Allow users to force TF availability #18650
Conversation
|
The documentation is not available anymore as the PR was closed or merged. |
| @@ -42,6 +42,8 @@ | |||
| USE_TORCH = os.environ.get("USE_TORCH", "AUTO").upper() | |||
| USE_JAX = os.environ.get("USE_FLAX", "AUTO").upper() | |||
|
|
|||
| FORCE_TF_AVAILABLE = os.environ.get("USE_TF", "AUTO").upper() | |||
There was a problem hiding this comment.
should be FORCE_TF_AVAILABLE
There was a problem hiding this comment.
Agreed with @kevint324, I'd put this is a FORCE_TF_AVAILABLE env var :)
There was a problem hiding this comment.
Agh, typo, I'm sorry! That's definitely more than a nit!
| @@ -42,6 +42,8 @@ | |||
| USE_TORCH = os.environ.get("USE_TORCH", "AUTO").upper() | |||
| USE_JAX = os.environ.get("USE_FLAX", "AUTO").upper() | |||
|
|
|||
| FORCE_TF_AVAILABLE = os.environ.get("USE_TF", "AUTO").upper() | |||
There was a problem hiding this comment.
Agreed with @kevint324, I'd put this is a FORCE_TF_AVAILABLE env var :)
|
Ah, additional nit: I'd document this somewhere. |
Rocketknight1
force-pushed
the
allow_force_tf_availability
branch
from
62f5bfb to
6adab4b
Compare
|
@LysandreJik I think a lot of these envvars aren't documented anywhere - I can't see any documentation for USE_TF or USE_TORCH! Maybe we should make a separate docs PR with a list of envvars that |
|
That would be fantastic :) Thanks for your contribution, merging! |
* Allow users to force TF availability * Correctly name the envvar!
We have a user report that with custom Tensorflow builds and package names that
_tf_availablecan returnFalseeven ifimport tensorflowsucceeds, because the user's package name isn't in the allowed list.This is quite niche, so I don't want to do anything that could affect other users and workflows, but I added a
FORCE_TF_AVAILABLEenvvar that will skip version checks and just make sure TF is treated as available. @sgugger WDYT, or is there a better solution?Fixed #18642