High-Throughput Data Loading | Dataford Interview Questions

Implement a data-loading routine optimized for high-throughput batching and minimal I/O overhead in ML pipelines.

Implement load_batches(shards, requests, batch_size). shards maps each shard ID to an array of sample values, and requests contains (shard_id, sample_index) pairs. Read each referenced shard at most once, process shards in their first-request order, preserve request order within each shard, and return consecutive batches of at most batch_size values.

Output

Return a list of lists containing the loaded values.

Constraints

1 <= len(requests) <= 1000
1 <= batch_size <= 1000
Every requested shard ID exists in shards
Every sample index is valid for its shard
Each request is a pair of a shard ID and an integer sample index

Function Signature

def load_batches(shards, requests, batch_size):

Your question is High-Throughput Data Loading. Start with the requirements in the Question tab.

Run and submit as often as you like. When you're ready, talk me through your approach or go straight to the code.

Problem