tenstorrent / tt-metal

:metal: TT-NN operator library, and TT-Metalium low level kernel programming model.
Apache License 2.0
387 stars 47 forks source link

Remove hf_modeling_falcon.py #5386

Closed cfjchu closed 2 weeks ago

cfjchu commented 6 months ago

After merging in https://github.com/tenstorrent-metal/tt-metal/pull/5360, we should have Falcon models in the transformers package accessible. Just adding a cleanup task to purge models/demos/falcon7b/reference/hf_modeling_falcon.py and instead reference Falcon classes directly through transformers lib.

fyi @skhorasganiTT

tt-aho commented 6 months ago

I saw the transformers update was reverted in https://github.com/tenstorrent-metal/tt-metal/commit/d38a22c3cf53f8824f676f18b7fd428b921a24a9.

Are we still trying to switch to newer version with falcon?

tt-aho commented 6 months ago

This branch updates Falcon40B to use Falcon from latest Transformers. This also updates to latest PT but no regression testing has been done on the branch so far to see if this breaks anything. @TT-BrianLiu do you want to update Falcon7B on this branch, test and merge? https://github.com/tenstorrent-metal/tt-metal/tree/aho/falcon-40b-transformers-uplift

cfjchu commented 3 weeks ago

fyi @skhorasganiTT