Files

T

Bubbliiiing 403f1f7b78 Fix Bug in Self Forcing in Multi-Gpus Infernece && Update Training Code and Docs && Update Reward Models (#498 )

2026-07-14 10:30:54 +08:00

20 KiB

Raw Blame History

FLUX.2 LoRA Fine-Tuning Training Guide

This document provides a complete workflow for FLUX.2 LoRA fine-tuning training, including environment configuration, data preparation, multiple distributed training strategies, and inference testing.

1. Environment Configuration
2. Data Preparation
3. LoRA Training
4. Inference Testing
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 training samples.

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

2.2 Dataset Structure

📦 datasets/
├── 📂 my_dataset/
│   ├── 📂 train/
│   │   ├── 📄 image001.jpg
│   │   ├── 📄 image002.png
│   │   └── 📄 ...
│   └── 📄 metadata.json

2.3 metadata.json Format

Relative Path Format (example):

[
  {
    "file_path": "train/image001.jpg",
    "text": "A beautiful sunset over the ocean, golden hour lighting",
    "width": 1024,
    "height": 1024
  },
  {
    "file_path": "train/image002.png",
    "text": "Portrait of a young woman, studio lighting, high quality",
    "width": 1328,
    "height": 1328
  }
]

Absolute Path Format:

[
  {
    "file_path": "/mnt/data/images/sunset.jpg",
    "text": "A beautiful sunset over the ocean",
    "width": 1024,
    "height": 1024
  }
]

Key Fields Description:

file_path: Image path (relative or absolute)
text: Image description (English prompt)
width / height: Image 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-Images-Demo/metadata.json --output_file datasets/X-Fun-Images-Demo/metadata_add_width_height.json

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-Images-Demo/"
export DATASET_META_NAME="datasets/X-Fun-Images-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. LoRA Training

3.1 Download Pretrained Model

# Create model directory
mkdir -p models/Diffusion_Transformer

# Download FLUX.2 official weights
modelscope download --model black-forest-labs/FLUX.2-dev --local_dir models/Diffusion_Transformer/FLUX.2-dev

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/FLUX.2-dev"
export DATASET_NAME="datasets/X-Fun-Images-Demo/"
export DATASET_META_NAME="datasets/X-Fun-Images-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/flux2/train_lora.py \
  --pretrained_model_name_or_path=$MODEL_NAME \
  --train_data_dir=$DATASET_NAME \
  --train_data_meta=$DATASET_META_NAME \
  --train_batch_size=1 \
  --image_sample_size=1328 \
  --gradient_accumulation_steps=1 \
  --dataloader_num_workers=8 \
  --num_train_epochs=100 \
  --checkpointing_steps=50 \
  --learning_rate=1e-04 \
  --seed=42 \
  --output_dir="output_dir_flux2_lora" \
  --gradient_checkpointing \
  --mixed_precision="bf16" \
  --adam_weight_decay=3e-2 \
  --adam_epsilon=1e-10 \
  --vae_mini_batch=1 \
  --max_grad_norm=0.05 \
  --enable_bucket \
  --rank=64 \
  --network_alpha=32 \
  --target_name="to_q,to_k,to_v,ff.0,ff.2,ff_context.0,ff_context.2" \
  --use_peft_lora \
  --uniform_sampling

3.3 LoRA-Specific Parameters

LoRA Key Parameters Description:

Parameter	Description	Example Value
`--pretrained_model_name_or_path`	Pretrained model path	`models/Diffusion_Transformer/FLUX.2-dev`
`--train_data_dir`	Training data directory	`datasets/X-Fun-Images-Demo/`
`--train_data_meta`	Training data metadata file	`datasets/X-Fun-Images-Demo/metadata_add_width_height.json`
`--train_batch_size`	Batch size per device	1
`--image_sample_size`	Maximum training resolution (auto bucketing)	1328
`--gradient_accumulation_steps`	Gradient accumulation steps (effective batch size increase)	1
`--dataloader_num_workers`	DataLoader subprocess count	8
`--num_train_epochs`	Number of training epochs	100
`--checkpointing_steps`	Save checkpoint every N steps	50
`--learning_rate`	Initial learning rate (recommended for LoRA)	1e-04
`--lr_warmup_steps`	Learning rate warmup steps	100
`--seed`	Random seed (reproducible training)	42
`--output_dir`	Output directory	`output_dir_flux2_lora`
`--gradient_checkpointing`	Enable gradient checkpointing	-
`--mixed_precision`	Mixed precision: `fp16/bf16`	`bf16`
`--enable_bucket`	Enable bucket training (no center crop, train full images grouped by resolution)	-
`--uniform_sampling`	Uniform timestep sampling (recommended)	-
`--resume_from_checkpoint`	Resume training path, use `"latest"` to auto-select latest checkpoint	None
`--rank`	LoRA update matrix dimension (higher rank = more expressive but more VRAM)	64
`--network_alpha`	LoRA update matrix scaling coefficient (typically half of rank or same)	32
`--target_name`	Components/modules to apply LoRA, comma-separated	`to_q,to_k,to_v,ff.0,ff.2,ff_context.0,ff_context.2`
`--use_peft_lora`	Use PEFT module to add LoRA (more memory efficient)	-
`--validation_steps`	Execute validation every N steps	100
`--validation_epochs`	Execute validation every N epochs	100
`--validation_prompts`	Prompts used during validation	`"1girl, black_hair, ..."`

3.4 Training Validation

You can configure validation parameters to periodically generate test images 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	100
`--validation_prompts`	Prompt for validation image generation. Use multiple space-separated prompt strings	Space-separated prompt strings

Example:

  --validation_steps=100 \
  --validation_epochs=100 \
  --validation_prompts="1girl, black_hair, brown_eyes, earrings, freckles, grey_background, jewelry, lips, long_hair, looking_at_viewer, nose, piercing, realistic, red_lips, solo, upper_body"

Notes:

Validation images 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.

✅ Recommended: FSDP has been thoroughly tested in this repository with fewer errors and more stability.

export MODEL_NAME="models/Diffusion_Transformer/FLUX.2-dev"
export DATASET_NAME="datasets/X-Fun-Images-Demo/"
export DATASET_META_NAME="datasets/X-Fun-Images-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 Flux2SingleTransformerBlock,Flux2TransformerBlock --fsdp_sharding_strategy "FULL_SHARD" --fsdp_state_dict_type=SHARDED_STATE_DICT --fsdp_backward_prefetch "BACKWARD_PRE" --fsdp_cpu_ram_efficient_loading False scripts/flux2/train_lora.py \
  --pretrained_model_name_or_path=$MODEL_NAME \
  --train_data_dir=$DATASET_NAME \
  --train_data_meta=$DATASET_META_NAME \
  --train_batch_size=1 \
  --image_sample_size=1328 \
  --gradient_accumulation_steps=1 \
  --dataloader_num_workers=8 \
  --num_train_epochs=100 \
  --checkpointing_steps=50 \
  --learning_rate=1e-04 \
  --seed=42 \
  --output_dir="output_dir_flux2_lora" \
  --gradient_checkpointing \
  --mixed_precision="bf16" \
  --adam_weight_decay=3e-2 \
  --adam_epsilon=1e-10 \
  --vae_mini_batch=1 \
  --max_grad_norm=0.05 \
  --enable_bucket \
  --rank=64 \
  --network_alpha=32 \
  --target_name="to_q,to_k,to_v,ff.0,ff.2,ff_context.0,ff_context.2" \
  --use_peft_lora \
  --uniform_sampling

3.6 Training without DeepSpeed or FSDP

This approach is not recommended as there is no memory-saving backend, which may cause insufficient VRAM. This is provided for reference only.

export MODEL_NAME="models/Diffusion_Transformer/FLUX.2-dev"
export DATASET_NAME="datasets/X-Fun-Images-Demo/"
export DATASET_META_NAME="datasets/X-Fun-Images-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/flux2/train_lora.py \
  --pretrained_model_name_or_path=$MODEL_NAME \
  --train_data_dir=$DATASET_NAME \
  --train_data_meta=$DATASET_META_NAME \
  --train_batch_size=1 \
  --image_sample_size=1328 \
  --gradient_accumulation_steps=1 \
  --dataloader_num_workers=8 \
  --num_train_epochs=100 \
  --checkpointing_steps=50 \
  --learning_rate=1e-04 \
  --seed=42 \
  --output_dir="output_dir_flux2_lora" \
  --gradient_checkpointing \
  --mixed_precision="bf16" \
  --adam_weight_decay=3e-2 \
  --adam_epsilon=1e-10 \
  --vae_mini_batch=1 \
  --max_grad_norm=0.05 \
  --enable_bucket \
  --rank=64 \
  --network_alpha=32 \
  --target_name="to_q,to_k,to_v,ff.0,ff.2,ff_context.0,ff_context.2" \
  --use_peft_lora \
  --uniform_sampling

3.7 Multi-Machine Distributed Training

Suitable for: Large-scale datasets, faster training speed

3.7.1 Environment Configuration

Assume 2 machines, each with 8 GPUs:

Machine 0 (Master):

export MODEL_NAME="models/Diffusion_Transformer/FLUX.2-dev"
export DATASET_NAME="datasets/X-Fun-Images-Demo/"
export DATASET_META_NAME="datasets/X-Fun-Images-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=$RANK --use_deepspeed --deepspeed_config_file config/zero_stage2_config.json --deepspeed_multinode_launcher standard scripts/flux2/train_lora.py \
  --pretrained_model_name_or_path=$MODEL_NAME \
  --train_data_dir=$DATASET_NAME \
  --train_data_meta=$DATASET_META_NAME \
  --train_batch_size=1 \
  --image_sample_size=1328 \
  --gradient_accumulation_steps=1 \
  --dataloader_num_workers=8 \
  --num_train_epochs=100 \
  --checkpointing_steps=50 \
  --learning_rate=1e-04 \
  --seed=42 \
  --output_dir="output_dir_flux2_lora" \
  --gradient_checkpointing \
  --mixed_precision="bf16" \
  --adam_weight_decay=3e-2 \
  --adam_epsilon=1e-10 \
  --vae_mini_batch=1 \
  --max_grad_norm=0.05 \
  --enable_bucket \
  --rank=64 \
  --network_alpha=32 \
  --target_name="to_q,to_k,to_v,ff.0,ff.2,ff_context.0,ff_context.2" \
  --use_peft_lora \
  --uniform_sampling

Machine 1 (Worker):

export MODEL_NAME="models/Diffusion_Transformer/FLUX.2-dev"
export DATASET_NAME="datasets/X-Fun-Images-Demo/"
export DATASET_META_NAME="datasets/X-Fun-Images-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 have access to the same data paths (NFS/shared storage)

4. Inference Testing

4.1 Inference Parameter Parsing

Key Parameters Description:

Parameter	Description	Example Value
`GPU_memory_mode`	GPU memory management mode, see table below	`model_cpu_offload`
`ulysses_degree`	Head dimension parallelism degree, 1 for single GPU	1
`ring_degree`	Sequence dimension parallelism 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 for faster inference (effective at fixed resolution)	`False`
`model_name`	Model path	`models/Diffusion_Transformer/FLUX.2-dev`
`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 image resolution `[height, width]`	`[1344, 768]`
`weight_dtype`	Model weight precision, use `torch.float16` for GPUs without bf16 support	`torch.bfloat16`
`prompt`	Positive prompt describing the content	`"1girl, black_hair..."`
`negative_prompt`	Negative prompt for content to avoid	`" "`
`guidance_scale`	Guidance strength	4.0
`seed`	Random seed for reproducibility	43
`num_inference_steps`	Inference steps	50
`lora_weight`	LoRA weight strength	0.55
`save_path`	Generated image save path	`samples/flux2-t2i`

GPU Memory Management Modes:

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 groups switch between CPU/CUDA	Low
`sequential_cpu_offload`	Layer-by-layer offload (slowest)	Lowest

4.2 Single GPU Inference

Run single GPU inference with the following command:

python examples/flux2/predict_t2i.py

Edit examples/flux2/predict_t2i.py according to your needs. For first-time inference, focus on these parameters. For other parameters, see the inference parameter section above.

# Choose based on GPU VRAM
GPU_memory_mode = "sequential_cpu_offload"
# Based on actual model path
model_name = "models/Diffusion_Transformer/FLUX.2-dev"  
# LoRA weights path, e.g., "output_dir_flux2_lora/checkpoint-xxx/lora_weights.safetensors"
lora_path = None
# LoRA weight strength
lora_weight = 0.55
# Write based on content to generate
prompt = "1girl, black_hair, brown_eyes, earrings, freckles, grey_background, jewelry, lips, long_hair, looking_at_viewer, nose, piercing, realistic, red_lips, solo, upper_body"  
# ...

4.3 Multi-GPU Parallel Inference

Suitable for: High-resolution generation, faster inference

Install Parallel Inference Dependencies

pip install xfuser==0.4.2 yunchang==0.6.2

Configure Parallel Strategy

Edit examples/flux2/predict_t2i.py:

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

Configuration Principles:

ulysses_degree must divide the model's head count evenly.
ring_degree splits on sequence dimension, affecting communication overhead. Avoid using it when head count can be divided.

Example Configurations:

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

Run Multi-GPU Inference

torchrun --nproc-per-node=2 examples/flux2/predict_t2i.py

5. Additional Resources

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

20 KiB Raw Blame History Unescape Escape