From 78275b04806c56de6fc3b62cec70b28a0fa67703 Mon Sep 17 00:00:00 2001
From: Yuxuan Zhang <2448370773@qq.com>
Date: Mon, 13 Jan 2025 20:02:06 +0800
Subject: [PATCH] add comment of bash scripts

---
 finetune/train_ddp_i2v.sh  | 10 +++++-----
 finetune/train_ddp_t2v.sh  | 10 +++++-----
 finetune/train_zero_i2v.sh | 12 ++++++------
 finetune/train_zero_t2v.sh | 12 ++++++------
 4 files changed, 22 insertions(+), 22 deletions(-)

diff --git a/finetune/train_ddp_i2v.sh b/finetune/train_ddp_i2v.sh
index ecd322b..d05661f 100644
--- a/finetune/train_ddp_i2v.sh
+++ b/finetune/train_ddp_i2v.sh
@@ -28,11 +28,11 @@ DATA_ARGS=(
 
 # Training Configuration
 TRAIN_ARGS=(
-    --train_epochs 10
+    --train_epochs 10 # number of training epochs
+    --seed 42 # random seed
     --batch_size 1
     --gradient_accumulation_steps 1
-    --mixed_precision "bf16"  # ["no", "fp16"]
-    --seed 42
+    --mixed_precision "bf16"  # ["no", "fp16"] # Only CogVideoX-2B supports fp16 training
 )
 
 # System Configuration
@@ -44,8 +44,8 @@ SYSTEM_ARGS=(
 
 # Checkpointing Configuration
 CHECKPOINT_ARGS=(
-    --checkpointing_steps 5
-    --checkpointing_limit 10
+    --checkpointing_steps 10 # save checkpoint every x steps
+    --checkpointing_limit 2 # maximum number of checkpoints to keep, after which the oldest one is deleted
     --resume_from_checkpoint "/absolute/path/to/checkpoint_dir"  # if you want to resume from a checkpoint, otherwise, comment this line
 )
 
diff --git a/finetune/train_ddp_t2v.sh b/finetune/train_ddp_t2v.sh
index 4340d56..994a440 100644
--- a/finetune/train_ddp_t2v.sh
+++ b/finetune/train_ddp_t2v.sh
@@ -27,11 +27,11 @@ DATA_ARGS=(
 
 # Training Configuration
 TRAIN_ARGS=(
-    --train_epochs 10
+    --train_epochs 10 # number of training epochs
+    --seed 42 # random seed
     --batch_size 1
     --gradient_accumulation_steps 1
-    --mixed_precision "bf16"  # ["no", "fp16"]
-    --seed 42
+    --mixed_precision "bf16"  # ["no", "fp16"] # Only CogVideoX-2B supports fp16 training
 )
 
 # System Configuration
@@ -43,8 +43,8 @@ SYSTEM_ARGS=(
 
 # Checkpointing Configuration
 CHECKPOINT_ARGS=(
-    --checkpointing_steps 5
-    --checkpointing_limit 10
+    --checkpointing_steps 10 # save checkpoint every x steps
+    --checkpointing_limit 2 # maximum number of checkpoints to keep, after which the oldest one is deleted
     --resume_from_checkpoint "/absolute/path/to/checkpoint_dir"  # if you want to resume from a checkpoint, otherwise, comment this line
 )
 
diff --git a/finetune/train_zero_i2v.sh b/finetune/train_zero_i2v.sh
index 2357a7e..03f2c93 100644
--- a/finetune/train_zero_i2v.sh
+++ b/finetune/train_zero_i2v.sh
@@ -23,18 +23,18 @@ DATA_ARGS=(
     --caption_column "prompt.txt"
     --video_column "videos.txt"
     # --image_column "images.txt"  # comment this line will use first frame of video as image conditioning
-    --train_resolution "81x768x1360"  # (frames x height x width), frames should be 8N+1
+    --train_resolution "81x768x1360"  # (frames x height x width), frames should be 8N+1 and height, width should be multiples of 16
 )
 
 # Training Configuration
 TRAIN_ARGS=(
-    --train_epochs 10
-    --seed 42
+    --train_epochs 10 # number of training epochs
+    --seed 42 # random seed
 
     #########   Please keep consistent with deepspeed config file ##########
     --batch_size 1
     --gradient_accumulation_steps 1
-    --mixed_precision "bf16"  # ["no", "fp16"]
+    --mixed_precision "bf16"  # ["no", "fp16"] Only CogVideoX-2B supports fp16 training
     ########################################################################
 )
 
@@ -47,8 +47,8 @@ SYSTEM_ARGS=(
 
 # Checkpointing Configuration
 CHECKPOINT_ARGS=(
-    --checkpointing_steps 10
-    --checkpointing_limit 2
+    --checkpointing_steps 10 # save checkpoint every x steps
+    --checkpointing_limit 2 # maximum number of checkpoints to keep, after which the oldest one is deleted
     # --resume_from_checkpoint "/absolute/path/to/checkpoint_dir"  # if you want to resume from a checkpoint, otherwise, comment this line
 )
 
diff --git a/finetune/train_zero_t2v.sh b/finetune/train_zero_t2v.sh
index 80dbca3..265b08d 100644
--- a/finetune/train_zero_t2v.sh
+++ b/finetune/train_zero_t2v.sh
@@ -22,18 +22,18 @@ DATA_ARGS=(
     --data_root "/absolute/path/to/your/data_root"
     --caption_column "prompt.txt"
     --video_column "videos.txt"
-    --train_resolution "81x768x1360"  # (frames x height x width), frames should be 8N+1
+    --train_resolution "81x768x1360"  # (frames x height x width), frames should be 8N+1 and height, width should be multiples of 16
 )
 
 # Training Configuration
 TRAIN_ARGS=(
-    --train_epochs 10
-    --seed 42
+    --train_epochs 10 # number of training epochs
+    --seed 42 # random seed
 
     #########   Please keep consistent with deepspeed config file ##########
     --batch_size 1
     --gradient_accumulation_steps 1
-    --mixed_precision "bf16"  # ["no", "fp16"]
+    --mixed_precision "bf16"  # ["no", "fp16"] Only CogVideoX-2B supports fp16 training
     ########################################################################
 )
 
@@ -46,8 +46,8 @@ SYSTEM_ARGS=(
 
 # Checkpointing Configuration
 CHECKPOINT_ARGS=(
-    --checkpointing_steps 10
-    --checkpointing_limit 2
+    --checkpointing_steps 10 # save checkpoint every x steps
+    --checkpointing_limit 2 # maximum number of checkpoints to keep, after which the oldest one is deleted
     # --resume_from_checkpoint "/absolute/path/to/checkpoint_dir"  # if you want to resume from a checkpoint, otherwise, comment this line
 )