Conversation

Notices

1. Tomeu Vizoso (tomeu@mastodon.social)'s status on Saturday, 27-Jan-2024 21:32:56 JST Tomeu Vizoso

   I'm so happy to announce that the first NPU driver has been accepted in Mesa.

   Have written a bigger update to celebrate this and properly thank everybody who has been supporting me in this amazing trip so far.

   https://blog.tomeuvizoso.net/2024/01/etnaviv-npu-update-15-we-are-upstream.html

   #vivante #npu #etnaviv #mesa

   • minute (mntmn@mastodon.social)'s status on Saturday, 27-Jan-2024 21:34:07 JST minute

     in reply to

     @tomeu congrats! i was looking a bit into foss speech recognition today, do you think this would make sense to use with e.g. DeepSpeech (it seems to use tflite)?

   • Tomeu Vizoso (tomeu@mastodon.social)'s status on Thursday, 08-Feb-2024 17:51:21 JST Tomeu Vizoso

     in reply to

     • minute

     @mntmn I think it should work well, but I have looked at the model and it uses lots of operations that aren't supported at the moment.

     I would need to take a closer look, but because there are so many operations that aren't convolutions, I think it is possible that this model would be more efficiently run on a GPU.

   • minute (mntmn@mastodon.social)'s status on Thursday, 08-Feb-2024 17:54:05 JST minute

     in reply to

     • minute

     @tomeu @mntmn ah, very interesting! thanks for looking into it.

   • Tomeu Vizoso (tomeu@mastodon.social)'s status on Friday, 09-Feb-2024 01:43:16 JST Tomeu Vizoso

     in reply to

     • minute
     • Collabora

     @mntmn Actually, I just ran the model and not a single operation was run on the systolic arrays. Only a few were run in the (underpowered) programmable core and the rest were rejected by Vivante's driver.

     The @collabora guys are working on open-source TTS models, maybe they have something that maps better to this HW.