aigc-apps-VideoX-Fun/scripts/ltx2/README_TRAIN.md

LTX-2 Full Parameter Training Guide

This document provides a complete workflow for full parameter training of LTX-2 (Lightricks Text-to-Video), including environment configuration, data preparation, distributed training, and inference testing.

Note

: LTX-2 is an audio-visual generative video model that can simultaneously generate video and corresponding audio. The training data requires both video and audio files.

---

Table of Contents

• 1. Environment Configuration
• 2. Data Preparation
  • 2.1 Quick Test Dataset
  • 2.2 Dataset Structure
  • 2.3 metadata.json Format
  • 2.4 Relative vs Absolute Path Usage
• 3. Full Parameter Training
  • 3.1 Download Pretrained Model
  • 3.2 Quick Start (DeepSpeed-Zero-2)
  • 3.3 Common Training Parameters
  • 3.4 Training Validation
  • 3.5 Training with FSDP
  • 3.6 Training Without DeepSpeed or FSDP
  • 3.7 Multi-Machine Distributed Training
• 4. Inference Testing
  • 4.1 Inference Parameters
  • 4.2 Single GPU Inference
  • 4.3 Multi-GPU Parallel Inference
• 5. Additional Resources

---

1. Environment Configuration

Method 1: Using requirements.txt

pip install -r requirements.txt

Method 2: Manual Dependency Installation

pip install Pillow einops safetensors timm tomesd librosa "torch>=2.1.2" torchdiffeq torchsde decord datasets numpy scikit-image
pip install omegaconf SentencePiece imageio[ffmpeg] imageio[pyav] tensorboard beautifulsoup4 ftfy func_timeout onnxruntime
pip install "peft>=0.17.0" "accelerate>=0.25.0" "gradio>=3.41.2" "diffusers>=0.30.1" "transformers>=4.46.2"
pip install yunchang xfuser modelscope openpyxl
pip uninstall opencv-python opencv-contrib-python opencv-python-headless -y
pip install opencv-python-headless
pip install deepspeed==0.17.0 numpy==1.26.4

Method 3: Using Docker

When using Docker, please ensure that the GPU driver and CUDA environment are correctly installed on your machine, then execute the following commands:

# pull image
docker pull mybigpai-public-registry.cn-beijing.cr.aliyuncs.com/easycv/torch_cuda:cogvideox_fun

# enter image
docker run -it -p 7860:7860 --network host --gpus all --security-opt seccomp:unconfined --shm-size 200g mybigpai-public-registry.cn-beijing.cr.aliyuncs.com/easycv/torch_cuda:cogvideox_fun

---

2. Data Preparation

2.1 Quick Test Dataset

We provide a test dataset containing several audio-video training samples.

# Download official demo dataset
modelscope download --dataset PAI/X-Fun-Videos-Audios-Demo --local_dir ./datasets/X-Fun-Videos-Audios-Demo

2.2 Dataset Structure

📦 datasets/
├── 📂 my_dataset/
│   ├── 📂 train/
│   │   ├── 📄 video001.mp4
│   │   ├── 📄 video002.mp4
│   │   └── 📄 ...
│   ├── 📂 wav/
│   │   ├── 📄 audio001.wav
│   │   ├── 📄 audio002.wav
│   │   └── 📄 ...
│   └── 📄 metadata.json

2.3 metadata.json Format

⚠️ Important: LTX-2 is an audio-visual generative model. Unlike regular video training, you must provide the audio_path field in metadata.json.

Relative Path Format (example):

[
  {
    "file_path": "train/video001.mp4",
    "audio_path": "wav/audio001.wav",
    "text": "A brown dog barks on a sofa, sitting on a light-colored couch in a cozy room",
    "type": "video",
    "width": 768,
    "height": 512
  },
  {
    "file_path": "train/video002.mp4",
    "audio_path": "wav/audio002.wav",
    "text": "A group of young men in suits and sunglasses are walking down a city street",
    "type": "video",
    "width": 640,
    "height": 640
  }
]

Absolute Path Format:

[
  {
    "file_path": "/mnt/data/videos/dog.mp4",
    "audio_path": "/mnt/data/wavs/dog.wav",
    "text": "A brown dog barks on a sofa",
    "type": "video",
    "width": 768,
    "height": 512
  }
]

Key Fields Description:

• file_path: Video file path (relative or absolute)
• audio_path: Audio file path (LTX-2 specific and required, main difference from regular video training)
  • Audio files are typically in .wav format
  • Path should correspond to file_path, e.g., train/video001.mp4 corresponds to wav/audio001.wav
• text: Video description (English prompt)
• type: Data type, fixed as "video"
• width / height: Video dimensions (recommended to provide for bucket training; if not provided, they will be automatically read during training, which may slow down training when data is stored on slow systems like OSS)
  • You can use scripts/process_json_add_width_and_height.py to add width and height fields to JSON files without these fields, supporting both images and videos
  • Usage: python scripts/process_json_add_width_and_height.py --input_file datasets/X-Fun-Videos-Audios-Demo/metadata.json --output_file datasets/X-Fun-Videos-Audios-Demo/metadata_add_width_height.json

Dataset Comparison: LTX-2 vs Regular Video Training:

Model Type	Required Fields	Audio Field
Regular Video (WAN, CogVideoX, etc.)	file_path, text, type	❌ Not needed
LTX-2 (Audio-Visual Generation)	file_path, audio_path, text, type	✅ Required

2.4 Relative vs Absolute Path Usage

Relative Paths:

If your data uses relative paths, configure the training script as follows:

export DATASET_NAME="datasets/X-Fun-Videos-Audios-Demo/"
export DATASET_META_NAME="datasets/X-Fun-Videos-Audios-Demo/metadata_add_width_height.json"

Absolute Paths:

If your data uses absolute paths, configure the training script as follows:

export DATASET_NAME=""
export DATASET_META_NAME="/mnt/data/metadata_add_width_height.json"

💡 Recommendation: If the dataset is small and stored locally, use relative paths. If the dataset is stored on external storage (e.g., NAS, OSS) or shared across multiple machines, use absolute paths.

---

3. Full Parameter Training

3.1 Download Pretrained Model

# Create model directory
mkdir -p models/Diffusion_Transformer

# Download LTX-2 official weights (ModelScope)
modelscope download --model Lightricks/LTX-2 --local_dir models/Diffusion_Transformer/LTX-2

# Or use Hugging Face
# hf download Lightricks/LTX-2 --local-dir models/Diffusion_Transformer/LTX-2

3.2 Quick Start (DeepSpeed-Zero-2)

If you have downloaded the data as per 2.1 Quick Test Dataset and the weights as per 3.1 Download Pretrained Model, you can directly copy and run the quick start command.

DeepSpeed-Zero-2 and FSDP are recommended for training. Here we use DeepSpeed-Zero-2 as an example.

The difference between DeepSpeed-Zero-2 and FSDP lies in whether the model weights are sharded. If VRAM is insufficient when using multiple GPUs with DeepSpeed-Zero-2, you can switch to FSDP.

export MODEL_NAME="models/Diffusion_Transformer/LTX-2"
export DATASET_NAME="datasets/X-Fun-Videos-Audios-Demo/"
export DATASET_META_NAME="datasets/X-Fun-Videos-Audios-Demo/metadata_add_width_height.json"
# NCCL_IB_DISABLE=1 and NCCL_P2P_DISABLE=1 are used in multi nodes without RDMA. 
# export NCCL_IB_DISABLE=1
# export NCCL_P2P_DISABLE=1
NCCL_DEBUG=INFO

accelerate launch --use_deepspeed --deepspeed_config_file config/zero_stage2_config.json --deepspeed_multinode_launcher standard scripts/ltx2/train.py \
  --pretrained_model_name_or_path=$MODEL_NAME \
  --train_data_dir=$DATASET_NAME \
  --train_data_meta=$DATASET_META_NAME \
  --image_sample_size=640 \
  --video_sample_size=640 \
  --token_sample_size=640 \
  --video_sample_stride=1 \
  --video_sample_n_frames=121 \
  --train_batch_size=1 \
  --video_repeat=1 \
  --gradient_accumulation_steps=1 \
  --dataloader_num_workers=8 \
  --num_train_epochs=100 \
  --checkpointing_steps=50 \
  --learning_rate=2e-05 \
  --lr_scheduler="constant_with_warmup" \
  --lr_warmup_steps=100 \
  --seed=42 \
  --output_dir="output_dir_ltx2" \
  --gradient_checkpointing \
  --mixed_precision="bf16" \
  --adam_weight_decay=3e-2 \
  --adam_epsilon=1e-10 \
  --vae_mini_batch=1 \
  --max_grad_norm=0.05 \
  --random_hw_adapt \
  --training_with_video_token_length \
  --enable_bucket \
  --uniform_sampling \
  --low_vram \
  --trainable_modules "."

3.3 Common Training Parameters

Key Parameter Descriptions:

Parameter	Description	Example Value
--pretrained_model_name_or_path	Path to pretrained model	models/Diffusion_Transformer/LTX-2
--train_data_dir	Training data directory	datasets/X-Fun-Videos-Audios-Demo/
--train_data_meta	Training data metadata file	datasets/X-Fun-Videos-Audios-Demo/metadata_add_width_height.json
--train_batch_size	Samples per batch	1
--image_sample_size	Maximum training resolution, auto bucketing	640
--video_sample_size	Maximum video resolution for training	640
--token_sample_size	Token length sampling size	640
--video_sample_stride	Frame sampling stride	1
--video_sample_n_frames	Number of frames to sample	81
--video_repeat	Number of times each video is repeated per epoch	1
--gradient_accumulation_steps	Gradient accumulation steps (equivalent to larger batch)	1
--dataloader_num_workers	DataLoader subprocesses	8
--num_train_epochs	Number of training epochs	100
--checkpointing_steps	Save checkpoint every N steps	50
--learning_rate	Initial learning rate	2e-05
--lr_scheduler	Learning rate scheduler	constant_with_warmup
--lr_warmup_steps	Learning rate warmup steps	100
--seed	Random seed	42
--output_dir	Output directory	output_dir_ltx2
--gradient_checkpointing	Enable activation checkpointing	-
--mixed_precision	Mixed precision: fp16/bf16	bf16
--adam_weight_decay	AdamW weight decay	3e-2
--adam_epsilon	AdamW epsilon value	1e-10
--vae_mini_batch	Mini-batch size for VAE encoding	1
--max_grad_norm	Gradient clipping threshold	0.05
--random_hw_adapt	Auto-scale videos to random size in range [512, video_sample_size]	-
--training_with_video_token_length	Train based on token length instead of fixed resolution	-
--enable_bucket	Enable bucket training: trains entire videos grouped by resolution without center cropping	-
--uniform_sampling	Uniform timestep sampling	-
--low_vram	Enable low VRAM optimizations	-
--resume_from_checkpoint	Resume training from checkpoint path, use "latest" to auto-select latest	None
--trainable_modules	Trainable modules ("." means all modules)	"."
--validation_steps	Execute validation every N steps	100
--validation_epochs	Execute validation every N epochs	500
--validation_prompts	Prompts used during validation	"A man in a blue blazer..."

3.4 Training Validation

You can configure validation parameters to periodically generate test videos during training, allowing you to monitor training progress and model quality.

Validation Parameters:

Parameter	Description	Recommended Value
--validation_steps	Execute validation every N steps	100
--validation_epochs	Execute validation every N epochs	500
--validation_prompts	Prompt for validation video generation. Use multiple space-separated prompt strings	Space-separated prompt strings

Example:

  --validation_steps=100 \
  --validation_epochs=500 \
  --validation_prompts="A man in a blue blazer and glasses speaks in a formal indoor setting, framed by wooden furniture and a filled bookshelf. Quiet room acoustics underscore his measured tone as he delivers his remarks. At one point, he says, \"Hi.\""

Notes:

• Validation videos will be saved to the output_dir directory
• For multi-prompt validation, use: --validation_prompts "prompt1" "prompt2" "prompt3"

3.5 Training with FSDP

If VRAM is insufficient when using multiple GPUs with DeepSpeed-Zero-2, you can switch to FSDP.

export MODEL_NAME="models/Diffusion_Transformer/LTX-2"
export DATASET_NAME="datasets/X-Fun-Videos-Audios-Demo/"
export DATASET_META_NAME="datasets/X-Fun-Videos-Audios-Demo/metadata_add_width_height.json"
# NCCL_IB_DISABLE=1 and NCCL_P2P_DISABLE=1 are used in multi nodes without RDMA. 
# export NCCL_IB_DISABLE=1
# export NCCL_P2P_DISABLE=1
NCCL_DEBUG=INFO

accelerate launch --mixed_precision="bf16" --use_fsdp --fsdp_auto_wrap_policy TRANSFORMER_BASED_WRAP --fsdp_transformer_layer_cls_to_wrap LTX2VideoTransformerBlock --fsdp_sharding_strategy "FULL_SHARD" --fsdp_state_dict_type=SHARDED_STATE_DICT --fsdp_backward_prefetch "BACKWARD_PRE" --fsdp_cpu_ram_efficient_loading False scripts/ltx2/train.py \
  --pretrained_model_name_or_path=$MODEL_NAME \
  --train_data_dir=$DATASET_NAME \
  --train_data_meta=$DATASET_META_NAME \
  --image_sample_size=640 \
  --video_sample_size=640 \
  --token_sample_size=640 \
  --video_sample_stride=1 \
  --video_sample_n_frames=121 \
  --train_batch_size=1 \
  --video_repeat=1 \
  --gradient_accumulation_steps=1 \
  --dataloader_num_workers=8 \
  --num_train_epochs=100 \
  --checkpointing_steps=50 \
  --learning_rate=2e-05 \
  --lr_scheduler="constant_with_warmup" \
  --lr_warmup_steps=100 \
  --seed=42 \
  --output_dir="output_dir_ltx2" \
  --gradient_checkpointing \
  --mixed_precision="bf16" \
  --adam_weight_decay=3e-2 \
  --adam_epsilon=1e-10 \
  --vae_mini_batch=1 \
  --max_grad_norm=0.05 \
  --random_hw_adapt \
  --training_with_video_token_length \
  --enable_bucket \
  --uniform_sampling \
  --low_vram \
  --trainable_modules "."

3.6 Training Without DeepSpeed or FSDP

This approach is not recommended as it lacks VRAM-saving backends and may easily cause out-of-memory errors. This is provided for reference only.

export MODEL_NAME="models/Diffusion_Transformer/LTX-2"
export DATASET_NAME="datasets/X-Fun-Videos-Audios-Demo/"
export DATASET_META_NAME="datasets/X-Fun-Videos-Audios-Demo/metadata_add_width_height.json"
# NCCL_IB_DISABLE=1 and NCCL_P2P_DISABLE=1 are used in multi nodes without RDMA. 
# export NCCL_IB_DISABLE=1
# export NCCL_P2P_DISABLE=1
NCCL_DEBUG=INFO

accelerate launch --mixed_precision="bf16" scripts/ltx2/train.py \
  --pretrained_model_name_or_path=$MODEL_NAME \
  --train_data_dir=$DATASET_NAME \
  --train_data_meta=$DATASET_META_NAME \
  --image_sample_size=640 \
  --video_sample_size=640 \
  --token_sample_size=640 \
  --video_sample_stride=1 \
  --video_sample_n_frames=121 \
  --train_batch_size=1 \
  --video_repeat=1 \
  --gradient_accumulation_steps=1 \
  --dataloader_num_workers=8 \
  --num_train_epochs=100 \
  --checkpointing_steps=50 \
  --learning_rate=2e-05 \
  --lr_scheduler="constant_with_warmup" \
  --lr_warmup_steps=100 \
  --seed=42 \
  --output_dir="output_dir_ltx2" \
  --gradient_checkpointing \
  --mixed_precision="bf16" \
  --adam_weight_decay=3e-2 \
  --adam_epsilon=1e-10 \
  --vae_mini_batch=1 \
  --max_grad_norm=0.05 \
  --random_hw_adapt \
  --training_with_video_token_length \
  --enable_bucket \
  --uniform_sampling \
  --low_vram \
  --trainable_modules "."

3.7 Multi-Machine Distributed Training

Suitable for: Ultra-large-scale datasets, faster training speed

3.7.1 Environment Configuration

Assuming 2 machines with 8 GPUs each:

Machine 0 (Master):

export MODEL_NAME="models/Diffusion_Transformer/LTX-2"
export DATASET_NAME="datasets/X-Fun-Videos-Audios-Demo/"
export DATASET_META_NAME="datasets/X-Fun-Videos-Audios-Demo/metadata_add_width_height.json"
export MASTER_ADDR="192.168.1.100"  # Master machine IP
export MASTER_PORT=10086
export WORLD_SIZE=2                  # Total number of machines
export NUM_PROCESS=16                # Total processes = machines × 8
export RANK=0                        # Current machine rank (0 or 1)
# NCCL_IB_DISABLE=1 and NCCL_P2P_DISABLE=1 are used in multi nodes without RDMA. 
# export NCCL_IB_DISABLE=1
# export NCCL_P2P_DISABLE=1
NCCL_DEBUG=INFO

accelerate launch --mixed_precision="bf16" --main_process_ip=$MASTER_ADDR --main_process_port=$MASTER_PORT --num_machines=$WORLD_SIZE --num_processes=$NUM_PROCESS --machine_rank=$RANK --use_deepspeed --deepspeed_config_file config/zero_stage2_config.json --deepspeed_multinode_launcher standard scripts/ltx2/train.py \
  --pretrained_model_name_or_path=$MODEL_NAME \
  --train_data_dir=$DATASET_NAME \
  --train_data_meta=$DATASET_META_NAME \
  --image_sample_size=640 \
  --video_sample_size=640 \
  --token_sample_size=640 \
  --video_sample_stride=1 \
  --video_sample_n_frames=121 \
  --train_batch_size=1 \
  --video_repeat=1 \
  --gradient_accumulation_steps=1 \
  --dataloader_num_workers=8 \
  --num_train_epochs=100 \
  --checkpointing_steps=50 \
  --learning_rate=2e-05 \
  --lr_scheduler="constant_with_warmup" \
  --lr_warmup_steps=100 \
  --seed=42 \
  --output_dir="output_dir_ltx2" \
  --gradient_checkpointing \
  --mixed_precision="bf16" \
  --adam_weight_decay=3e-2 \
  --adam_epsilon=1e-10 \
  --vae_mini_batch=1 \
  --max_grad_norm=0.05 \
  --random_hw_adapt \
  --training_with_video_token_length \
  --enable_bucket \
  --uniform_sampling \
  --low_vram \
  --trainable_modules "."

Machine 1 (Worker):

export MODEL_NAME="models/Diffusion_Transformer/LTX-2"
export DATASET_NAME="datasets/X-Fun-Videos-Audios-Demo/"
export DATASET_META_NAME="datasets/X-Fun-Videos-Audios-Demo/metadata_add_width_height.json"
export MASTER_ADDR="192.168.1.100"  # Same as Master
export MASTER_PORT=10086
export WORLD_SIZE=2
export NUM_PROCESS=16
export RANK=1  # Note this is 1
# NCCL_IB_DISABLE=1 and NCCL_P2P_DISABLE=1 are used in multi nodes without RDMA. 
# export NCCL_IB_DISABLE=1
# export NCCL_P2P_DISABLE=1
NCCL_DEBUG=INFO

# Use the same accelerate launch command as Machine 0

3.7.2 Multi-Machine Training Notes

• Network Requirements:

  • RDMA/InfiniBand recommended (high performance)
  • Without RDMA, add environment variables:

    export NCCL_IB_DISABLE=1
    export NCCL_P2P_DISABLE=1
    

• Data Synchronization: All machines must be able to access the same data paths (NFS/shared storage)

4. Inference Testing

4.1 Inference Parameters

Key Parameter Descriptions:

Parameter	Description	Example Value
GPU_memory_mode	GPU memory mode, see table below for options	sequential_cpu_offload
ulysses_degree	Head dimension parallelization degree, 1 for single GPU	1
ring_degree	Sequence dimension parallelization degree, 1 for single GPU	1
fsdp_dit	Use FSDP for Transformer in multi-GPU inference to save VRAM	False
fsdp_text_encoder	Use FSDP for text encoder in multi-GPU inference	False
compile_dit	Compile Transformer to accelerate inference (effective at fixed resolution)	False
model_name	Model path	models/Diffusion_Transformer/LTX-2
sampler_name	Sampler type: Flow, Flow_Unipc, Flow_DPM++	Flow
transformer_path	Path to trained Transformer weights	None
vae_path	Path to trained VAE weights	None
lora_path	LoRA weights path	None
sample_size	Generated video resolution [height, width]	[512, 768]
video_length	Number of frames to generate	121
fps	Frames per second	24
weight_dtype	Model weight precision, use torch.float16 for GPUs without bf16 support	torch.bfloat16
prompt	Positive prompt describing the content to generate	"A brown dog barks..."
negative_prompt	Negative prompt for content to avoid	"worst quality, inconsistent motion..."
guidance_scale	Guidance strength	6.0
seed	Random seed for reproducibility	43
num_inference_steps	Inference steps	50
lora_weight	LoRA weight strength	0.55
save_path	Generated video save path	samples/ltx2-videos-t2v
audio_sample_rate	Audio sample rate (read from vocoder config)	24000

GPU Memory Mode Description:

Mode	Description	VRAM Usage
model_full_load	Load entire model to GPU	Highest
model_full_load_and_qfloat8	Full load + FP8 quantization	High
model_cpu_offload	Offload model to CPU after use	Medium
model_cpu_offload_and_qfloat8	CPU offload + FP8 quantization	Medium-Low
model_group_offload	Layer group offload between CPU/CUDA	Low
sequential_cpu_offload	Offload each layer individually (slowest)	Lowest

4.2 Single GPU Inference

Run single GPU inference with:

python examples/ltx2/predict_t2v.py

Edit examples/ltx2/predict_t2v.py according to your needs. For first-time inference, focus on these parameters. For other parameters, see the Inference Parameters section above.

# Choose based on your GPU VRAM
GPU_memory_mode = "sequential_cpu_offload"
# Your actual model path
model_name = "models/Diffusion_Transformer/LTX-2"  
# Trained weights path, e.g. "output_dir_ltx2/checkpoint-xxx/diffusion_pytorch_model.safetensors"
transformer_path = None  
# Write based on content to generate
prompt = "A brown dog barks on a sofa, sitting on a light-colored couch in a cozy room"  
# ...

4.3 Multi-GPU Parallel Inference

Suitable for: High-resolution generation, accelerated inference

Install Parallel Inference Dependencies

pip install xfuser==0.4.2 yunchang==0.6.2

Configure Parallel Strategy

Edit examples/ltx2/predict_t2v.py:

# Ensure ulysses_degree × ring_degree = number of GPUs
# For example, using 2 GPUs:
ulysses_degree = 2  # Head dimension parallelization
ring_degree = 1     # Sequence dimension parallelization

Configuration Principles:

• ulysses_degree must evenly divide the model's number of heads
• ring_degree splits on sequence dimension, affecting communication overhead; avoid using it when heads can be divided

Example Configurations:

GPU Count	ulysses_degree	ring_degree	Description
1	1	1	Single GPU
4	4	1	Head parallelization
8	2	4	Hybrid parallelization
8	8	1	Head parallelization

Run Multi-GPU Inference

torchrun --nproc-per-node=2 examples/ltx2/predict_t2v.py

5. Additional Resources

• Official GitHub: https://github.com/aigc-apps/VideoX-Fun