roxxel 0.3.0

A zero-RAM, multi-modal, sharded binary dataset manager

Roxxel 🚀

Zero-RAM, Multi-Modal, Sharded Binary Dataset Manager

Roxxel is an ultra-lightweight, minimal-dependency, binary dataset format and reader designed for high-performance deep learning pipelines.

By implementing the standard Python sequence protocol over native numpy.memmap views, Roxxel virtualizes massive, multi-sharded, variable-length datasets on-disk as a simple, continuous in-memory list.

---

💡 Motivation

Mainstream deep learning data loaders—such as PyTorch's DataLoader, Google's Grain, and TensorFlow's tf.data—attempt to handle every aspect of the data pipeline (I/O, caching, multiprocessing, shuffling, collation, and transformations) in a single, massive monolithic system. This inevitably leads to severe operational friction:

• PyTorch DataLoader: Relying on multiple workers (num_workers > 0) spawns child processes that trigger Python's fork mechanism. This frequently results in massive memory leaks due to copy-on-write page sharing bugs in Python's GIL. Furthermore, debugging opaque subprocess deadlocks and socket/IPC exhaustion is incredibly frustrating.
• Google Grain: While powerful, it introduces a heavyweight dependency footprint and complex pipeline building abstractions that are difficult to customize or run outside of JAX-specific training pipelines.
• TensorFlow tf.data: Building robust tf.data pipelines is highly complex. Additionally, it forces you to use the opaque TFRecord binary format, which cannot be easily inspected or read without pulling in the massive, multi-gigabyte TensorFlow library as a dependency.

The Roxxel Philosophy

Roxxel shifts the architectural boundary by practicing the Unix philosophy of doing one thing and doing it well. It handles only the hardest, most critical parts of storage—safe contiguous file packing, zero-RAM memory mapping, and O(1) seek indexing—and leaves all batching, threading, and transformations to plain, standard Python and NumPy code.

---

🌟 Unique Benefits of Roxxel

1. Zero-RAM Overhead: Roxxel maps your dataset directly into virtual memory via the operating system's kernel page cache using numpy.memmap. Even for multi-terabyte datasets, it consumes exactly 0 bytes of Python RAM for the data.
2. 100% Framework Agnostic: Because Roxxel is built purely on Python standard libraries and NumPy, it is entirely decoupled from any ML framework. You can use the exact same Roxxel dataset across PyTorch, JAX, TensorFlow, or pure CPU environments with zero code changes.
3. No Multiprocessing Deadlocks: Because reading from memory maps is natively thread-safe and extremely fast, you can implement high-performance, asynchronous loading using simple Python threads (threading.Thread) or thread pools. You never have to worry about subprocess IPC bottlenecks or fork-related deadlocks.
4. Modality-Agnostic Variable-Length Records: Unlike rigid formats, Roxxel accepts arbitrary variable-length binary payloads contiguously. You can store JPEGs, MP4 clips, text token arrays, or audio samples in a single, unified structure with zero padding waste.
5. Clean Sharded Portability: Roxxel automatically splits massive datasets into sequentially numbered shards during writes. During reads, it seamlessly virtualizes them into a single continuous sequence using fast binary search boundaries. Shards are easy to distribute, copy, and stream over networks.

---

🛠️ File Format Architecture

To prevent header contamination (where inline metadata blocks corrupt flat memory maps), Roxxel writes your entire dataset into a single contiguous binary file with a trailing index table:

+-------------------------------------------------------------+
|                                                             |
| 1. RAW CONTIGUOUS PAYLOAD DATA SECTION                      |
|    (No headers, no prefixes, completely clean bytes)        |
|                                                             |
+-------------------------------------------------------------+
|                                                             |
| 2. TRAILING INDEX TABLE SECTION                             |
|    (Flat array of uint64 offsets pointing to record ends)   |
|                                                             |
+-------------------------------------------------------------+
| 3. FOOTER (Exactly 24 bytes)                                |
|    [Total Records (8B)] [Raw Data Size (8B)] [MAGIC (8B)]  |
+-------------------------------------------------------------+

Because the raw data section is completely uninterrupted, you can interpret the entire archive as a single contiguous array in one line (e.g. for LLM token pre-training) or resolve individual records in $O(1)$ constant time.

📦 Installation

Roxxel can be installed via pip directly from PyPI:

pip install roxxel

---

🚀 Getting Started

1. Compiling raw data into uniform blocks

Pass an iterable stream of strings or raw bytes directly into the write() API. Roxxel will automatically group them into strictly uniform blocks (e.g., 4096-byte blocks) and write them to disk.

from roxxel import Roxxel

# Generator yielding text documents of variable lengths
def text_generator():
    yield "The quick brown fox jumps over the lazy dog."
    yield "Generative AI and SSMs like Xenron are transforming sequences."
    yield "Roxxel delivers zero-RAM, highly efficient data loading."

# Instantiate and write (automatically shards at 1GB, blocks of 4KB)
rox = Roxxel("./wiki_*.rox")
rox.write(
    data_generator=text_generator(),
    block_size=4096,
    max_shard_bytes=1024**3,
    separator=b"\xff"
)

2. High-Performance Deterministic Streaming (NumPy & JAX)

Opening the dataset and streaming globally shuffled, JAX-sharded batches takes just a few lines. The stream() API handles circular prefetch buffering and PCIe hardware transfers asynchronously in the background:

import jax
from roxxel import Roxxel

# 1. Open the virtualized multi-sharded dataset
with Roxxel("./wiki_*.rox") as dataset:
    # 2. Yield globally shuffled JAX-sharded device arrays automatically!
    # If JAX is installed, it outputs JAX arrays. Otherwise, standard NumPy arrays.
    dataloader = dataset.stream(
        seq_len=1024,
        batch_size=32,
        seed=42,
        start_step=0  # Supports instant O(1) checkpoint fast-forwarding!
    )
    
    for batch in dataloader:
        # batch is a JAX device-put array of shape (32, 1024) ready for TPU/GPU!
        outputs = train_step(state, batch)

---

🍳 Cookbooks

A. Flat Token Reshaping (e.g. LLM Training)

If you want to bypass sequence boundaries entirely and treat the whole sharded database as a single, contiguous token stream:

with Roxxel("./wiki_*.rox") as dataset:
    # Get a 1D memory-mapped view of the entire dataset across all shards
    # (Since len(dataset[0]) is uniform, this represents one solid contiguous sequence)
    flat_tokens = dataset.raw_data.view(np.uint16)
    
    # Reshape and train dynamically in NumPy
    total_sequences = len(flat_tokens) // 2048
    reshaped_dataset = flat_tokens[:total_sequences * 2048].reshape(total_sequences, 2048)

B. Instant O(1) Checkpoint Resuming

To save training progress, simply checkpoint your current step. Resuming takes less than 1 millisecond as Roxxel instantly jumps past the consumed blocks using basic index arithmetic—completely skipping the need to execute dummy fast-forward loops:

# During save:
step = current_step  # Save this integer in your checkpoint manager

# During restore:
with Roxxel("./wiki_*.rox") as dataset:
    # Instantly fast-forward and resume streaming from step 4500
    dataloader = dataset.stream(
        seq_len=1024,
        batch_size=32,
        seed=42,
        start_step=4500
    )

---

⚖️ License

MIT License. Feel free to use, modify, and distribute.