Cleanup 117GB Disk Space | Generated by AI
Here’s the full breakdown of what’s eating 184GB:
TOP SPACE CONSUMERS:
====================
1. /root/nanoGPT/data/fineweb 28GB - old nanoGPT FineWeb dataset
2. /root/nanoGPT/out-fineweb-760m 9GB - old nanoGPT checkpoint
3. /root/zz/datasets/fineweb-edu 40GB - FineWeb-Edu dataset
4. /root/.cache/nanochat/base_ckpt 19GB - nanochat cached checkpoints
5. /root/.cache/nanochat/data 2.7GB - nanochat cached data
6. /root/.cache/uv/archive-v0 31GB - uv package cache (pip equivalent)
7. /root/.cache/pip 4.7GB - pip cache
8. /opt/rocm-7.2.0 22GB - ROCm (NEEDED - keep)
9. /tmp/flash-attention 1.8GB - flash-attention build dir
10. /tmp/torchinductor_root 1.5GB - torch compile cache
RECOMMENDED TO DELETE (saves ~117GB):
# 1. OLD nanoGPT FineWeb dataset (~28GB)
# You have nanochat now, this is the old pipeline's data
rm -rf /root/nanoGPT/data/fineweb
# 2. OLD nanoGPT checkpoint (~9GB)
# Replaced by nanochat checkpoints
rm -rf /root/nanoGPT/out-fineweb-760m
# 3. FineWeb-Edu dataset in zz (~40GB)
# Check if you still need this - was for earlier experiments
rm -rf /root/zz/datasets/fineweb-edu
# 4. uv package cache (~31GB)
# Safe to delete - uv will re-download on next install
rm -rf /root/.cache/uv
# 5. pip cache (~4.7GB)
rm -rf /root/.cache/pip
# 6. /tmp build caches (~3.3GB)
rm -rf /tmp/flash-attention /tmp/torchinductor_root
KEEP:
/opt/rocm-7.2.0(22GB) - ROCm runtime, required for GPU/root/.cache/nanochat/- your nanochat checkpoints/data cache/root/nanochat/- your current working project
Want me to run the cleanup commands?