[Unity][MSC][M2.1] Add pruner for model pruning #16186
Merged
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Outdated suggestions cannot be applied.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.
This is a pull request for MSC(Multi-System Compile)
RFC: https://discuss.tvm.apache.org/t/rfc-unity-msc-introduction-to-multi-system-compiler/15251/5
Tracking issue: #15233
This is the Milestone 2 for MSC: Use msc.runtime.Manager to manage the compiling pipeline && tools.
To limit each PR in reviewable size, the Milestone 2 will be split into some steps:
M2.1 Add Manager for compile pipeline
M2.2 Add pruner for model pruning
M2.3 Add tracker for track layer datas
M2.4 Add quantizer for quantize model
The tools are basic modules for optimize the model. They works in model rewriting as well as runtime controlling. Tool can be used in training system as well as in deploy system.
This PR add basic part of tool and pruner for all frameworks. The pruner in torch, tvm and tensorrt are tested. Pruner compress the model by given density. It works together with gym(auto optimization system, which will be added in later PRs) to prune the model automatically. It can also works together with distiller (will be added in later PRs) to update weights in pruned model.
WeightGraph is used in pruning. A WeightGraph describe the topology of weights in model. This graph can also be used in model parallel, which requires split the weights depends on topology.
cc @Hzfengsy @junrushao