Timur Abishev

@rahulbatra85 I've tried above code with Jax image you provided and it failed with following error: ``` NotImplementedError: MLIR translation rule for primitive 'pallas_call' not found for platform cpu ```...

@rahulbatra85 it seems like in my case jax doesn't recognize gpu: ``` > sudo docker run -it --cap-add=SYS_PTRACE --security-opt seccomp=unconfined --device=/dev/kfd --device=/dev/dri --group-add video --ipc=host --shm-size 8G docker.io/rocm/jax-build:rocm6.0.0-jax0.4.20-py3.10.0-jax_triton > python...

@rahulbatra85 2x 7900 XTX `rocm-smi`: ``` ====================================== ROCm System Management Interface ====================================== ================================================ Concise Info ================================================ Device [Model : Revision] Temp Power Partitions SCLK MCLK Fan Perf PwrCap VRAM% GPU%...

Hi, any updates on the issue? There are some projects showing viability of using INT4 for compute (i.e. https://github.com/efeslab/Atom), and it would be really awesome if it's supported in Triton.

@JackCaoG thanks for the response! Couple of follow up questions: - This way time of `randn` creation gets into the operation time, is there way to avoid it? Put `xm.mark_step()`...

@miladm I was able to benchmark my code and get consistent and probably more or less correct results. I'm trying to get access to TPU v5e for a week now...

@ultrons I would really appreciate if you can help me. I'm trying to create v5e nodes to try it out and got some quota approved, but it doesn't seem working....

@mrwyattii something like current `generate` method but `async` would be enough. I would like to call this method from another method which is declared `async` itself (with a possibility of...

Removing `.orElse(composeSame)` from default map reduce optimization rules fixes the test.

I think I know what's wrong. The test I wrote doing next: ``` val crossedMapped = TypedPipe.from(source) .cross(mapPipe) .map { case (value, map) => map(value) } crossedMapped.toPipe('value).write(sink1) crossedMapped.map(identity).toPipe('value).write(sink2) ``` During...