max pixels argument

[sharonsalabiglossai]

Reminder

• I have read the README and searched the existing issues.

System Info

[2024-09-17 10:58:53,418] [INFO] [real_accelerator.py:203:get_accelerator] Setting ds_accelerator to cuda (auto detect)

• llamafactory version: 0.9.1.dev0
• Platform: Linux-5.15.0-94-generic-x86_64-with-glibc2.35
• Python version: 3.8.18
• PyTorch version: 2.3.1+cu121 (GPU)
• Transformers version: 4.45.0.dev0
• Datasets version: 2.21.0
• Accelerate version: 0.34.0
• PEFT version: 0.12.0
• TRL version: 0.9.6
• GPU type: NVIDIA A40
• DeepSpeed version: 0.15.1
• Bitsandbytes version: 0.43.3

Reproduction

args = dict(
stage="sft", # do supervised fine-tuning
do_train=True,
model_name_or_path="Qwen/Qwen2-VL-2B-Instruct", # use bnb-4bit-quantized Llama-3-8B-Instruct model
dataset="small_ferrari", # use alpaca and identity datasets
template="qwen2_vl", # use llama3 prompt template
finetuning_type="lora", # use LoRA adapters to save memory
lora_target="all", # attach LoRA adapters to all linear layers
output_dir="qwen2vl_lora_25im_ds2_4", # the path to save LoRA adapters
per_device_train_batch_size=1, # the batch size
gradient_accumulation_steps=1, # the gradient accumulation steps
lr_scheduler_type="cosine", # use cosine learning rate scheduler
logging_steps=10, # log every 10 steps
warmup_ratio=0.1, # use warmup scheduler
save_steps=1000, # save checkpoint every 1000 steps
learning_rate=5e-5, # the learning rate
num_train_epochs=3.0, # the epochs of training
max_samples=500, # use 500 examples in each dataset
max_grad_norm=1.0, # clip gradient norm to 1.0
loraplus_lr_ratio=16.0, # use LoRA+ algorithm with lambda=16.0
bf16=True, # use float16 mixed precision training
use_liger_kernel=True, # use liger kernel for efficient training
cutoff_len=20000, # maximum sequence length
plot_loss=True, # plot loss during training
preprocessing_num_workers=4,
ddp_timeout=9000, # timeout for distributed data parallel
)

# Save args to JSON file

json.dump(args, open("train_qwen2.json", "w", encoding="utf-8"), indent=2)

!llamafactory-cli train train_qwen2.json

Expected behavior

How can we add max_pixels, min_pixels parameters in the args for finetuning qwen2vl in order to control the amount of token per image?

Others

No response