refactor dynamic training

scripts/accelerate_configs/ddp_config.yaml

@@ -1,11 +1,11 @@
 compute_environment: LOCAL_MACHINE
-distributed_type: MULTI_GPU
+distributed_type: NO 
 fsdp_config: {}
 machine_rank: 0
 main_process_ip: null
 main_process_port: 29501
 main_training_function: main
 num_machines: 1
-num_processes: 8
-gpu_ids: 0,1,2,3,4,5,6,7
+num_processes: 1
+gpu_ids: 0,
 use_cpu: false

scripts/text_condition/train_videoae_65x512x512.sh

@@ -2,7 +2,7 @@ export WANDB_KEY=""
 export ENTITY=""
 export PROJECT="test"
 accelerate launch \
-    --config_file scripts/accelerate_configs/deepspeed_zero2_config.yaml \
+    --config_file scripts/accelerate_configs/ddp_config.yaml \
     opensora/train/train_t2v.py \
     --model LatteT2V-XL/122 \
     --text_encoder_name DeepFloyd/t5-v1_1-xxl \
@@ -17,15 +17,15 @@ accelerate launch \
     --max_image_size 512 \
     --gradient_checkpointing \
     --attention_mode xformers \
-    --train_batch_size=2 \
-    --dataloader_num_workers 10 \
+    --train_batch_size=4 \
+    --dataloader_num_workers 0 \
     --gradient_accumulation_steps=1 \
     --max_train_steps=1000000 \
     --learning_rate=2e-05 \
     --lr_scheduler="constant" \
     --lr_warmup_steps=0 \
     --mixed_precision="bf16" \
-    --report_to="wandb" \
+    --report_to="tensorboard" \
     --checkpointing_steps=500 \
     --output_dir="test" \
     --allow_tf32 \