1 |
On Wed, Oct 14, 2020 at 11:06 AM Neil Bothwick <neil@××××××××××.uk> wrote: |
2 |
> |
3 |
> On Wed, 14 Oct 2020 10:44:35 -0700, Mark Knecht wrote: |
4 |
> |
5 |
> > I could be wrong but I don't think tensorflow 2 supports python 3.9 |
6 |
> > yet... |
7 |
> > |
8 |
> > https://www.tensorflow.org/install/ |
9 |
> |
10 |
> It doesn't, at least not from the portage ebuild: |
11 |
> |
12 |
> % eix -e tensorflow |
13 |
> * sci-libs/tensorflow |
14 |
> Available versions: (~)2.1.0 (~)2.2.0-r1 (~)2.2.0-r2 (~)2.3.1 {cuda |
15 |
mpi +python xla CPU_FLAGS_X86="avx avx2 fma3 fma4 sse sse2 sse3 sse4_1 |
16 |
sse4_2" PYTHON_TARGETS="python3_6 python3_7 python3_8"} |
17 |
> Homepage: https://www.tensorflow.org/ |
18 |
> Description: Computation framework using data flow graphs |
19 |
for scalable machine learning |
20 |
> |
21 |
|
22 |
So the ebuild is consistent with Google's requirements which is sensible. |
23 |
That doesn't mean tensorflow wouldn't work with 3.9 but that would require |
24 |
the OP to modify the ebuild if there isn't some testing ebuild out in the |
25 |
ether somewhere. |
26 |
|
27 |
I'm sadly stuck running tensorflow on Colab because my monster i7 980 |
28 |
Extreme which has been such a wonderful workhorse for probably 8-10 years |
29 |
now doesn't have the required instruction sets and I cannot justify |
30 |
the cost of a new machine for this one program... <sniff> |