GitLab

* 1

* change to channel 1

* cogview4 control training

* add CacheMixin

* 1

* remove initial_input_channels change for val

* 1

* update

* use 3.5

* new loss

* 1

* use imagetoken

* for megatron convert

* 1

* train con and uc

* 2

* remove guidance_scale

* Update pipeline_cogview4_control.py

* fix

* use cogview4 pipeline with timestep

* update shift_factor

* remove the uncond

* add max length

* change convert and use GLMModel instead of GLMForCasualLM

* fix

* [cogview4] Add attention mask support to transformer model

* [fix] Add attention mask for padded token

* update

* remove padding type

* Update train_control_cogview4.py

* resolve conflicts with #10981

* add control convert

* use control format

* fix

* add missing import

* update with cogview4 formate

* make style

* Update pipeline_cogview4_control.py

* Update pipeline_cogview4_control.py

* remove

* Update pipeline_cogview4_control.py

* put back

* Apply style fixes

---------
Co-authored-by: OleehyO <leehy0357@gmail.com>
Co-authored-by: yiyixuxu <yixu310@gmail.com>
Co-authored-by: github-actions[bot] <github-actions[bot]@users.noreply.github.com>