• https://git.ecker.tech/ aims to provide a place to share my efforts while maintaining true ownership of my code, as I do not trust GitHub.

    XMR: 4B9TQdkAkBFYrbj5ztvTx89e5LpucPeTSPzemCihdDi9EBnx7btn8RDNZTBz2zihWsjMnDkzn5As1LU6gLv3KQy8BLsZ8SG

  • Joined on 2022-10-10
mrq pushed to master at mrq/vall-e 2025-02-13 21:46:23 +00:00
02fc67d00c what has science done
mrq pushed to master at mrq/vall-e 2025-02-13 18:57:36 +00:00
mrq pushed to master at mrq/vall-e 2025-02-13 18:44:55 +00:00
e3becec0e8 more better-er loss calc I suppose
mrq pushed to master at mrq/vall-e 2025-02-13 15:30:32 +00:00
e8f182b634 cleaned up loss calc code (it REALLY hates ignore_loss_for_inputs, but is fine with splitting with loss factors)
mrq pushed to master at mrq/vall-e 2025-02-13 05:32:02 +00:00
319ca09a4f cleanup
mrq pushed to master at mrq/vall-e 2025-02-12 22:11:44 +00:00
b52c5c5d80 this seems to work in testing
mrq pushed to master at mrq/vall-e 2025-02-12 17:12:10 +00:00
e029a8804d ironically none of this cruft gets the loss lower than the original way
mrq pushed to master at mrq/vall-e 2025-02-12 06:31:57 +00:00
4b31f5c808 this seems preferable
mrq pushed to master at mrq/vall-e 2025-02-12 06:13:23 +00:00
04fef5dad5 agony
mrq pushed to master at mrq/vall-e 2025-02-11 22:17:25 +00:00
1c0ed6abac added notes on this unfruitful experiment
mrq pushed to master at mrq/vall-e 2025-02-11 20:44:54 +00:00
e5916ea519 for my sanity it seems having extraneous tokens in the embedding/classifier has the loss/acc a little higher than it should
mrq pushed to master at mrq/vall-e 2025-02-11 19:41:31 +00:00
d4a6709fb4 stopgap cringe to get this training session working (it does not seem fruitful)
mrq pushed to master at mrq/vall-e 2025-02-11 03:43:26 +00:00
c0b46b82eb tweaks
mrq pushed to master at mrq/vall-e 2025-02-11 03:42:19 +00:00
f788faca80 tweaks
mrq pushed to master at mrq/vall-e 2025-02-11 03:40:24 +00:00
f306d52ec4 tweaks
mrq pushed to master at mrq/vall-e 2025-02-11 03:40:11 +00:00
38c955fc31 tweaks
mrq pushed to master at mrq/vall-e 2025-02-11 02:49:08 +00:00
d6a679ca5c tweaks
mrq pushed to master at mrq/vall-e 2025-02-11 01:13:15 +00:00
276a2342a4 tweaks to processing script
mrq pushed to master at mrq/vall-e 2025-02-10 21:26:43 +00:00
b3f9b76fd9 invalidate a path if loading via metadata and entry is not in hdf5 (to avoid reparsing my metadata since I'm using a partial copy of my dataset at the moment)
mrq pushed to master at mrq/vall-e 2025-02-09 18:58:45 +00:00