Skip to content
This repository was archived by the owner on Jul 1, 2023. It is now read-only.

Make seed type to be Int32 so that it is compatible with TPUs. #169

Merged
merged 1 commit into from
Jun 3, 2019

Conversation

bgogul
Copy link
Contributor

@bgogul bgogul commented Jun 3, 2019

Otherwise, @burmako and I encountered the following error when running MNIST on TPUs:

 (OpKernel was found, but attributes didn't match) Requested Attributes: T=DT_INT32, Tseed=DT_INT64, dtype=DT_FLOAT
	.  Registered:  device='TPU'; Tseed in [DT_INT32]; T in [DT_INT32, DT_INT64]; dtype in [DT_FLOAT, DT_BFLOAT16]

Otherwise, we get the following error:

```
 (OpKernel was found, but attributes didn't match) Requested Attributes: T=DT_INT32, Tseed=DT_INT64, dtype=DT_FLOAT
	.  Registered:  device='TPU'; Tseed in [DT_INT32]; T in [DT_INT32, DT_INT64]; dtype in [DT_FLOAT, DT_BFLOAT16]
```
@bgogul bgogul requested review from rxwei and burmako June 3, 2019 19:02
@bgogul bgogul merged commit 23cdfd3 into tensorflow:master Jun 3, 2019
@bgogul bgogul deleted the tpu_compatible_seed branch June 3, 2019 20:18
@Shashi456
Copy link
Contributor

Shashi456 commented Jun 3, 2019

@bgogul just purely curious here, you were trying to use s4tf to train a model on a tpu?

That's so cool :o.

@bgogul
Copy link
Contributor Author

bgogul commented Jun 3, 2019

@bgogul just purely curious here, you were trying to use s4tf to train a model on a tpu?

That's so cool :o.

@Shashi456, yes. :) now that GPE is deprecated, we are experimenting with alternative ways of extracting graphs. I will have more PRs on this shortly.

Sign up for free to subscribe to this conversation on GitHub. Already have an account? Sign in.
Labels
None yet
Projects
None yet
Development

Successfully merging this pull request may close these issues.

3 participants