only process module's own parameters in Zero context
add zero hooks for all modules that contrain parameters
gather parameters only belonging to module itself
* fix merge conflict
modify unit test
remove unnessesary log info
reformat file
* remove unused module
* remove unnecessary sync function
* change doc string style from Google to Sphinx
* add zero1 (#209)
* add zero1
* add test zero1
* update zero stage 1 develop (#212)
* Implement naive zero3 (#240)
* naive zero3 works well
* add zero3 param manager
* add TODOs in comments
* add gather full param ctx
* fix sub module streams
* add offload
* fix bugs of hook and add unit tests
* fix bugs of hook and add unit tests (#252)
* add gather full param ctx
* fix sub module streams
* add offload
* fix bugs of hook and add unit tests
* polish code and add state dict hook
* fix bug
* update unit test
* refactor reconstructed zero code
* clip_grad support zero3 and add unit test
* add unit test for Zero3ParameterManager
* [WIP] initialize the shard param class
* [WIP] Yet another sharded model implementation (#274)
* [WIP] initialize the shard param class
* [WIP] Yes another implementation of shardModel. Using a better hook method.
* torch.concat -> torch.cat
* fix test_zero_level_1.py::test_zero_level_1 unitest
* remove deepspeed implementation and refactor for the reconstructed zero module
* polish zero dp unittests
Co-authored-by: ver217 <lhx0217@gmail.com>
Co-authored-by: Frank Lee <somerlee.9@gmail.com>
added branch context;
added vocab parallel layers;
moved split_batch from load_batch to tensor parallel embedding layers;
updated gpt model;
updated unit test cases;
fixed few collective communicator bugs
* add pipeline shared module wrapper and update load batch
* added model parallel process group for amp and clip grad (#86)
* added model parallel process group for amp and clip grad
* update amp and clip with model parallel process group
* remove pipeline_prev/next group (#88)
* micro batch offload
* optimize pipeline gpu memory usage
* pipeline can receive tensor shape (#93)
* optimize pipeline gpu memory usage
* fix grad accumulation step counter
* rename classes and functions
Co-authored-by: Frank Lee <somerlee.9@gmail.com>
* Add gradient accumulation, fix lr scheduler
* fix FP16 optimizer and adapted torch amp with tensor parallel (#18)
* fixed bugs in compatibility between torch amp and tensor parallel and performed some minor fixes
* fixed trainer
* Revert "fixed trainer"
This reverts commit 2e0b0b7699.
* improved consistency between trainer, engine and schedule (#23)
Co-authored-by: 1SAA <c2h214748@gmail.com>
* Split conv2d, class token, positional embedding in 2d, Fix random number in ddp
Fix convergence in cifar10, Imagenet1000
* Integrate 1d tensor parallel in Colossal-AI (#39)
* fixed 1D and 2D convergence (#38)
* optimized 2D operations
* fixed 1D ViT convergence problem
* Feature/ddp (#49)
* remove redundancy func in setup (#19) (#20)
* use env to control the language of doc (#24) (#25)
* Support TP-compatible Torch AMP and Update trainer API (#27)
* Add gradient accumulation, fix lr scheduler
* fix FP16 optimizer and adapted torch amp with tensor parallel (#18)
* fixed bugs in compatibility between torch amp and tensor parallel and performed some minor fixes
* fixed trainer
* Revert "fixed trainer"
This reverts commit 2e0b0b7699.
* improved consistency between trainer, engine and schedule (#23)
Co-authored-by: 1SAA <c2h214748@gmail.com>
Co-authored-by: 1SAA <c2h214748@gmail.com>
Co-authored-by: ver217 <lhx0217@gmail.com>
* add an example of ViT-B/16 and remove w_norm clipping in LAMB (#29)
* add explanation for ViT example (#35) (#36)
* support torch ddp
* fix loss accumulation
* add log for ddp
* change seed
* modify timing hook
Co-authored-by: Frank Lee <somerlee.9@gmail.com>
Co-authored-by: 1SAA <c2h214748@gmail.com>
Co-authored-by: binmakeswell <binmakeswell@gmail.com>
* Feature/pipeline (#40)
* remove redundancy func in setup (#19) (#20)
* use env to control the language of doc (#24) (#25)
* Support TP-compatible Torch AMP and Update trainer API (#27)
* Add gradient accumulation, fix lr scheduler
* fix FP16 optimizer and adapted torch amp with tensor parallel (#18)
* fixed bugs in compatibility between torch amp and tensor parallel and performed some minor fixes
* fixed trainer
* Revert "fixed trainer"
This reverts commit 2e0b0b7699.
* improved consistency between trainer, engine and schedule (#23)
Co-authored-by: 1SAA <c2h214748@gmail.com>
Co-authored-by: 1SAA <c2h214748@gmail.com>
Co-authored-by: ver217 <lhx0217@gmail.com>
* add an example of ViT-B/16 and remove w_norm clipping in LAMB (#29)
* add explanation for ViT example (#35) (#36)
* optimize communication of pipeline parallel
* fix grad clip for pipeline
Co-authored-by: Frank Lee <somerlee.9@gmail.com>
Co-authored-by: 1SAA <c2h214748@gmail.com>
Co-authored-by: binmakeswell <binmakeswell@gmail.com>
* optimized 3d layer to fix slow computation ; tested imagenet performance with 3d; reworked lr_scheduler config definition; fixed launch args; fixed some printing issues; simplified apis of 3d layers (#51)
* Update 2.5d layer code to get a similar accuracy on imagenet-1k dataset
* update api for better usability (#58)
update api for better usability
Co-authored-by: 1SAA <c2h214748@gmail.com>
Co-authored-by: ver217 <lhx0217@gmail.com>
Co-authored-by: puck_WCR <46049915+WANG-CR@users.noreply.github.com>
Co-authored-by: binmakeswell <binmakeswell@gmail.com>
Co-authored-by: アマデウス <kurisusnowdeng@users.noreply.github.com>
Co-authored-by: BoxiangW <45734921+BoxiangW@users.noreply.github.com>
* Add gradient accumulation, fix lr scheduler
* fix FP16 optimizer and adapted torch amp with tensor parallel (#18)
* fixed bugs in compatibility between torch amp and tensor parallel and performed some minor fixes
* fixed trainer
* Revert "fixed trainer"
This reverts commit 2e0b0b7699.
* improved consistency between trainer, engine and schedule (#23)
Co-authored-by: 1SAA <c2h214748@gmail.com>
Co-authored-by: 1SAA <c2h214748@gmail.com>
Co-authored-by: ver217 <lhx0217@gmail.com>
HELSON