huggingface--transformers/docs/source/en/reference/environment_variables.md

Environment Variables

HF_ENABLE_PARALLEL_LOADING

By default, this option is disabled. When enabled, it allows Torch and Safetensors weight files to be loaded in parallel during model initialization. This can significantly reduce the time required to load large, multi-shard models, often resulting in speedups of around ~50% in supported environments.

Can be set to a string equal to "false" or "true". e.g. os.environ["HF_ENABLE_PARALLEL_LOADING"] = "true".

e.g. facebook/opt-30b on an AWS EC2 g4dn.metal instance can be made to load in ~30s with this enabled vs ~55s without it.

Profile before committing to using this environment variable, this will not produce speed ups for smaller models.

import os

os.environ["HF_ENABLE_PARALLEL_LOADING"] = "true"

from transformers import pipeline

model = pipeline(task="text-generation", model="facebook/opt-30b", device_map="auto")

HF_PARALLEL_LOADING_WORKERS

Determines how many threads should be used when parallel loading is enabled. Default is 8.

If the number of files that are being loaded is less than the number of threads specified, the number that is actually spawned will be equal to the number of files.

e.g. If you specify 8 workers, and there are only 2 files, only 2 workers will be spawned.

Tune as you see fit.

import os

os.environ["HF_ENABLE_PARALLEL_LOADING"] = "true"
os.environ["HF_PARALLEL_LOADING_WORKERS"] = "4"

from transformers import pipeline

model = pipeline(task="text-generation", model="facebook/opt-30b", device_map="auto")