tracel-ai/burn · error

Failed to create cache file

Error message

Failed to create cache file

What it means

`download_if_needed` in the DISTS weights module panics with `File::create(cache_path).expect("Failed to create cache file")` when the cached VGG16 weights file (`vgg16-397923af.pth`) cannot be created after downloading. The cache directory was already ensured writable by `get_cache_dir`, so failures usually come from name collisions (path is a directory), races with another process, or sudden permission/quota changes. Triggered only when the file is not already cached.

Source

Thrown at crates/burn-train/src/metric/vision/dists/weights.rs:37

/// Get the cache directory for DISTS weights.
fn get_cache_dir() -> PathBuf {
    let cache_dir = dirs::cache_dir()
        .expect("Could not get cache directory")
        .join("burn-dataset")
        .join("dists");

    if !cache_dir.exists() {
        create_dir_all(&cache_dir).expect("Failed to create cache directory");
    }

    cache_dir
}

/// Download file if not cached.
fn download_if_needed(url: &str, cache_path: &PathBuf, message: &str) {
    if !cache_path.exists() {
        let bytes = download_file_as_bytes(url, message);
        let mut file = File::create(cache_path).expect("Failed to create cache file");
        file.write_all(&bytes).expect("Failed to write weights");
    }
}

/// Download and load pretrained weights into a DISTS module.
///
/// This loads both:
/// 1. ImageNet pretrained VGG16 backbone weights
/// 2. DISTS trained alpha/beta weights
///
/// Weights are cached in the user's cache directory to avoid re-downloading.
///
/// # Arguments
///
/// * `dists` - The DISTS module to load weights into.
///
/// # Returns
///

View on GitHub (pinned to d16f7ba2ed)

Solutions

1. Remove the stale entry: `rm -rf ~/.cache/burn-dists-cache-path/vgg16-397923af.pth` (or the whole dists cache dir) and retry
2. Ensure the cache directory is writable by the running user (`chmod`/`chown`)
3. Give each concurrent job its own XDG_CACHE_HOME, or pre-populate the cache once before parallel runs
4. Check quota/free space on the cache volume

Example fix

// before (shell)
mpirun -np 4 trainer   # races on cache
// after (shell)
export XDG_CACHE_HOME=$HOME/.cache
python -c "pre_populate()"  # or run one warm-up job first
rm -rf ~/.cache/burn-dataset/dists/vgg16-397923af.pth
mpirun -np 4 trainer

let dists_dir = dirs::cache_dir().unwrap().join("burn-dataset").join("dists");
let vgg = dists_dir.join("vgg16-397923af.pth");
if vgg.is_dir() { eprintln!("{} is a directory; remove it", vgg.display()); }
if !dists_dir.is_dir() || dists_dir.metadata().unwrap().permissions().readonly() {
    eprintln!("dists cache dir not writable");
}

fn cache_file_creatable(dir: &std::path::Path, name: &str) -> bool {
    let p = dir.join(name);
    !p.is_dir() && std::fs::write(dir.join(".probe"), b"").is_ok()
}

let weights = std::panic::catch_unwind(load_pretrained_weights)
    .map_err(|_| anyhow!("VGG16 cache file creation failed; clear the dists cache and retry"))?;

When it happens

Trigger: DISTS `load_pretrained_weights` with no cached VGG16 file: download succeeds, `File::create` fails because `vgg16-397923af.pth` exists as a directory, the dir became read-only, or quota was hit.

Common situations: Corrupted cache from an interrupted run; concurrent training jobs racing to populate the same cache; read-only mount appearing after startup; quota exhaustion on shared scratch volumes.

Related errors

• Failed to create cache file
• Failed to create cache directory
• Failed to write weights
• Failed to create cache directory
• Failed to write weights

AI-assisted analysis of tracel-ai/burn@d16f7ba2ed (2026-09-05). Data as JSON: /api/errors/8f3d3a754b3b855e. Report an issue: GitHub.