Run:AI Model Streamer for Azure

0.15.9 verified Fri May 01 auth: no python

A library for streaming AI model artifacts from Azure Blob Storage to GPU nodes in Run:AI environments. Supports asynchronous prefetching, caching, and direct loading into PyTorch/TensorFlow. Current version 0.15.9 with weekly releases.

pip install runai-model-streamer-azure

Common errors

error ModuleNotFoundError: No module named 'runai_model_streamer' ↓

cause Missing dependency: runai-model-streamer must be installed separately.

fix

pip install runai-model-streamer

error AttributeError: 'StreamConfig' object has no attribute 'block_size' ↓

cause block_size renamed to read_chunk_size in 0.14.0.

fix

Use 'read_chunk_size' instead of 'block_size' in StreamConfig.

error azure.core.exceptions.ClientAuthenticationError: No connection string supplied. ↓

cause Missing or empty AZURE_STORAGE_CONNECTION_STRING environment variable.

fix

Set AZURE_STORAGE_CONNECTION_STRING to a valid Azure Blob Storage connection string.

Warnings

breaking Version 0.13.0 changed the base class of AzureModelStreamer; custom subclasses must inherit from the new base or override fetch() explicitly. ↓

fix Update subclasses: inherit from runai_model_streamer._base.BaseStreamer instead.

deprecated The 'StreamConfig.block_size' parameter will be removed in 0.16.0; use 'StreamConfig.read_chunk_size' instead. ↓

fix Replace block_size with read_chunk_size in StreamConfig.

gotcha Connection string authentication (AZURE_STORAGE_CONNECTION_STRING) is required; using DefaultAzureCredential silently fails if no connection string is set. ↓

fix Set environment variable AZURE_STORAGE_CONNECTION_STRING or pass connection_string explicitly.

Imports

AzureModelStreamer

wrong

from runai.model_streamer.azure import ...

correct

from runai_model_streamer_azure import AzureModelStreamer

Old dotted path was used in early betas, now flat import.

StreamConfig

from runai_model_streamer_azure import StreamConfig

Quickstart

Initializes a streamer to fetch a model checkpoint from Azure Blob Storage.

from runai_model_streamer_azure import AzureModelStreamer, StreamConfig

config = StreamConfig(
    container="my-models",
    blob_prefix="models/resnet50/",
    connection_string=os.environ.get("AZURE_STORAGE_CONNECTION_STRING", "")
)
streamer = AzureModelStreamer(config)
local_path = streamer.fetch("model.pt")
print(local_path)