Skip to content

Actions: yhirose/cpp-tensorlib

Actions

All workflows

Actions

Loading...
Loading

Showing runs from all workflows
145 workflow runs
145 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
webgpu: the kernels take the canonical ABI, so marshal_ is gone CI #145: Commit 43809a2 pushed by yhirose
8m 54s master
master
8m 54s
webgpu: the kernels take the canonical ABI, so marshal_ is gone CI #144: Pull request #6 synchronize by yhirose
6m 30s backend-layer-4
backend-layer-4
6m 30s
gpu: the mirror table and its buffer pool live once, in gpu_abi.h CI #143: Commit da085de pushed by yhirose
8m 18s master
master
8m 18s
webgpu: the kernels take the canonical ABI, so marshal_ is gone CI #142: Pull request #5 opened by yhirose
6m 55s backend-layer-4
backend-layer-4
6m 55s
gpu: three of the four remaining backend-layer gaps CI #141: Pull request #4 opened by yhirose
6m 35s backend-layer-3
backend-layer-3
6m 35s
cuda: the wave plan's fill stays long, the one width every arch agr… CI #140: Commit 6abecc6 pushed by yhirose
8m 11s master
master
8m 11s
gpu: launch recording, split policy and the model path, all in one place CI #139: Pull request #3 synchronize by yhirose
8m 55s backend-layer-2
backend-layer-2
8m 55s
gpu: launch recording, split policy and the model path, all in one place CI #138: Pull request #3 opened by yhirose
5m 43s backend-layer-2
backend-layer-2
5m 43s
gpu: a fourth backend, written the way the docs say one is CI #137: Commit 5911d39 pushed by yhirose
8m 48s master
master
8m 48s
gpu: a shared op layer over the backends, views as the currency, and a fourth backend to prove it CI #136: Pull request #2 opened by yhirose
8m 21s backend-layer
backend-layer
8m 21s
metal: the [N,K] weight layouts, so a prompt prefills as GEMMs CI #135: Commit 62289b2 pushed by yhirose
5m 11s master
master
5m 11s
models: the Qwen2.5 driver runs on whatever backend gpu:: resolves to CI #134: Commit 0cff4da pushed by yhirose
8m 18s master
master
8m 18s
parity: compare each op's signature, not only whether it is a stub CI #133: Commit 1ed6f56 pushed by yhirose
8m 18s master
master
8m 18s
caps: say whether the model path is real, and skip the test where it … CI #132: Commit 9e13a0b pushed by yhirose
7m 59s master
master
7m 59s
kv_cache: the model header and the WebGPU stub follow the new signature CI #131: Commit 8d4614a pushed by yhirose
8m 18s master
master
8m 18s
kv_cache: one cache over the gpu:: facade, and Metal's model path CI #130: Commit f171307 pushed by yhirose
5m 40s master
master
5m 40s
tokenizer: name it for the BPE family it implements, not for one model CI #129: Commit 88e72b4 pushed by yhirose
8m 27s master
master
8m 27s
metal: causal attention and its pullback on simdgroup_matrix CI #128: Commit afff16d pushed by yhirose
7m 23s master
master
7m 23s
storage: byte-sized allocations go through the runtime hook too CI #127: Commit e4d7b0b pushed by yhirose
6m 39s master
master
6m 39s
metal: a deep sum_to per threadgroup, index_add gathered per output row CI #126: Commit 83daa26 pushed by yhirose
8m 17s master
master
8m 17s
view: a q4 array can't be viewed, since its packed layout has no per-… CI #125: Commit 003fbad pushed by yhirose
8m 35s master
master
8m 35s
index_add reads its indices through their stride CI #124: Commit 8c37c9a pushed by yhirose
6m 22s master
master
6m 22s
The fused pullbacks run on the own CPU too CI #123: Commit 15479f0 pushed by yhirose
8m 15s master
master
8m 15s
softmax uses exp_shifted's sum; the attention tile folds its max in l… CI #122: Commit c33dadf pushed by yhirose
5m 38s master
master
5m 38s
broadcast_to says where elementwise work goes; softmax checks against… CI #121: Commit 14cb1ca pushed by yhirose
7m 56s master
master
7m 56s