phaeusler/llama-lol

llama-lol's Issues

CUDA out of memory. Tried to allocate 3.00 GiB (GPU 0; 23.69 GiB total capacity; 14.73 GiB already allocated;

torch.cuda.OutOfMemoryError: CUDA out of memory. Tried to allocate 3.00 GiB (GPU 0; 23.69 GiB total capacity; 14.73 GiB already allocated; 722.00 MiB free; 20.91 GiB reserved in total by PyTorch) If reserved memory is >> allocated memory try setting max_split_size_mb to avoid fragmentation. See documentation for Memory Management and PYTORCH_CUDA_ALLOC_CONF

training_arguments = TrainingArguments(
    output_dir="./results",
    per_device_train_batch_size=40,
    gradient_accumulation_steps=1,
    optim="paged_adamw_32bit",
    save_steps=50,
    logging_steps=10,
    learning_rate=2e-4,
    fp16=True,
    max_grad_norm=0.3,
    max_steps=1000,
    warmup_ratio=0.03,
    lr_scheduler_type="constant",
)

changing

per_device_train_batch_size=40

to this fixed problem

per_device_train_batch_size=10,

Hi Phil

Are you available for hire?
I'm working on a startup - and we're taking app to market.

https://philliphaeusler.com/posts/llama_lol/
I love the seinfeld generated text.

I want to do achive

take an mp4 - lip sync - reworded as seinfeld.

done.