A recursive transformer version of the Anima 2b model (relaxed variant, per-recursion LoRA rank 64, currently increased to 256 via SVD), trained in pure fp16 mode on two T4 GPUs. The number of recursion passes is currently only 2, with plans to increase it to 3–4. At the moment, the vibecoded slop implementation of the model is located in a fork of kohya-ss sd-scripts here:...
Model source
Source description
Sources & provenance
1 active sourcehighest_eligible_score
Verified Jul 11
Source evidence
3 excerptslicense: other basemodel: circlestone-labs/Anima A recursive transformer version of the Anima 2b model (relaxed variant, per-recursion LoRA rank 64, currently increased to 256 via SVD), trained in pure fp16 mode on two T4 GPUs. The number of recursion passes…
A recursive transformer version of the Anima 2b model (relaxed variant, per-recursion LoRA rank 64, currently increased to 256 via SVD), trained in pure fp16 mode on two T4 GPUs. The number of recursion passes is currently only 2, with plans to increase it…
femboysLover/anima-looped-test-lowlr4