982 Commits

Author SHA1 Message Date
sdbds
5f9047c8cf add truncation when > max_length 2025-02-26 01:00:35 +08:00
Kohya S
67fde015f7 Merge branch 'dev' into sd3 2025-02-24 18:56:15 +09:00
Kohya S.
386b7332c6 Merge pull request #1918 from tsukimiya/fix_vperd_warning
Remove v-pred warning.
2025-02-24 18:55:25 +09:00
Kohya S
905f081798 Merge branch 'dev' into sd3 2025-02-24 18:54:28 +09:00
sdbds
fc772affbe 1、Implement cfg_trunc calculation directly using timesteps, without intermediate steps.
2、Deprecate and remove the guidance_scale parameter because it used in inference not train

3、Add inference command-line arguments --ct for cfg_trunc_ratio and --rc for renorm_cfg to control CFG truncation and renormalization during inference.
2025-02-24 14:10:24 +08:00
rockerBOO
2c94d17f05 Fix typo 2025-02-23 20:21:06 -05:00
rockerBOO
48e7da2d4a Add sample batch size for Lumina 2025-02-23 20:19:24 -05:00
rockerBOO
ba725a84e9 Set default discrete_flow_shift to 6.0. Remove default system prompt. 2025-02-23 18:01:09 -05:00
rockerBOO
42a801514c Fix system prompt in datasets 2025-02-23 13:48:37 -05:00
rockerBOO
6d7bec8a37 Remove non-used code 2025-02-23 01:46:47 -05:00
rockerBOO
025cca699b Fix samples, LoRA training. Add system prompt, use_flash_attn 2025-02-23 01:29:18 -05:00
Kohya S
efb2a128cd fix wandb val logging 2025-02-21 22:07:35 +09:00
Yidi
13df47516d Remove position_ids for V2
The postions_ids cause errors for the newer version of transformer.
This has already been fixed in convert_ldm_clip_checkpoint_v1() but
not in v2.
The new code applies the same fix to convert_ldm_clip_checkpoint_v2().
2025-02-20 04:49:51 -05:00
rockerBOO
7f2747176b Use resize_image where resizing is required 2025-02-19 14:20:40 -05:00
rockerBOO
545425c13e Typo 2025-02-19 14:20:40 -05:00
rockerBOO
d0128d18be Add resize interpolation CLI option 2025-02-19 14:20:40 -05:00
rockerBOO
58e9e146a3 Add resize interpolation configuration 2025-02-19 14:20:40 -05:00
Kohya S
dc7d5fb459 Merge branch 'sd3' into val-loss-improvement 2025-02-18 21:34:30 +09:00
rockerBOO
bd16bd13ae Remove unused attention, fix typo 2025-02-18 01:21:18 -05:00
rockerBOO
98efbc3bb7 Add documentation to model, use SDPA attention, sample images 2025-02-18 00:58:53 -05:00
rockerBOO
9436b41061 Fix validation split and add test 2025-02-17 14:28:41 -05:00
rockerBOO
3ed7606f88 Clear sizes for validation reg images to be consistent 2025-02-17 12:07:23 -05:00
rockerBOO
3365cfadd7 Fix sizes for validation split 2025-02-17 12:07:23 -05:00
sdbds
aa36c48685 update for always use gemma2 mask 2025-02-17 19:00:18 +08:00
rockerBOO
f3a010978c Clear sizes for validation reg images to be consistent 2025-02-16 22:28:34 -05:00
rockerBOO
3c7496ae3f Fix sizes for validation split 2025-02-16 22:18:14 -05:00
rockerBOO
16015635d2 Update metadata.resolution for Lumina 2 2025-02-16 01:36:29 -05:00
rockerBOO
60a76ebb72 Add caching gemma2, add gradient checkpointing, refactor lumina model code 2025-02-16 01:06:34 -05:00
rockerBOO
a00b06bc97 Lumina 2 and Gemma 2 model loading 2025-02-15 14:56:11 -05:00
sdbds
c0caf33e3f update 2025-02-15 16:38:59 +08:00
sdbds
d154e76c45 init 2025-02-12 16:30:05 +08:00
Kohya S
a24db1d532 fix: validation timestep generation fails on SD/SDXL training 2025-02-04 22:02:42 +09:00
tsukimiya
4a71687d20 不要な警告の削除
(おそらく be14c06267 の修正漏れ )
2025-02-04 00:42:27 +09:00
Kohya S
58b82a576e Fix to work with validation dataset 2025-01-26 21:21:21 +09:00
rockerBOO
c04e5dfe92 Fix loss recorder on 0. Fix validation for cached runs. Assert on validation dataset 2025-01-23 09:57:24 -05:00
rockerBOO
b489082495 Disable repeats for validation datasets 2025-01-12 16:42:04 -05:00
rockerBOO
2bbb40ce51 Fix regularization images with validation
Adding metadata recording for validation arguments
Add comments about the validation split for clarity of intention
2025-01-12 14:29:50 -05:00
rockerBOO
264167fa16 Apply is_training_dataset only to DreamBoothDataset. Add validation_split check and warning 2025-01-09 12:43:58 -05:00
rockerBOO
9fde0d7972 Handle tuple return from generate_dataset_group_by_blueprint 2025-01-08 18:38:20 -05:00
rockerBOO
556f3f1696 Fix documentation, remove unused function, fix bucket reso for sd1.5, fix multiple datasets 2025-01-08 13:41:15 -05:00
rockerBOO
1231f5114c Remove unused train_util code, fix accelerate.log for wandb, add init_trackers library code 2025-01-07 22:31:41 -05:00
Kohya S
a9c5aa1f93 add CFG to FLUX.1 sample image 2025-01-05 22:28:51 +09:00
rockerBOO
695f38962c Move get_huber_threshold_if_needed 2025-01-03 15:25:12 -05:00
rockerBOO
0522070d19 Fix training, validation split, revert to using upstream implemenation 2025-01-03 15:20:25 -05:00
rockerBOO
6604b36044 Remove duplicate assignment 2025-01-03 02:04:59 -05:00
rockerBOO
fbfc2753eb Update text for train/reg with repeats 2025-01-03 01:53:12 -05:00
rockerBOO
c8c3569df2 Cleanup order, types, print to logger 2025-01-03 01:26:45 -05:00
rockerBOO
534059dea5 Typos and lingering is_train 2025-01-03 01:18:15 -05:00
rockerBOO
d23c7322ee Merge remote-tracking branch 'hina/feature/val-loss' into validation-loss-upstream
Modified implementation for process_batch and cleanup validation
recording
2025-01-03 00:48:08 -05:00
rockerBOO
7f6e124c7c Merge branch 'gesen2egee/val' into validation-loss-upstream
Modified various implementations to restore original behavior
2025-01-02 23:04:38 -05:00