DeepSpeed
DeepSpeed copied to clipboard
Add support for T5 for deepspeed-inference
This is the PR to add support to T5. Currently it is still work in progress. A lot of the codes are adapted from https://github.com/microsoft/DeepSpeed/pull/2451
Hi @HeyangQin, thanks for your great work! Is there any update on the timeline of merging T5 support?
For whatever reason, it seems it is requiring a lot more VRAM when this policy is injected.
@HeyangQin @loadams I take it that this is not on your roadmap?
Hello @alexcoca. T5 support should already be live with https://github.com/microsoft/DeepSpeed/pull/2962. Please feel free to let us know if you need more information