clean up

2025-06-05 13:43:04 -07:00
parent 2768c94977
commit 43cb9075f2
4 changed files with 6 additions and 9 deletions
@@ -115,7 +115,6 @@ class ParquetVideoTextDataset(Dataset):
        else:
            pass

-        # get_world_group().barrier()
        dist.barrier()
        if validation:
            with open(self.plan_output_dir) as f:
@@ -126,4 +126,3 @@ class ForwardBatch:
        # Set do_classifier_free_guidance based on guidance scale and negative prompt
        if self.guidance_scale > 1.0:
            self.do_classifier_free_guidance = True
-            # self.negative_prompt_embeds = []
@@ -272,7 +272,6 @@ class WanTrainingPipeline(TrainingPipeline):
        logger.info("GPU memory usage before train_one_step: %s MB",
                    gpu_memory_usage)

-        # self._log_validation(self.transformer, self.training_args, 1)
        for step in range(self.init_steps + 1,
                          self.training_args.max_train_steps + 1):
            start_time = time.perf_counter()
@@ -2,15 +2,15 @@
 GPU_NUM=1 # 2,4,8
 MODEL_PATH="Wan-AI/Wan2.1-T2V-1.3B-Diffusers"
 MODEL_TYPE="wan"
-DATA_MERGE_PATH="data/cats_480/merge.txt"
-OUTPUT_DIR="cats_480_2_latents_parq_neg2"
-VALIDATION_PATH="assets/2_cat_prompt.txt"
+DATA_MERGE_PATH="your/path/to/Mixkit-Src/merge.txt"
+OUTPUT_DIR="your/path"
+VALIDATION_PATH="assets/prompt.txt"

 torchrun --nproc_per_node=$GPU_NUM \
    fastvideo/data_preprocess/preprocess.py \
    --model_path $MODEL_PATH \
    --data_merge_path $DATA_MERGE_PATH \
-    --preprocess_video_batch_size 1 \
+    --preprocess_video_batch_size 8 \
    --max_height 480 \
    --max_width 832 \
    --num_frames 81 \
@@ -19,6 +19,6 @@ torchrun --nproc_per_node=$GPU_NUM \
    --model_type $MODEL_TYPE \
    --train_fps 16 \
    --validation_prompt_txt $VALIDATION_PATH \
-    --samples_per_file 1 \
-    --flush_frequency 1 \
+    --samples_per_file 32 \
+    --flush_frequency 128 \
    --preprocess_task "t2v"