Skip to content

Actions: liuliu/ccv

cuda-int-tests

Actions

Loading...
Loading

Show workflow options

Create status badge

Loading
61 workflow runs
61 workflow runs

Filter by Event

Filter by Status

Filter by Branch

Filter by Actor

Fix issue with mul_back that not check input size.
cuda-int-tests #61: Commit c8d8187 pushed by liuliu
November 5, 2024 00:55 1h 37m 55s unstable
November 5, 2024 00:55 1h 37m 55s
Switch D to BF16, also need to gate for macOS 14.
cuda-int-tests #60: Commit de59d3e pushed by liuliu
October 31, 2024 04:38 26m 29s unstable
October 31, 2024 04:38 26m 29s
Fix various issues related to running backprop of sdpa on fp16.
cuda-int-tests #59: Commit fe99c48 pushed by liuliu
October 30, 2024 05:17 28m 18s unstable
October 30, 2024 05:17 28m 18s
Add support for MFA backprop.
cuda-int-tests #58: Commit 18f1860 pushed by liuliu
October 30, 2024 02:01 26m 34s unstable
October 30, 2024 02:01 26m 34s
Support backward of cmul.
cuda-int-tests #57: Commit d69ee17 pushed by liuliu
October 29, 2024 04:44 1h 29m 0s unstable
October 29, 2024 04:44 1h 29m 0s
Remove ccv_nnc_mfa_cmul kernel, move that to CMulKernel.
cuda-int-tests #56: Commit e4d13c2 pushed by liuliu
October 28, 2024 21:29 1h 4m 7s unstable
October 28, 2024 21:29 1h 4m 7s
Fix some minor issues related to NHWC on CUDA with more than 4 dims.
cuda-int-tests #55: Commit 34d496a pushed by liuliu
October 25, 2024 17:30 44m 38s unstable
October 25, 2024 17:30 44m 38s
When compute whether a tensor is fully covered, derive new shape from
cuda-int-tests #54: Commit c6ce5b1 pushed by liuliu
October 22, 2024 05:16 1h 20m 47s unstable
October 22, 2024 05:16 1h 20m 47s
Allow to disable gradient checkpointing inside a gradient checkpointi…
cuda-int-tests #53: Commit 20d998d pushed by liuliu
October 21, 2024 18:35 45m 48s unstable
October 21, 2024 18:35 45m 48s
Revert previous change, also disable flashattn in the unit tests for …
cuda-int-tests #52: Commit b361b30 pushed by liuliu
October 18, 2024 23:52 48m 52s unstable
October 18, 2024 23:52 48m 52s
Add the logic to tighten-up the inputs.
cuda-int-tests #51: Commit a003506 pushed by liuliu
October 18, 2024 21:43 24m 52s unstable
October 18, 2024 21:43 24m 52s
Improve gradient checkpointing implementation to:
cuda-int-tests #50: Commit a02a510 pushed by liuliu
October 18, 2024 06:46 5m 3s unstable
October 18, 2024 06:46 5m 3s
Make sure when dequant weights, take into account cublas' workspace m…
cuda-int-tests #49: Commit bed628c pushed by liuliu
October 16, 2024 23:54 49m 51s unstable
October 16, 2024 23:54 49m 51s
Use dequant datatype throughout in gemm kernels.
cuda-int-tests #48: Commit 1d2beb4 pushed by liuliu
October 16, 2024 00:03 6h 31m 44s unstable
October 16, 2024 00:03 6h 31m 44s
Add cancel function for concrete graph.
cuda-int-tests #47: Commit e7c7639 pushed by liuliu
October 13, 2024 22:23 25m 35s unstable
October 13, 2024 22:23 25m 35s
Add ccv_cnnp_debug.
cuda-int-tests #46: Commit ef7bd53 pushed by liuliu
September 28, 2024 05:59 6h 10m 40s unstable
September 28, 2024 05:59 6h 10m 40s
Fix a typo.
cuda-int-tests #45: Commit d2622a3 pushed by liuliu
September 16, 2024 04:50 1h 50m 20s unstable
September 16, 2024 04:50 1h 50m 20s
Fix bug where the shader cache is not used properly.
cuda-int-tests #44: Commit 67887cd pushed by liuliu
September 16, 2024 04:46 41m 32s unstable
September 16, 2024 04:46 41m 32s
Fix a bug on the flag is not inspected properly.
cuda-int-tests #43: Commit d53469e pushed by liuliu
September 16, 2024 00:12 3h 9m 3s unstable
September 16, 2024 00:12 3h 9m 3s
Make sure we do low precision intermediate the same way as Swift repo.
cuda-int-tests #42: Commit eda29de pushed by liuliu
September 15, 2024 23:47 1h 38m 17s unstable
September 15, 2024 23:47 1h 38m 17s
Temporarily gate against BF16.
cuda-int-tests #41: Commit ffd6604 pushed by liuliu
September 15, 2024 22:39 2h 6m 9s unstable
September 15, 2024 22:39 2h 6m 9s
Switch SDPA to default to high precision and only you can optionally …
cuda-int-tests #40: Commit 2134068 pushed by liuliu
September 15, 2024 22:04 1h 36m 6s unstable
September 15, 2024 22:04 1h 36m 6s
Add code to support switching load offset to be computed immediately …
cuda-int-tests #39: Commit 546e9bc pushed by liuliu
September 15, 2024 21:52 40m 19s unstable
September 15, 2024 21:52 40m 19s
Pass in lse from the op.
cuda-int-tests #38: Commit a58acd2 pushed by liuliu
September 15, 2024 16:57 50m 14s unstable
September 15, 2024 16:57 50m 14s
Fix a bug caused low precision intermediates not working.
cuda-int-tests #37: Commit c67441f pushed by liuliu
September 14, 2024 18:17 41m 36s unstable
September 14, 2024 18:17 41m 36s