feat: fast random access for streamingDataset without chunk downloading

[deependujha]

Before submitting

• Was this discussed/agreed via a Github issue? (no need for typos and docs improvements)
• Did you read the contributor guideline, Pull Request section?
• Did you make sure to update the docs?
• Did you write any new necessary tests?

What does this PR do?

Fixes #14

• Supports AWS & GCloud.
• This pr only aims at adding support for PyTreeLoader and encrypted/compressed datasets are not supported.

Visualize

import litdata as ld

uri = "gs://deependu-gcp-first-bucket/simple_data"
ds = ld.StreamingDataset(uri, cache_dir="my_cache")

# when iterating, check `my_cache`. it shouldn't download chunks
for i in range(0,1000, 100):
    print(i, ds[i])

# it should download chunks now
for data in ds:
    print(data)

---

Benchmark

for the fast-random-access pr, chunk size: 64MB (Done on lightning studio):

• main branch: 20-22 seconds for 10 random access
• fast-random-access : 5-6 seconds for 10 random access

Note: I tried random access to have enough separation to make sure not all random accesses are from the same chunk.

---

• For single item

0.83 seconds v/s 2 seconds

PR review

Anyone in the community is free to review the PR once the tests have passed.
If we didn't discuss your PR in GitHub issues there's a high chance it will not be merged.

Did you have fun?

Make sure you had fun coding 🙃

[codecov]

Codecov Report

Attention: Patch coverage is 61.38614% with 39 lines in your changes missing coverage. Please review.

Project coverage is 83%. Comparing base (82bf020) to head (29fefcd).
Report is 1 commits behind head on main.

Additional details and impacted files

@@         Coverage Diff         @@
##           main   #631   +/-   ##
===================================
- Coverage    83%    83%   -0%     
===================================
  Files        43     43           
  Lines      6611   6696   +85     
===================================
+ Hits       5507   5553   +46     
- Misses     1104   1143   +39     

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[Copilot]

Pull Request Overview

Adds fast random access for StreamingDataset by introducing a no_store flag and byte-range fetching to avoid downloading entire chunks.

• Introduces no_store in BinaryReader, Cache, and StreamingDataset to conditionally fetch only requested item bytes.
• Implements download_bytes support in S3Downloader and GCPDownloader, and adds read_item_bytes in BinaryReader.
• Adds tests for reader.read_item_bytes and config.download_chunk_bytes_from_index.

Reviewed Changes

Copilot reviewed 8 out of 8 changed files in this pull request and generated no comments.

Show a summary per file

File	Description
tests/streaming/test_reader.py	Added test_reader_read_bytes to validate per-item byte reads.
tests/streaming/test_config.py	Added test_config_download_chunk_bytes for config byte reads.
src/litdata/streaming/reader.py	Added no_store, read_item_bytes, and refactored read.
src/litdata/streaming/item_loader.py	Added default and concrete load_item_from_bytes methods.
src/litdata/streaming/downloader.py	Added base download_bytes, S3 and GCP implementations.
src/litdata/streaming/dataset.py	Propagated no_store and added slice support in __getitem__.
src/litdata/streaming/config.py	Added download_chunk_bytes_from_index for local/remote files.
src/litdata/streaming/cache.py	Propagated no_store through Cache.

Comments suppressed due to low confidence (2)

tests/streaming/test_reader.py:7

• The test uses os.path.join but os is not imported. Please add import os at the top of this file.

import pytest

src/litdata/streaming/item_loader.py:110

• There are two load_item_from_bytes definitions in this class (one stub and one concrete). Consolidate into a single method to avoid unexpected overrides.

    def load_item_from_bytes(

[bhimrazy]

Looking good so far!

I’ve added a few comments and questions.
It would also be helpful to include benchmarks comparing a few individual item loads with and without on_demand_bytes in the description.

[tchaton]

really neat !