feat: use NVRTC for DeepGEMM JIT compilation #3239
Conversation
lucifer1004
force-pushed
the
user/zihuaw/deep_gemm_nvrtc
branch
from
a411cdb to
7dafe95
Compare
|
/bot run |
|
PR_Github #1012 [ run ] triggered by Bot |
|
PR_Github #1012 [ run ] completed with state |
lucifer1004
force-pushed
the
user/zihuaw/deep_gemm_nvrtc
branch
from
7dafe95 to
7ec7850
Compare
|
/bot run |
|
PR_Github #1068 [ run ] triggered by Bot |
|
PR_Github #1068 [ run ] completed with state |
lucifer1004
force-pushed
the
user/zihuaw/deep_gemm_nvrtc
branch
from
26926bc to
3fa9f31
Compare
|
/bot run |
|
PR_Github #1103 [ run ] triggered by Bot |
|
I am approving this. The overall code LGTM. The on-by-default of DeepGemm for DS model should be converged before merging. |
|
PR_Github #1103 [ run ] completed with state |
|
/bot run --reuse-test |
|
PR_Github #1118 [ run ] triggered by Bot |
|
PR_Github #1118 [ run ] completed with state |
|
/bot run --reuse-test |
1 similar comment
|
/bot run --reuse-test |
|
PR_Github #1134 [ run ] triggered by Bot |
|
PR_Github #1134 [ run ] completed with state |
|
/bot run |
|
PR_Github #1149 [ run ] triggered by Bot |
|
PR_Github #1149 [ run ] completed with state |
|
/bot run |
|
PR_Github #1169 [ run ] triggered by Bot |
lucifer1004
force-pushed
the
user/zihuaw/deep_gemm_nvrtc
branch
from
8b9c133 to
1c39a52
Compare
|
/bot run |
|
PR_Github #1238 [ run ] triggered by Bot |
|
PR_Github #1238 [ run ] completed with state |
lucifer1004
force-pushed
the
user/zihuaw/deep_gemm_nvrtc
branch
from
1c39a52 to
08ff654
Compare
|
/bot run --stage-list "DGX_H100-4_GPUs-PyTorch-1, DGX_H100-4_GPUs-PyTorch-2" |
|
PR_Github #1291 [ run ] triggered by Bot |
|
PR_Github #1291 [ run ] completed with state |
lucifer1004
force-pushed
the
user/zihuaw/deep_gemm_nvrtc
branch
from
08ff654 to
a04862b
Compare
|
/bot skip --comment "L0_MergeRequest_PR/931 + L0_MergeRequest_PR/973" |
|
PR_Github #1318 [ skip ] triggered by Bot |
|
PR_Github #1318 [ skip ] completed with state |
Signed-off-by: Zihua Wu <[email protected]>
Signed-off-by: Zihua Wu <[email protected]>
Signed-off-by: Zihua Wu <[email protected]>
Signed-off-by: Zihua Wu <[email protected]>
Signed-off-by: Zihua Wu <[email protected]>
Signed-off-by: Zihua Wu <[email protected]>
juney-nvidia
force-pushed
the
user/zihuaw/deep_gemm_nvrtc
branch
from
a04862b to
8c51c85
Compare
|
/bot reuse-pipeline |
|
PR_Github #1322 [ reuse-pipeline ] triggered by Bot |
Thanks, Tao. I am okay to firstly merge this MR, then have another MR to enable DeepGEMM by default. Thus to amortize the overhead for Zihua a little bit :) June |
|
PR_Github #1322 [ reuse-pipeline ] completed with state |
* feat: use NVRTC for DeepGEMM JIT compilation Signed-off-by: Zihua Wu * fix: add license Signed-off-by: Zihua Wu * feat: store NVRTC JIT results in memory by default Signed-off-by: Zihua Wu * feat: refinement Signed-off-by: Zihua Wu * feat: refinement Signed-off-by: Zihua Wu * test: set timeout to 7200 Signed-off-by: Zihua Wu --------- Signed-off-by: Zihua Wu Signed-off-by: sarattha <[email protected]>
* feat: use NVRTC for DeepGEMM JIT compilation Signed-off-by: Zihua Wu * fix: add license Signed-off-by: Zihua Wu * feat: store NVRTC JIT results in memory by default Signed-off-by: Zihua Wu * feat: refinement Signed-off-by: Zihua Wu * feat: refinement Signed-off-by: Zihua Wu * test: set timeout to 7200 Signed-off-by: Zihua Wu --------- Signed-off-by: Zihua Wu
* feat: use NVRTC for DeepGEMM JIT compilation Signed-off-by: Zihua Wu * fix: add license Signed-off-by: Zihua Wu * feat: store NVRTC JIT results in memory by default Signed-off-by: Zihua Wu * feat: refinement Signed-off-by: Zihua Wu * feat: refinement Signed-off-by: Zihua Wu * test: set timeout to 7200 Signed-off-by: Zihua Wu --------- Signed-off-by: Zihua Wu
No description provided.