nodes/edenartlab-sd-lora-trainer
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity
388 Commits 4 Branches 0 Tags
85264bd004af76181a2d08df237c5635f2c1609f
Commit Graph
388 Commits
This Branch
This Branch
All Branches
Author SHA1 Message Date
mayukhdeb 85264bd004 wandb 2024-07-15 07:38:27 -07:00
mayukhdeb d46976f3dd sweep params set 2024-07-15 07:33:45 -07:00
mayukhdeb 8ed5996c1f train and inference on same device 2024-07-15 07:32:27 -07:00
mayukhdeb 8247b4b9dd fix OOM during inference (vae.decode) 2024-07-15 07:20:13 -07:00
mayukhdeb dc28fcb988 inference on same device 2024-07-15 03:28:31 -07:00
mayukhdeb 3c0a3be55f save different sh files for each gpu 2024-07-15 03:28:20 -07:00
mayukhdeb 274ec5c90b disable textual inversion if config.ti_lr is None 2024-07-11 12:46:52 -07:00
mayukhdeb fc54948eff more tweaks 2024-07-11 12:39:30 -07:00
mayukhdeb 32a7ae248d prompts for sweep 2024-07-11 12:34:42 -07:00
mayukhdeb 3cbabd2634 sweep dry runs 2024-07-11 12:26:07 -07:00
mayukhdeb 1cdacbc63a remove old todos 2024-07-11 11:16:22 -07:00
mayukhdeb 4668f755ec switch to adamw_8bit for sd3 transformer lora params 2024-07-11 00:15:09 -07:00
mayukhdeb dd093616b5 fix inference prompts bug 2024-07-09 09:09:56 -07:00
mayukhdeb 2c824d6807 more progress 2024-07-09 05:35:21 -07:00
mayukhdeb 022d51f53a inference fixed 2024-07-09 00:25:02 -07:00
mayukhdeb ecacf815ec re-impl textual inversion for first 2 text encoders 2024-07-08 12:45:55 -07:00
mayukhdeb 90d2269572 testing textual inversion training with frozen transformer 2024-07-04 07:40:41 -07:00
mayukhdeb 4955261cae new checkpoint + deterministic inference 2024-07-03 04:28:06 -07:00
mayukhdeb e84932af29 some small changes to text with main_sd3.py 2024-07-03 04:06:21 -07:00
mayukhdeb 115da83e2d cleaner output dir with checkpoints and generated samples in one folder 2024-07-03 03:43:30 -07:00
mayukhdeb 98b78cd6d6 cleanup + save training samples in output dir 2024-07-03 03:11:56 -07:00
mayukhdeb 7c5e0949ba keep changes 2024-07-02 05:19:34 -07:00
mayukhdeb ba2b7532ad more tweaks 2024-07-02 05:16:22 -07:00
mayukhdeb 2c0939a733 run inference less often 2024-07-02 04:25:26 -07:00
mayukhdeb 3b41479aa1 bfloat16 training + inference 2024-07-02 04:12:23 -07:00
mayukhdeb 6e589c76a8 impement some upstream changes and save a sample every 10 train steps 2024-07-02 01:50:14 -07:00
mayukhdeb 95f78d7b91 completely comement out TI for now 2024-07-01 01:56:49 -07:00
mayukhdeb 3c0ebd7d70 apply mask to loss + some hardcoding for banny debugging 2024-06-28 04:38:43 -07:00
mayukhdeb 6350b5344b more progress 2024-06-28 03:22:49 -07:00
mayukhdeb 2c4bd43044 temporarily remove ti grad norms 2024-06-28 03:12:16 -07:00
mayukhdeb b850a1615d watch grad norms 2024-06-28 03:04:40 -07:00
mayukhdeb 6d6dd98bfa dynamic ti lr 2024-06-28 02:03:34 -07:00
mayukhdeb 0444e35729 clip grad norms 2024-06-26 03:01:34 -07:00
mayukhdeb 0748e1d14e better prompt 2024-06-22 02:32:19 -07:00
mayukhdeb db7507c849 handle T5EncoderModel 2024-06-22 01:38:48 -07:00
mayukhdeb b379a28715 update todos 2024-06-22 01:38:12 -07:00
mayukhdeb 54ff8c4977 sd3 concept inference 2024-06-22 01:25:16 -07:00
mayukhdeb 64f28c8589 save TI embeds and lora adapters 2024-06-22 01:24:30 -07:00
mayukhdeb b12ce26fc9 update command 2024-06-20 03:52:30 -07:00
mayukhdeb b3da65dd39 ignore wandb stuff 2024-06-20 03:46:26 -07:00
mayukhdeb f64da5d4f1 update todo 2024-06-20 03:44:42 -07:00
mayukhdeb 5cc09092a4 tweak param 2024-06-20 03:43:44 -07:00
mayukhdeb 3d3d0ee4cb smash more todos 2024-06-20 03:42:32 -07:00
mayukhdeb 4b1efce0e3 compute loss and update weights 2024-06-20 03:28:49 -07:00
mayukhdeb d115c52274 do just forward passes 2024-06-20 03:20:41 -07:00
mayukhdeb 22127c9917 typo 2024-06-20 01:41:28 -07:00
mayukhdeb 4844845d5c more progress 2024-06-20 01:40:59 -07:00
mayukhdeb 3b15250bbc init train dataloader + update todos for training 2024-06-20 00:54:41 -07:00
mayukhdeb fd448e5437 accomodate T5EncoderModel 2024-06-20 00:54:21 -07:00
mayukhdeb a41cce1486 small cleanup 2024-06-20 00:27:51 -07:00