[Mobile] Add support for dtypes and custom classes in model tracer #84795
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Summary: Currently, the model tracer generates the selected features YAML file only with used operators. This change adds support for dtypes and custom classes as well. We need to add the flag `-DENABLE_RECORD_KERNEL_FUNCTION_DTYPE` when building PyTorch in Instrumentation Mode (i.e. with `TRACING_BASED=1` for server builds) to enable capturing this data. Test Plan: Built using `USE_NUMPY=0 USE_DISTRIBUTED=0 USE_CUDA=0 TRACING_BASED=1 python setup.py develop` Ran the model tracer to observe this generated file: https://gist.github.com/dhruvbird/50e1860b39ae065e57d58f17e0912136 [ghstack-poisoned]
🔗 Helpful Links🧪 See artifacts and rendered test results at hud.pytorch.org/pr/84795
Note: Links to docs will display an error until the docs builds have been completed. ✅ No FailuresAs of commit bd8f366: This comment was automatically generated by Dr. CI and updates every 15 minutes. |
facebook-github-bot
added
the
oncall: jit
|
Hmm I think we need to have the pr together with mobile build as well. Have we tried that? |
Thanks for flagging! Yes, I just updated the PR summary to include it - I seem to have missed it earlier! |
…l tracer" Summary: Currently, the model tracer generates the selected features YAML file only with used operators. This change adds support for dtypes and custom classes as well. We need to add the flag `-DENABLE_RECORD_KERNEL_FUNCTION_DTYPE` when building PyTorch in Instrumentation Mode (i.e. with `TRACING_BASED=1` for server builds) to enable capturing this data. Test Plan: Built using `USE_NUMPY=0 USE_DISTRIBUTED=0 USE_CUDA=0 TRACING_BASED=1 python setup.py develop` Ran the model tracer to observe this generated file: https://gist.github.com/dhruvbird/50e1860b39ae065e57d58f17e0912136 Then used the generated YAML to built pytorch (minimal build) using the command ``` BUILD_PYTORCH_MOBILE_WITH_HOST_TOOLCHAIN=1 \ USE_LIGHTWEIGHT_DISPATCH=0 BUILD_LITE_INTERPRETER=1 \ SELECTED_OP_LIST=/tmp/selected_ops.yaml \ TRACING_BASED=1 \ ./scripts/build_mobile.sh ``` After that I generated a binary using this command: ``` g++ /tmp/main.cpp -L build_mobile/lib/ -I build_mobile/install/include/ -ffunction-sections -fdata-sections -Wl,--gc-sections \ -lpthread -lc10 -Wl,--whole-archive -ltorch_cpu -Wl,--no-whole-archive -ltorch -lXNNPACK \ -lpytorch_qnnpack -lcpuinfo -lclog -lpthreadpool -lkineto -lfmt -ldl -lc10 ``` The table below shows the size reduction in all build modes. | Build Type | Unstripped | Stripped | | ----------- | ----------- | ----------- | | Standard | 49MiB | 34MiB | | Minimal w/o dtype | 6.1MiB (12%) | 4.5MiB (18%) | | Minimal w/ dtype | 3.7MiB (7%) | 2.7MiB (11%) | [ghstack-poisoned]
…l tracer" Summary: Currently, the model tracer generates the selected features YAML file only with used operators. This change adds support for dtypes and custom classes as well. We need to add the flag `-DENABLE_RECORD_KERNEL_FUNCTION_DTYPE` when building PyTorch in Instrumentation Mode (i.e. with `TRACING_BASED=1` for server builds) to enable capturing this data. Test Plan: Built using `USE_NUMPY=0 USE_DISTRIBUTED=0 USE_CUDA=0 TRACING_BASED=1 python setup.py develop` Ran the model tracer to observe this generated file: https://gist.github.com/dhruvbird/50e1860b39ae065e57d58f17e0912136 Then used the generated YAML to built pytorch (minimal build) using the command ``` BUILD_PYTORCH_MOBILE_WITH_HOST_TOOLCHAIN=1 \ USE_LIGHTWEIGHT_DISPATCH=0 BUILD_LITE_INTERPRETER=1 \ SELECTED_OP_LIST=/tmp/selected_ops.yaml \ TRACING_BASED=1 \ ./scripts/build_mobile.sh ``` After that I generated a binary using this command: ``` g++ /tmp/main.cpp -L build_mobile/lib/ -I build_mobile/install/include/ -ffunction-sections -fdata-sections -Wl,--gc-sections \ -lpthread -lc10 -Wl,--whole-archive -ltorch_cpu -Wl,--no-whole-archive -ltorch -lXNNPACK \ -lpytorch_qnnpack -lcpuinfo -lclog -lpthreadpool -lkineto -lfmt -ldl -lc10 ``` The table below shows the size reduction in all build modes. | Build Type | Unstripped | Stripped | | ----------- | ----------- | ----------- | | Standard | 49MiB | 34MiB | | Minimal w/o dtype | 6.1MiB (12%) | 4.5MiB (18%) | | Minimal w/ dtype | 3.7MiB (7%) | 2.7MiB (11%) | [ghstack-poisoned]
…l tracer" Summary: Currently, the model tracer generates the selected features YAML file only with used operators. This change adds support for dtypes and custom classes as well. We need to add the flag `-DENABLE_RECORD_KERNEL_FUNCTION_DTYPE` when building PyTorch in Instrumentation Mode (i.e. with `TRACING_BASED=1` for server builds) to enable capturing this data. Test Plan: Built using `USE_NUMPY=0 USE_DISTRIBUTED=0 USE_CUDA=0 TRACING_BASED=1 python setup.py develop` Ran the model tracer to observe this generated file: https://gist.github.com/dhruvbird/50e1860b39ae065e57d58f17e0912136 Then used the generated YAML to built pytorch (minimal build) using the command ``` BUILD_PYTORCH_MOBILE_WITH_HOST_TOOLCHAIN=1 \ USE_LIGHTWEIGHT_DISPATCH=0 BUILD_LITE_INTERPRETER=1 \ SELECTED_OP_LIST=/tmp/selected_ops.yaml \ TRACING_BASED=1 \ ./scripts/build_mobile.sh ``` After that I generated a binary using this command: ``` g++ /tmp/main.cpp -L build_mobile/lib/ -I build_mobile/install/include/ -ffunction-sections -fdata-sections -Wl,--gc-sections \ -lpthread -lc10 -Wl,--whole-archive -ltorch_cpu -Wl,--no-whole-archive -ltorch -lXNNPACK \ -lpytorch_qnnpack -lcpuinfo -lclog -lpthreadpool -lkineto -lfmt -ldl -lc10 ``` The table below shows the size reduction in all build modes. | Build Type | Unstripped | Stripped | | ----------- | ----------- | ----------- | | Standard | 49MiB | 34MiB | | Minimal w/o dtype | 6.1MiB (12%) | 4.5MiB (18%) | | Minimal w/ dtype | 3.7MiB (7%) | 2.7MiB (11%) | [ghstack-poisoned]
dhruvbird
added a commit
that referenced
this pull request
Sep 10, 2022
Summary: Currently, the model tracer generates the selected features YAML file only with used operators. This change adds support for dtypes and custom classes as well. We need to add the flag `-DENABLE_RECORD_KERNEL_FUNCTION_DTYPE` when building PyTorch in Instrumentation Mode (i.e. with `TRACING_BASED=1` for server builds) to enable capturing this data. Test Plan: Built using `USE_NUMPY=0 USE_DISTRIBUTED=0 USE_CUDA=0 TRACING_BASED=1 python setup.py develop` Ran the model tracer to observe this generated file: https://gist.github.com/dhruvbird/50e1860b39ae065e57d58f17e0912136 ghstack-source-id: 76fe50b Pull Request resolved: #84795
|
The error |
…l tracer" Summary: Currently, the model tracer generates the selected features YAML file only with used operators. This change adds support for dtypes and custom classes as well. We need to add the flag `-DENABLE_RECORD_KERNEL_FUNCTION_DTYPE` when building PyTorch in Instrumentation Mode (i.e. with `TRACING_BASED=1` for server builds) to enable capturing this data. Test Plan: Built using `USE_NUMPY=0 USE_DISTRIBUTED=0 USE_CUDA=0 TRACING_BASED=1 python setup.py develop` Ran the model tracer to observe this generated file: https://gist.github.com/dhruvbird/50e1860b39ae065e57d58f17e0912136 Then used the generated YAML to built pytorch (minimal build) using the command ``` BUILD_PYTORCH_MOBILE_WITH_HOST_TOOLCHAIN=1 \ USE_LIGHTWEIGHT_DISPATCH=0 BUILD_LITE_INTERPRETER=1 \ SELECTED_OP_LIST=/tmp/selected_ops.yaml \ TRACING_BASED=1 \ ./scripts/build_mobile.sh ``` After that I generated a binary using this command: ``` g++ /tmp/main.cpp -L build_mobile/lib/ -I build_mobile/install/include/ -ffunction-sections -fdata-sections -Wl,--gc-sections \ -lpthread -lc10 -Wl,--whole-archive -ltorch_cpu -Wl,--no-whole-archive -ltorch -lXNNPACK \ -lpytorch_qnnpack -lcpuinfo -lclog -lpthreadpool -lkineto -lfmt -ldl -lc10 ``` The table below shows the size reduction in all build modes. | Build Type | Unstripped | Stripped | | ----------- | ----------- | ----------- | | Standard | 49MiB | 34MiB | | Minimal w/o dtype | 6.1MiB (12%) | 4.5MiB (18%) | | Minimal w/ dtype | 3.7MiB (7%) | 2.7MiB (11%) | [ghstack-poisoned]
|
@pytorchbot merge |
|
@pytorchbot successfully started a merge job. Check the current status here. |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Stack from ghstack (oldest at bottom):
Summary: Currently, the model tracer generates the selected features YAML file only with used operators. This change adds support for dtypes and custom classes as well.
We need to add the flag
-DENABLE_RECORD_KERNEL_FUNCTION_DTYPEwhen building PyTorch in Instrumentation Mode (i.e. withTRACING_BASED=1for server builds) to enable capturing this data.Test Plan: Built using
USE_NUMPY=0 USE_DISTRIBUTED=0 USE_CUDA=0 TRACING_BASED=1 python setup.py developRan the model tracer to observe this generated file: https://gist.github.com/dhruvbird/50e1860b39ae065e57d58f17e0912136
Then used the generated YAML to built pytorch (minimal build) using the command
After that I generated a binary using this command:
The table below shows the size reduction in all build modes.