Piotr Żelasko
Piotr Żelasko
I don't have the bandwidth atm to add it. If you'd like to make a PR I'm happy to review it.
Does it actually help you achieve greater GPU utilization? Would be great if you can share any benchmarks. I wasn't able to verify the improvements in the original PR.
> It will be around 2x faster on 8xGPU. Wow!! That's fantastic! Let me know if you need any help along the way.
I'm closing this PR in favor of #1341 which has a broader support for lazy cutsets (webdataset, lhotse shar) and both map-style and iterable-style datasets. Thank you for your contributions...
Can you also set `max_steps` to something else than `-1`? E.g. `100000`. Let us know if this helps.
> I had to tune down the batch_duration to 750, with batch_size removed from the configuration. It seems that your actual batch sizes became larger after removing batch_size constraint, leading...
It means that if you keep the “pseudoepoch” size constant, the amount of data seen during a “pseudoepoch” is proportional to the number of GPUs. Generally I don’t encourage thinking...
Thanks for your feedback. I’ll try to improve the documentation to be clearer about this. The option is there if you keep scrolling down, but its indeed missing in the...
Could you fix the formatting checks?
We're seeing some CI instability so the tests are taking longer than ususal. I merged main and re-triggered the failing tests (they don't seem related to your changes). Will keep...