kabachuha
90b18795fc
Add option to use Scheduled Huber Loss in all training pipelines to improve resilience to data corruption ( #1228 )
...
* add huber loss and huber_c compute to train_util
* add reduction modes
* add huber_c retrieval from timestep getter
* move get timesteps and huber to own function
* add conditional loss to all training scripts
* add cond loss to train network
* add (scheduled) huber_loss to args
* fixup twice timesteps getting
* PHL-schedule should depend on noise scheduler's num timesteps
* *2 multiplier to huber loss cause of 1/2 a^2 conv.
The Taylor expansion of sqrt near zero gives 1/2 a^2, which differs from a^2 of the standard MSE loss. This change scales them better against one another
* add option for smooth l1 (huber / delta)
* unify huber scheduling
* add snr huber scheduler
---------
Co-authored-by: Kohya S <52813779+kohya-ss@users.noreply.github.com >
2024-04-07 13:54:21 +09:00
ykume
cd587ce62c
verify command line args if wandb is enabled
2024-04-05 08:23:03 +09:00
Kohya S
c86e356013
Merge branch 'dev' into dataset-cache
2024-03-26 19:43:40 +09:00
Kohya S
ab1e389347
Merge branch 'dev' into masked-loss
2024-03-26 19:39:30 +09:00
Kohya S
a2b8531627
make each script consistent, fix to work w/o DeepSpeed
2024-03-25 22:28:46 +09:00
Kohya S
993b2ab4c1
Merge branch 'dev' into deep-speed
2024-03-24 18:45:59 +09:00
Kohya S
8d5858826f
Merge branch 'dev' into masked-loss
2024-03-24 18:19:53 +09:00
Kohya S
025347214d
refactor metadata caching for DreamBooth dataset
2024-03-24 18:09:32 +09:00
Kohaku-Blueleaf
ae97c8bfd1
[Experimental] Add cache mechanism for dataset groups to avoid long waiting time for initilization ( #1178 )
...
* support meta cached dataset
* add cache meta scripts
* random ip_noise_gamma strength
* random noise_offset strength
* use correct settings for parser
* cache path/caption/size only
* revert mess up commit
* revert mess up commit
* Update requirements.txt
* Add arguments for meta cache.
* remove pickle implementation
* Return sizes when enable cache
---------
Co-authored-by: Kohya S <52813779+kohya-ss@users.noreply.github.com >
2024-03-24 15:40:18 +09:00
Kohya S
381c44955e
update readme and typing hint
2024-03-24 11:27:18 +09:00
Kohya S
ad97410ba5
Merge pull request #1205 from feffy380/patch-1
...
register reg images with correct subset
2024-03-24 11:14:07 +09:00
Kohya S
79d1c12ab0
disable sample_every_n_xxx if value less than 1 ref #1202
2024-03-24 11:06:37 +09:00
feffy380
0c7baea88c
register reg images with correct subset
2024-03-23 17:28:02 +01:00
Kohya S
f4a4c11cd3
support multiline captions ref #1155
2024-03-23 18:51:37 +09:00
Kohya S
fbb98f144e
Merge branch 'dev' into deep-speed
2024-03-20 18:15:26 +09:00
Kohya S
9b6b39f204
Merge branch 'dev' into masked-loss
2024-03-20 18:14:36 +09:00
Kohya S
855add067b
update option help and readme
2024-03-20 18:14:05 +09:00
Kohya S
bf6cd4b9da
Merge pull request #1168 from gesen2egee/save_state_on_train_end
...
Save state on train end
2024-03-20 18:02:13 +09:00
Kohya S
119cc99fb0
Merge pull request #1167 from Horizon1704/patch-1
...
Add "encoding='utf-8'" for --config_file
2024-03-20 17:39:08 +09:00
Kohya S
3419c3de0d
common masked loss func, apply to all training script
2024-03-17 19:30:20 +09:00
Kohya S
7081a0cf0f
extension of src image could be different than target image
2024-03-17 18:09:15 +09:00
gesen2egee
b5e8045df4
fix control net
2024-03-16 11:51:41 +08:00
kblueleaf
53954a1e2e
use correct settings for parser
2024-03-13 18:21:49 +08:00
kblueleaf
86399407b2
random noise_offset strength
2024-03-13 18:21:49 +08:00
kblueleaf
948029fe61
random ip_noise_gamma strength
2024-03-13 18:21:49 +08:00
gesen2egee
5d7ed0dff0
Merge remote-tracking branch 'kohya-ss/dev' into val
2024-03-13 18:00:49 +08:00
gesen2egee
095b8035e6
save state on train end
2024-03-10 23:33:38 +08:00
Horizon1704
124ec45876
Add "encoding='utf-8'"
2024-03-10 22:53:05 +08:00
gesen2egee
b558a5b73d
val
2024-03-10 04:37:16 +08:00
Kohya S
e3ccf8fbf7
make deepspeed_utils
2024-02-27 21:30:46 +09:00
Kohya S
eefb3cc1e7
Merge branch 'deep-speed' into deepspeed
2024-02-27 18:57:42 +09:00
Kohya S
f2c727fc8c
add minimal impl for masked loss
2024-02-26 23:19:58 +09:00
Kohya S
577e9913ca
add some new dataset settings
2024-02-26 20:01:25 +09:00
BootsofLagrangian
4d5186d1cf
refactored codes, some function moved into train_utils.py
2024-02-22 16:20:53 +09:00
Kohya S
d1fb480887
format by black
2024-02-18 09:13:24 +09:00
Kohya S
358ca205a3
Merge branch 'dev' into dev_device_support
2024-02-12 13:01:54 +09:00
Kohya S
672851e805
Merge branch 'dev' into dev_improve_log
2024-02-12 11:24:33 +09:00
Kohya S
e579648ce9
fix help for highvram arg
2024-02-12 11:12:41 +09:00
Kohya S
e24d9606a2
add clean_memory_on_device and use it from training
2024-02-12 11:10:52 +09:00
Kohya S
75ecb047e2
Merge branch 'dev' into dev_device_support
2024-02-11 19:51:28 +09:00
BootsofLagrangian
03f0816f86
the reason not working grad accum steps found. it was becasue of my accelerate settings
2024-02-09 17:47:49 +09:00
BootsofLagrangian
a98fecaeb1
forgot setting mixed_precision for deepspeed. sorry
2024-02-07 17:19:46 +09:00
BootsofLagrangian
62556619bd
fix full_fp16 compatible and train_step
2024-02-07 16:42:05 +09:00
BootsofLagrangian
3970bf4080
maybe fix branch to run offloading
2024-02-05 22:40:43 +09:00
BootsofLagrangian
2824312d5e
fix vae type error during training sdxl
2024-02-05 20:13:28 +09:00
BootsofLagrangian
64873c1b43
fix offload_optimizer_device typo
2024-02-05 17:11:50 +09:00
Yuta Hayashibe
5f6bf29e52
Replace print with logger if they are logs ( #905 )
...
* Add get_my_logger()
* Use logger instead of print
* Fix log level
* Removed line-breaks for readability
* Use setup_logging()
* Add rich to requirements.txt
* Make simple
* Use logger instead of print
---------
Co-authored-by: Kohya S <52813779+kohya-ss@users.noreply.github.com >
2024-02-04 18:14:34 +09:00
Kohya S
e793d7780d
reduce peak VRAM in sample gen
2024-02-04 17:31:01 +09:00
BootsofLagrangian
dfe08f395f
support deepspeed
2024-02-04 03:12:42 +09:00
Kohya S
2f9a344297
fix typo
2024-02-03 23:26:57 +09:00
