summaryrefslogtreecommitdiffstats
path: root/train_lora.py
Commit message (Expand)AuthorAgeFilesLines
* UpdateVolpeon2023-04-161-1/+5
* TI via LoRAVolpeon2023-04-151-3/+4
* Added cycle LR decayVolpeon2023-04-131-5/+7
* UpdateVolpeon2023-04-131-27/+46
* UpdateVolpeon2023-04-111-7/+16
* Randomize dataset across cyclesVolpeon2023-04-101-1/+3
* UpdateVolpeon2023-04-101-22/+44
* UpdateVolpeon2023-04-091-7/+14
* UpdateVolpeon2023-04-091-6/+1
* UpdateVolpeon2023-04-091-128/+12
* FixVolpeon2023-04-091-4/+6
* Made Lora script interactiveVolpeon2023-04-091-40/+71
* UpdateVolpeon2023-04-081-8/+32
* UpdateVolpeon2023-04-081-10/+22
* UpdateVolpeon2023-04-081-34/+41
* FixVolpeon2023-04-071-35/+42
* Fixed Lora PTIVolpeon2023-04-071-17/+20
* FixVolpeon2023-04-071-5/+8
* Run PTI only if placeholder tokens arg isn't emptyVolpeon2023-04-071-54/+55
* FixVolpeon2023-04-071-1/+1
* UpdateVolpeon2023-04-071-7/+22
* UpdateVolpeon2023-04-071-42/+261
* UpdateVolpeon2023-04-061-19/+32
* Add color jitterVolpeon2023-04-051-3/+12
* Fix choice argsVolpeon2023-04-041-7/+7
* Bring back Lion optimizerVolpeon2023-04-031-3/+27
* Lora: Only register params with grad to optimizerVolpeon2023-04-021-3/+7
* UpdateVolpeon2023-04-011-1/+0
* Add support for Adafactor, add TI initializer noiseVolpeon2023-04-011-1/+15
* UpdateVolpeon2023-03-311-1/+3
* UpdateVolpeon2023-03-311-0/+7
* FixVolpeon2023-03-311-2/+2
* Support Dadaptation d0, adjust sample freq when steps instead of epochs are usedVolpeon2023-03-311-4/+11
* FixVolpeon2023-03-311-1/+2
* FixVolpeon2023-03-281-1/+1
* Support num_train_steps arg againVolpeon2023-03-281-6/+11
* Improved inverted tokensVolpeon2023-03-261-0/+1
* UpdateVolpeon2023-03-251-5/+11
* UpdateVolpeon2023-03-241-0/+7
* Refactoring, fixed Lora trainingVolpeon2023-03-241-1/+72
* UpdateVolpeon2023-03-231-3/+0
* Log DAdam/DAdan dVolpeon2023-03-211-2/+2
* Added dadaptationVolpeon2023-03-211-0/+28
* Fixed SNR weighting, re-enabled xformersVolpeon2023-03-211-27/+9
* UpdateVolpeon2023-03-071-7/+16
* Pipeline: Perlin noise for init imageVolpeon2023-03-041-1/+1
* Implemented different noise offsetVolpeon2023-03-031-1/+1
* UpdateVolpeon2023-03-011-1/+1
* Don't rely on Accelerate for gradient accumulationVolpeon2023-02-211-1/+1
* Embedding normalization: Ignore tensors with grad = 0Volpeon2023-02-211-9/+2