rivershah

@mattdangerw I agree with you. `SymbolicTensor` is a better alias over my initial suggestion.

This issue still persists. `MirroredStrategy` with `jit_compile=True` appears broken. To reproduce task count explosion please run the following code. It appears that mirror_strategy run will keep creating threads until system...

I can confirm this as well. `timeout` is not working.

Likewise. Can we please release `jax` that allows `nvidia-cublas-cu12==12.9.0.13`

Hi guys, we really want to move to `jax` backend from `tensorflow`. Any luck with this? We are an intermediate sized tpu customer (for example training runs on `v6e-256`). Should...

@mehtamansi29 Thanks for the pointer but this is incomplete. My request is specifically for `multi-node tpu`. Also many critical details such as tpu runtimes etc are missing.

@jeffcarp This was helpful: [fine_tuning_on_tpu](https://ai.google.dev/gemma/docs/core/distributed_tuning?_gl=1*1wx74lj*_up*MQ..*_ga*NTI0OTYwNTY4LjE3NDU0OTEwOTA.*_ga_P1DBVKWT6V*MTc0NTQ5MTA5MC4xLjAuMTc0NTQ5MTEwMC4wLjAuMTcyNDkxNDAzOA..) We should consider using this as an inspiration for a guide either in `keras` or `keras_hub`. For example fine tuning (without or without LoRA) `gemma3-27b`...

Requesting that this please be added. In ML environments we have non `blosc2` managed threadpool and need to be able to get a handle `blosc2_schunk *schunk = blosc2_schunk_from_buffer_ctx(dctx, src_ptr, src_size,...

Thank you for your great work. Fully agreed with this model and I've done so.

This is a duplicate. Please fix https://github.com/tensorflow/io/issues/1912 The issue is closed but the problem is not resolved