Jupyter kernel crashes during olmocr installation on Vast.ai PyTorch instances

Problem Description
The Jupyter notebook for olmOCR-2 LoRA fine-tuning crashes consistently during package installation on Vast.ai instances. The kernel dies when attempting to install the olmocr library and its dependencies, specifically during the flash-attention compilation step.

Environment
Platform: Vast.ai cloud instances
Template: PyTorch (pre-installed PyTorch environment)
GPU: RTX 3090 (24GB VRAM)
RAM: 12-24GB system RAM
CPU: 8 cores
Storage: 100GB disk
OS: Ubuntu (Vast.ai default)
Python: 3.12
CUDA: 12.8

Steps to Reproduce
Rent Vast.ai instance with PyTorch template (RTX 3090)
Clone repository and

!pip install git+https://github.com/allenai/olmocr.git
!pip install .[train]
!pip install transformers==4.52.4
!pip install flash-attn>=2.8.0.post2 --no-build-isolation
!pip install peft>=0.8.0
!pip install accelerate>=0.25.0
!pip install wandb

Attempted Solutions
Tried installing packages sequentially
Used --no-cache-dir flag
Skipped flash-attention installation
Used conda instead of pip
Checked for PyTorch version conflicts