7m 21s
adam_step: the optimizer's update in one pass CI #108: Commit f467b4d pushed by yhirose
7m 10s master
master
7m 10s
the widened push-down covers every elementwise builder CI #107: Commit 8679b4c pushed by yhirose
7m 7s master
master
7m 7s
elementwise scalars push down through a widened view CI #106: Commit 733fb6e pushed by yhirose
6m 54s master
master
6m 54s
test: the fused pullback needs a device, not just a CUDA build CI #105: Commit b7b2d83 pushed by yhirose
7m 0s master
master
7m 0s
cross-entropy without a [rows, classes] intermediate CI #104: Commit 2678c24 pushed by yhirose
6m 48s master
master
6m 48s
census: one scope for what the measurement must not disturb CI #103: Commit 0b08154 pushed by yhirose
5m 52s master
master
5m 52s
auto census: measure outside the caller's defer_flush scope CI #102: Commit 95d7013 pushed by yhirose
5m 43s master
master
5m 43s
test: only a CUDA build requires the fused pullback CI #101: Commit 41e2420 pushed by yhirose
6m 53s master
master
6m 53s
attention: a fused pullback that keeps the scores in registers CI #100: Commit 89029b8 pushed by yhirose
6m 50s master
master
6m 50s
sum and mean off the last axis take the blocked kernel, not the host CI #99: Commit 47ab945 pushed by yhirose
6m 21s master
master
6m 21s
sum_to: a deep reduction gets a block per output, not a thread CI #98: Commit e8d412e pushed by yhirose
6m 40s master
master
6m 40s
broadcast_to: widening a gradient is a view, not a materialized copy CI #97: Commit 95d0fe8 pushed by yhirose
4m 41s master
master
4m 41s