Require an increase in file size for Intel's cpu-optimized tensorflow lib

[s1113950]

I am having the same issue as #103. I'm a part of Intel as well and I'm trying to publish tensorflow-mkl, essentially a copy of tensorflow but optimized for CPU using the Intel MKL library. AFAIK previously we didn't publish it and only have dockerfiles with a compile flag for users, but now we've been asked to provide our own library.

I'm currently trying to do this with our test.pypi.org account of IntelAI before I push the real thing to our prod account nervanasys. I either get this:

robertso@robertso-mac02:~/Nervana$ twine upload --repository-url https://test.pypi.org/legacy/ tensorflow_mkl-1.11.0-cp*
Enter your username: IntelAI
Enter your password:
Uploading distributions to https://test.pypi.org/legacy/
Uploading tensorflow_mkl-1.11.0-cp27-cp27mu-linux_x86_64.whl
100%|███████████████████████████████████████████████████████████████████████████████████████████████████████████████| 413M/413M [03:08<00:00, 2.30MB/s]
Received "502: Bad Gateway" Package upload appears to have failed.  Retry 1 of 5
Uploading tensorflow_mkl-1.11.0-cp27-cp27mu-linux_x86_64.whl

Or I get Received "503: first byte timeout" Package upload appears to have failed (often on subsequent retries).

The whl we're trying to publish is 413MB in size

[s1113950]

I just learned that we have another mkl account here: https://pypi.org/project/mkl/#files with very large files. I am waiting on getting credentials to that account and then would like to publish this mkl package there if possible

[s1113950]

We've decided to rename our package to be intel-tensorflow which better aligns with our previous naming of intel-numpy.

[s1113950]

Ahh I screwed something up with wheel creation. The real wheel size should be closer to 106 MB

[s1113950]

Latest wheel sizes:

-rw-r--r--. 1 robertso jenkins 107066238 Nov  1 13:41 intel_tensorflow-1.11.0-cp27-cp27mu-linux_x86_64.whl
-rw-r--r--. 1 robertso jenkins 107058950 Nov  1 13:42 intel_tensorflow-1.11.0-cp34-cp34m-linux_x86_64.whl
-rw-r--r--. 1 robertso jenkins 107057417 Nov  1 13:43 intel_tensorflow-1.11.0-cp36-cp36m-linux_x86_64.whl
-rw-r--r--. 1 robertso jenkins 107057167 Nov  1 13:44 intel_tensorflow-1.11.0-cp35-cp35m-linux_x86_64.whl

[claynerobison]

@dstufft Any chance you can help? 120 MB should do it.

[di]

The warehouse tracker is the right place for these requests.