Clean up relationship between deployment_tar and deployment

[bfineran]

Feature Description

Now, any directory that can be downloaded as tar.gz and then unzipped is by default downloaded in such a manner.
This does not prevent us from downloading single, "loose" files from the directory in question.

Downloading the deployment directory

Note, that the deployment is downloaded as tar and then unzipped:

stub = "zoo:mobilenet_v1-1.0-imagenet-pruned.4block_quantized"

model = Model(stub)
deployment_path = model.deployment.path
print(os.listdir(deployment_path))
print(os.listdir(model._path))

Downloading (…)ed/deployment.tar.gz: 100%|██████████| 4.48M/4.48M [00:00<00:00, 12.7MB/s]
['model.onnx']
['deployment', 'deployment.tar.gz']

Downloading the whole model

Note, that the appropriate folders get downloaded as tars and then unzipped

model = Model(stub)
print(os.listdir(model.path))

Downloading (…)d/training/model.pth: 100%|██████████| 32.6M/32.6M [00:02<00:00, 12.5MB/s]
Downloading (…)training-metric.yaml: 100%|██████████| 168/168 [00:00<00:00, 89.7kB/s]
Downloading (…)ed/deployment.tar.gz: 100%|██████████| 4.48M/4.48M [00:00<00:00, 12.0MB/s]
Downloading (…)ple-originals.tar.gz: 100%|██████████| 4.07M/4.07M [00:00<00:00, 9.58MB/s]
Downloading (…)sample-inputs.tar.gz: 100%|██████████| 3.31M/3.31M [00:00<00:00, 11.1MB/s]
Downloading (…)ample-outputs.tar.gz: 100%|██████████| 152k/152k [00:00<00:00, 4.29MB/s]
Downloading (…)sample-labels.tar.gz: 100%|██████████| 170k/170k [00:00<00:00, 3.43MB/s]
Downloading (…)k_quantized/model.md: 100%|██████████| 1.98k/1.98k [00:00<00:00, 1.19MB/s]
Downloading (…)ed/model.onnx.tar.gz: 100%|██████████| 4.48M/4.48M [00:00<00:00, 10.6MB/s]
['sample-inputs', 'sample-labels.tar.gz', 'deployment', 'sample-originals', 'sample-originals.tar.gz', 'deployment.tar.gz', 'model.md', 'sample-outputs.tar.gz', 'sample-inputs.tar.gz', 'model.onnx', 'sample-outputs', 'training', 'model.onnx.tar.gz', 'sample-labels']

Downloading a single "loose" deployment file

model = Model(stub)
deployment_onnx_file = model.deployment.get_file("model.onnx").path
print(os.listdir(os.path.dirname(deployment_onnx_file)))
print(os.listdir(model._path))

Downloading (…)eployment/model.onnx: 100%|██████████| 7.00M/7.00M [00:00<00:00, 11.4MB/s]
['model.onnx']
['deployment']

[dbogunowicz]

Reminder: once this PR lands, we need to update the relevant pathways to the deployment directory in sparseml and deepsparse.

[bfineran]

needs better name

[bfineran]

like download_from_tar_if_available

[dbogunowicz]

best pr ever