From efcb3c4058d2d345215ab8d8980827c371c859ff Mon Sep 17 00:00:00 2001 From: RandNMR73 Date: Thu, 25 Dec 2025 02:00:45 +0000 Subject: [PATCH] 8 gpus --- .../finetune/wan_t2v_1.3B/crush_smol/finetune_t2v_sage3.sh | 6 +++--- 1 file changed, 3 insertions(+), 3 deletions(-) diff --git a/examples/training/finetune/wan_t2v_1.3B/crush_smol/finetune_t2v_sage3.sh b/examples/training/finetune/wan_t2v_1.3B/crush_smol/finetune_t2v_sage3.sh index c80063591..f7767bda1 100644 --- a/examples/training/finetune/wan_t2v_1.3B/crush_smol/finetune_t2v_sage3.sh +++ b/examples/training/finetune/wan_t2v_1.3B/crush_smol/finetune_t2v_sage3.sh @@ -10,7 +10,7 @@ export WANDB_API_KEY=2f25ad37933894dbf0966c838c0b8494987f9f2f MODEL_PATH="Wan-AI/Wan2.1-T2V-1.3B-Diffusers" DATA_DIR=data/Wan-Syn_77x448x832_600k VALIDATION_DATASET_FILE="examples/training/finetune/wan_t2v_1.3B/crush_smol/validation.json" -NUM_GPUS=4 +NUM_GPUS=8 # export CUDA_VISIBLE_DEVICES=4,5 @@ -19,7 +19,7 @@ training_args=( --tracker_project_name "wan_t2v_finetune_qat" --output_dir "checkpoints/wan_t2v_finetune_sage3" --max_train_steps 5000 - --train_batch_size 32 + --train_batch_size 16 --train_sp_batch_size 1 --gradient_accumulation_steps 1 --num_latent_t 12 @@ -33,7 +33,7 @@ training_args=( # Parallel arguments parallel_args=( --num_gpus $NUM_GPUS - --sp_size 4 + --sp_size $NUM_GPUS --tp_size 1 --hsdp_replicate_dim 1 --hsdp_shard_dim $NUM_GPUS