add debertav2

ppfleetx/configs/multimodal/imagen/imagen_super_resolution_1024.yaml

@@ -8,7 +8,7 @@ Global:
 
 Model:
   name: imagen_SR1024
-  text_encoder_name: t5/t5-11b
+  text_encoder_name: projects/imagen/t5/t5-11b
   text_embed_dim: 1024 
   timesteps: 1000 
   in_chans: 3
@@ -25,12 +25,13 @@ Model:
   dynamic_thresholding_percentile: 0.95
   only_train_unet_number: 1 
   use_recompute: False
+  recompute_granularity:
 
 Data:
   Train:
     dataset:
       name: ImagenDataset
-      input_path: ./projects/imagen/filelist/cc12m_base64.lst
+      input_path: ./projects/imagen/filelist/laion_400M/train
       shuffle: True
       input_resolution: 1024 
       max_seq_len: 128 

ppfleetx/configs/multimodal/imagen/imagen_super_resolution_256.yaml

@@ -53,7 +53,7 @@ Loss:
   p2_loss_weight_k: 1.0
 
 Distributed:
-  dp_degree: 1
+  dp_degree: 128
   mp_degree: 1
   pp_degree: 1
   sharding:

ppfleetx/configs/multimodal/imagen/imagen_text2im_64x64_DebertaV2.yaml

@@ -25,6 +25,7 @@ Model:
   dynamic_thresholding_percentile: 0.95
   only_train_unet_number: 1 
   use_recompute: False
+  recompute_granularity:
 
 Loss:
   name: mse_loss

ppfleetx/models/multimodal_model/imagen/modeling.py

@@ -23,6 +23,7 @@
 
 from .unet import Unet
 from ppfleetx.models.language_model.t5 import *
+from ppfleetx.models.language_model.debertav2 import *
 from ppfleetx.data.tokenizers import get_t5_tokenizer
 from .utils import (
     GaussianDiffusionContinuousTimes, default, exists, cast_tuple, first,

projects/imagen/run_text2im_2B_64x64_T5-11B_sharding8_dp32.sh

@@ -22,5 +22,5 @@ python -m paddle.distributed.launch --log_dir $log_dir --devices "0,1,2,3,4,5,6,
     ./tools/train.py \
     -c ./ppfleetx/configs/multimodal/imagen/imagen_text2im_64x64_T5-11B.yaml \
     -o Distributed.sharding.sharding_stage=2 \
-    -o Distributed.dp_degree=32 \ 
+    -o Distributed.dp_degree=32 \
     -o Distributed.sharding.sharding_degree=8