aboutsummaryrefslogtreecommitdiffstats
path: root/modules/hypernetworks/hypernetwork.py
Commit message (Collapse)AuthorAgeFilesLines
* patch bug (SeverianVoid's comment on 5245c7a)timntorres2022-10-261-1/+1
|
* remove duplicate keys and lowercaseAngelBottomless2022-10-261-1/+1
|
* Weight initialization and More activation funcAngelBottomless2022-10-261-9/+38
| | | | | | | | | | | | | | | | add weight init add weight init option in create_hypernetwork fstringify hypernet info save weight initialization info for further debugging fill bias with zero for He/Xavier initialize LayerNorm with Normal fix loading weight_init
* check length for varianceAngelBottomless2022-10-241-2/+10
|
* convert deque -> listAngelBottomless2022-10-241-1/+1
| | | I don't feel this being efficient
* statistics for pbarAngelBottomless2022-10-241-2/+10
|
* cleanup some codeAngelBottomless2022-10-241-11/+3
|
* Hypernetworks - fix KeyError in statistics cachingAngelBottomless2022-10-241-2/+2
| | | Statistics logging has changed to {filename : list[losses]}, so it has to use loss_info[key].pop()
* Update hypernetwork.pyDepFA2022-10-231-4/+7
|
* Allow tracking real-time lossAngelBottomless2022-10-221-1/+1
| | | | Someone had 6000 images in their dataset, and it was shown as 0, which was confusing. This will allow tracking real time dataset-average loss for registered objects.
* Update hypernetwork.pyAngelBottomless2022-10-221-11/+44
|
* small fixdiscus04342022-10-221-4/+3
|
* Merge branch 'AUTOMATIC1111:master' into masterdiscus04342022-10-221-0/+11
|\
| * added a guard for hypernet training that will stop early if weights are ↵AUTOMATIC2022-10-221-0/+11
| | | | | | | | getting no gradients
* | small fixdiscus04342022-10-221-7/+5
| |
* | add an option to avoid dying reludiscus04342022-10-221-6/+6
| |
* | Merge branch 'master' of upstreamdiscus04342022-10-221-8/+18
|\|
| * Remove unused variable.timntorres2022-10-211-1/+0
| |
| * Match hypernet name with filename in all cases.timntorres2022-10-211-1/+7
| |
| * turns out LayerNorm also has weight and bias and needs to be pre-multiplied ↵AUTOMATIC2022-10-211-2/+2
| | | | | | | | and trained for hypernets
| * Merge branch 'master' into training-help-textAUTOMATIC11112022-10-211-40/+56
| |\
| | * Revise comments.timntorres2022-10-211-1/+1
| | |
| | * Issue #2921-Give PNG info to Hypernet previews.timntorres2022-10-211-2/+7
| | |
| | * a more strict check for activation type and a more reasonable check for type ↵AUTOMATIC2022-10-211-3/+9
| | | | | | | | | | | | of layer in hypernets
| * | change html outputDepFA2022-10-191-1/+1
| | |
* | | add dropoutdiscus04342022-10-221-26/+42
| | |
* | | Revert "fix bugs and optimizations"aria1th2022-10-201-59/+46
| | | | | | | | | | | | This reverts commit 108be15500aac590b4e00420635d7b61fccfa530.
* | | fix bugs and optimizationsAngelBottomless2022-10-201-46/+59
| | |
* | | only linearAngelBottomless2022-10-201-5/+5
| | |
* | | generalized some functions and option for ignoring first layerAngelBottomless2022-10-201-8/+15
| |/ |/|
* | updatediscus04342022-10-201-10/+19
| |
* | fix for #3086 failing to load any previous hypernetdiscus04342022-10-191-32/+28
|/
* layer options moves into create hnet uidiscus04342022-10-191-32/+32
|
* Merge branch 'master' into masterdiscus04342022-10-191-3/+2
|\
| * Use training width/height when training hypernetworks.Silent2022-10-191-2/+2
| |
* | updatediscus04342022-10-181-2/+2
| |
* | updatediscus04342022-10-181-6/+8
| |
* | add options to custom hypernetwork layer structurediscus04342022-10-181-21/+67
|/
* check NaN for hypernetwork tuningAngelBottomless2022-10-151-4/+6
|
* add option to use batch size for trainingAUTOMATIC2022-10-151-9/+24
|
* remove duplicate code for log loss, add step, make it read from options ↵AUTOMATIC2022-10-141-14/+6
| | | | rather than gradio input
* Merge remote-tracking branch 'Melanpan/master'AUTOMATIC2022-10-141-0/+15
|\
| * Add learn_rate to csv and removed a left-over debug statementMelan2022-10-131-3/+3
| |
| * Save a csv containing the loss while trainingMelan2022-10-121-1/+16
| |
* | add option to read generation params for learning previews from txt2imgAUTOMATIC2022-10-141-5/+16
| |
* | add hypernetwork multipliersAUTOMATIC2022-10-131-1/+7
|/
* train: change filename processing to be more simple and configurableAUTOMATIC2022-10-121-24/+16
| | | | | | train: make it possible to make text files with prompts train: rework scheduler so that there's less repeating code in textual inversion and hypernets train: move epochs setting to options
* change textual inversion tab to trainAUTOMATIC2022-10-121-1/+1
| | | | remake train interface to use tabs
* xy_grid: Find hypernetwork by closest nameMilly2022-10-121-0/+11
|
* apply lr schedule to hypernetsAUTOMATIC2022-10-111-4/+15
|