From 9dbe1ea261b6cccbccd3498e897999854f9fadcd Mon Sep 17 00:00:00 2001 From: Bubbliiiing <47347516+bubbliiiing@users.noreply.github.com> Date: Wed, 22 Oct 2025 11:23:26 +0800 Subject: [PATCH] Update video_sample_stride in fantasy talking (#359) --- scripts/fantasytalking/README_TRAIN.md | 8 ++++---- scripts/fantasytalking/train.sh | 2 +- 2 files changed, 5 insertions(+), 5 deletions(-) diff --git a/scripts/fantasytalking/README_TRAIN.md b/scripts/fantasytalking/README_TRAIN.md index 0edf4b7..c476abc 100755 --- a/scripts/fantasytalking/README_TRAIN.md +++ b/scripts/fantasytalking/README_TRAIN.md @@ -51,7 +51,7 @@ accelerate launch --mixed_precision="bf16" scripts/fantasytalking/train.py \ --train_data_meta=$DATASET_META_NAME \ --video_sample_size=512 \ --token_sample_size=512 \ - --video_sample_stride=2 \ + --video_sample_stride=1 \ --video_sample_n_frames=81 \ --train_batch_size=1 \ --gradient_accumulation_steps=1 \ @@ -96,7 +96,7 @@ accelerate launch --use_deepspeed --deepspeed_config_file config/zero_stage2_con --train_data_meta=$DATASET_META_NAME \ --video_sample_size=512 \ --token_sample_size=512 \ - --video_sample_stride=2 \ + --video_sample_stride=1 \ --video_sample_n_frames=81 \ --train_batch_size=1 \ --gradient_accumulation_steps=1 \ @@ -146,7 +146,7 @@ accelerate launch --zero_stage 3 --zero3_save_16bit_model true --zero3_init_flag --train_data_meta=$DATASET_META_NAME \ --video_sample_size=512 \ --token_sample_size=512 \ - --video_sample_stride=2 \ + --video_sample_stride=1 \ --video_sample_n_frames=81 \ --train_batch_size=1 \ --gradient_accumulation_steps=1 \ @@ -192,7 +192,7 @@ accelerate launch --mixed_precision="bf16" --use_fsdp --fsdp_auto_wrap_policy TR --train_data_meta=$DATASET_META_NAME \ --video_sample_size=512 \ --token_sample_size=512 \ - --video_sample_stride=2 \ + --video_sample_stride=1 \ --video_sample_n_frames=81 \ --train_batch_size=1 \ --gradient_accumulation_steps=1 \ diff --git a/scripts/fantasytalking/train.sh b/scripts/fantasytalking/train.sh index 225a556..b7eac02 100644 --- a/scripts/fantasytalking/train.sh +++ b/scripts/fantasytalking/train.sh @@ -13,7 +13,7 @@ accelerate launch --mixed_precision="bf16" scripts/fantasytalking/train.py \ --train_data_meta=$DATASET_META_NAME \ --video_sample_size=512 \ --token_sample_size=512 \ - --video_sample_stride=2 \ + --video_sample_stride=1 \ --video_sample_n_frames=81 \ --train_batch_size=1 \ --gradient_accumulation_steps=1 \