summaryrefslogtreecommitdiffstats
path: root/models
Commit message (Expand)AuthorAgeFilesLines
* UpdateVolpeon2023-04-092-6/+14
* UpdateVolpeon2023-04-082-2/+2
* TI: Bring back old embedding decayVolpeon2023-04-042-14/+15
* Improved sparse embeddingsVolpeon2023-04-032-31/+78
* TI: Delta learningVolpeon2023-04-031-17/+33
* RevertVolpeon2023-04-011-38/+4
* Combined TI with embedding and LoRAVolpeon2023-04-011-4/+15
* Experimental: TI via LoRAVolpeon2023-04-011-15/+38
* Add support for Adafactor, add TI initializer noiseVolpeon2023-04-011-1/+9
* UpdateVolpeon2023-03-311-24/+9
* Fix TIVolpeon2023-03-271-9/+25
* Fix TIVolpeon2023-03-271-25/+9
* Revert to regular embeddingsVolpeon2023-03-271-19/+15
* Sparse TI embeddings without sparse tensorsVolpeon2023-03-271-17/+23
* Fix TI embeddings initVolpeon2023-03-261-1/+1
* Improved TI embeddingsVolpeon2023-03-261-7/+23
* UpdateVolpeon2023-03-231-3/+3
* UpdateVolpeon2023-03-011-1/+1
* Embedding normalization: Ignore tensors with grad = 0Volpeon2023-02-211-8/+0
* Optimized embedding normalizationVolpeon2023-01-171-5/+2
* UpdateVolpeon2023-01-171-0/+3
* CleanupVolpeon2023-01-141-1/+4
* More modularizationVolpeon2023-01-131-5/+1
* Removed PromptProcessor, modularized training loopVolpeon2023-01-133-39/+39
* Code deduplicationVolpeon2023-01-131-2/+2
* Fixed TI decayVolpeon2023-01-121-1/+9
* UpdateVolpeon2023-01-071-5/+0
* UpdateVolpeon2023-01-061-1/+1
* Added EMA to TIVolpeon2023-01-051-3/+3
* UpdateVolpeon2023-01-051-4/+6
* Various cleanupsVolpeon2023-01-052-7/+7
* UpdateVolpeon2023-01-042-2/+4
* Added vector dropoutVolpeon2023-01-031-2/+25
* UpdateVolpeon2023-01-011-11/+11
* CleanupVolpeon2023-01-012-11/+1
* Fix MultiCLIPTokenizer (forgot to override encode)Volpeon2023-01-011-13/+20
* UpdatesVolpeon2023-01-012-16/+71
* Fixed accuracy calc, other improvementsVolpeon2023-01-011-7/+11
* FixVolpeon2023-01-011-1/+1
* Better token shufflingVolpeon2023-01-011-2/+3
* FixVolpeon2022-12-311-5/+7
* UpdateVolpeon2022-12-312-18/+18
* Bugfixes for multi-vector token handlingVolpeon2022-12-312-22/+44
* Simplified multi-vector embedding codeVolpeon2022-12-311-12/+11
* FixesVolpeon2022-12-311-1/+1
* Added multi-vector embeddingsVolpeon2022-12-313-3/+176
* Fixed Textual InversionVolpeon2022-12-221-3/+3
* Support attention_mask of text encoderVolpeon2022-12-101-3/+7
* Make prompt processor compatible with any modelVolpeon2022-11-271-1/+2
* UpdateVolpeon2022-10-243-0/+362