75 Commits (main)

Author SHA1 Message Date
Brett Kuprel f951424e38 is_reusable 2 years ago
Brett Kuprel b55bcba4c0 removed deepcopy, delete expendable parameters after use 2 years ago
Brett Kuprel 41a44068d0 keep params in expendable mode 2 years ago
Brett Kuprel df9aa6f915 sort -> topk, prev_token_and_index -> prev_token, token_index 2 years ago
Brett Kuprel fb97ba5e20 update readme, cleanup 2 years ago
Brett Kuprel 1e18ba0ffa is_expendable argument reduces memory usage for command line script 2 years ago
Brett Kuprel d99828a239 simplified flax attention and matched torch attention 2 years ago
Brett Kuprel 61cc99c13c read tokenizer files with utf8 encoding 2 years ago
Brett Kuprel 661ec976ac simplified attention for torch model 2 years ago
Brett Kuprel ed91ab4a30 refactored to load models once and run multiple times 2 years ago
Adam Novak 28c812c832 Use all logical cores in Torch mode 2 years ago
Brett Kuprel 1fbb209623 fixed bug with cuda in detokenizer 2 years ago
Brett Kuprel 764b0bc685 cuda in detokenizer from previous commit broke colab flax model, fixed 2 years ago
Brett Kuprel 17c96fe110 works with cuda 2 years ago
Haydn Jones a3a247e6ec Fixed disabling of gradients in the torch code 2 years ago
Brett Kuprel 9d6b6dcc92 previous commit broke flax model, fixed now 2 years ago
Brett Kuprel 5aa6fe49bf use cuda if available 2 years ago
Brett Kuprel 8544f59576 use cuda if available 2 years ago
Brett Kuprel aef24ea157 torch.no_grad(), cleanup 2 years ago
Brett Kuprel 34df2b97df previous commit broke colab example, so adjusting flax requirement to 0.4.2 for now 2 years ago
Brett Kuprel 38ebe54a38 works with latest flax version 0.5.2 now 2 years ago
Brett Kuprel a014dccc05 fixed an issue with argument parser 2 years ago
Brett Kuprel e7001f063c simplified 2 years ago
Brett Kuprel 18e6a9852f license and cleanup 2 years ago
Brett Kuprel c936d26102 back to linear attention 2 years ago
Brett Kuprel 018414a5c3 fixed relative imports 2 years ago