GitLab

Co-authored-by: Yuanchen Xu <yuanchen.xu00@gmail.com>

* refactor: separate log_probs fn from Actor forward fn

* refactor: separate generate fn from Actor class

* feat: update unwrap_model and get_base_model
* unwrap_model returns model not wrapped by Strategy
* get_base_model returns HF model for Actor, Critic and RewardModel

* feat: simplify Strategy.prepare

* style: remove get_base_model method of Actor

* perf: tokenize text in batches

* refactor: move calc_action_log_probs to utils of model

* test: update test with new forward fn

* style: rename forward fn args

* fix: do not unwrap model in save_model fn of naive strategy

* test: add gemini test for train_prompts

* fix: fix _set_default_generate_kwargs

Revert "[sync] sync feature/shardformer with develop"

[sync] sync feature/shardformer with develop

[example] Adding an example of training dreambooth with the new booster API

Co-authored-by: Yuanchen Xu <yuanchen.xu00@gmail.com>

* add gpt2 policy and modify shard and slicer to support

* remove unused code

* polish code

* add dropout layer, add dropout test

* modify seed manager as context manager

* add a copy of col_nn.layer

* add dist_crossentropy loss; separate module test

* polish the code

* fix dist crossentropy loss

* update readme with modules content

* remove img

* [shardformer] refactored the user api

* polish code

* init shardformer code structure

* add implement of sharder (inject and replace)

* add implement of replace layer to colossal layer

* separate different layer policy, add some notion

* implement 1d and 2d slicer, can tell col or row

* fix bug when slicing and inject model

* fix some bug; add inference test example

* add share weight and train example

* add train

* add docstring and readme

* add docstring for other files

* pre-commit

* init shardformer code structure

* add implement of sharder (inject and replace)

* add implement of replace layer to colossal layer

* separate different layer policy, add some notion

* implement 1d and 2d slicer, can tell col or row

* fix bug when slicing and inject model

* fix some bug; add inference test example

[feature] updated device mesh and dtensor

* [doc] add lazy init en doc

* [doc] add lazy init zh doc

* [doc] add lazy init doc in sidebar

* [doc] add lazy init doc test

* [doc] fix lazy init doc link

* Modify torch version requirement to adapt torch 2.0

* modify palm example using new booster API

* roll back

* fix port

* polish

* polish

[sync] sync feature/dtensor with develop