Commit Graph

60 Commits

Author SHA1 Message Date
Brett Kuprel
c199507a7a vqgan needs to be float32 2022-07-07 08:53:27 -04:00
Brett Kuprel
da62298f06 support bfloat16 2022-07-07 08:21:20 -04:00
Brett Kuprel
f071b31bdd properly limit input to 64 tokens 2022-07-05 22:14:19 -04:00
Brett Kuprel
89a125b4b9 control top_k value 2022-07-05 17:23:05 -04:00
Brett Kuprel
ba6080668f update readme 2022-07-05 07:44:42 -04:00
Brett Kuprel
9a4d9dc118 remove unused imports 2022-07-05 06:31:01 -04:00
Brett Kuprel
ccdcbc7d46 forgot missing 2** 2022-07-04 23:29:48 -04:00
Brett Kuprel
6d81d514a0 control super condition factor 2022-07-04 21:30:27 -04:00
Brett Kuprel
5f4815775b generate_image_stream 2022-07-04 20:02:33 -04:00
Brett Kuprel
97d22f0fdd handle log2_mid_count=0 2022-07-04 17:46:31 -04:00
Brett Kuprel
1702d3c439 log2_mid_count 2022-07-04 17:27:23 -04:00
Brett Kuprel
0d9998926d display intermediate images 2022-07-04 16:06:49 -04:00
Brett Kuprel
deefd24919 decode_row 2022-07-04 09:58:00 -04:00
Brett Kuprel
6f617fe98f inplace attention state, faster and less memory 2022-07-04 09:14:37 -04:00
Brett Kuprel
377d15cb16 faster decoder self attention 2022-07-04 08:05:55 -04:00
Brett Kuprel
256fd1aab0 empty cache if cuda available 2022-07-04 07:21:54 -04:00
w4ffl35
7d549505da Fixes #16 - mega model running out of memory
Prior to this fix the mega model will (more often than not) fail when running in succession.

Clearing the cache seems to fix the issue.
2022-07-04 00:26:17 -06:00
Brett Kuprel
313635e914 fixed typing error for older python versions 2022-07-02 09:06:22 -04:00
Brett Kuprel
1eb56737d8 added grid_size parameter to generate a grid of images 2022-07-02 08:45:49 -04:00
Brett Kuprel
cf9656baa2 added is_verbose flag 2022-07-01 20:17:20 -04:00
Brett Kuprel
35e97768a5 v0.2.0, MinDalleTorch -> MinDalle, breaking change 2022-07-01 19:44:24 -04:00
Brett Kuprel
fff44d683e update readme, random seed if none specified 2022-07-01 18:50:11 -04:00
Brett Kuprel
be2beca7c0 added to pypi 2022-07-01 18:16:55 -04:00
Brett Kuprel
69683fec94 simplify import 2022-07-01 16:02:05 -04:00
Brett Kuprel
18c72ed34d simplified MinDalleTorch 2022-07-01 15:53:39 -04:00
Brett Kuprel
07ce93d5f8 moved flax model and conversion code to separate repository 2022-07-01 14:06:50 -04:00
Brett Kuprel
85f5866eff remove config.json dependency, default to torch in image_from_text.py 2022-07-01 12:03:37 -04:00
Brett Kuprel
09a0f85b8e separate setup processes for flax and torch 2022-07-01 11:08:33 -04:00
Brett Kuprel
7bf76deafb fixed wrong file path 2022-07-01 10:58:29 -04:00
Brett Kuprel
e4c2be54cb save converted detokenizer params 2022-07-01 10:17:29 -04:00
Brett Kuprel
b40fd83a0d mega works with latest flax version 0.5.2 now, removing 0.4.2 pin 2022-07-01 02:58:43 -04:00
Brett Kuprel
08b158d580 updated readme 2022-06-30 16:50:04 -04:00
Brett Kuprel
2311a1af7b delete cache 2022-06-30 15:48:20 -04:00
Brett Kuprel
b913b58353 pre converting params to torch allows mega to run in standard colab runtime 2022-06-30 14:54:08 -04:00
Brett Kuprel
c2a3858c96 delete params sooner 2022-06-30 11:44:36 -04:00
Brett Kuprel
f951424e38 is_reusable 2022-06-30 11:25:24 -04:00
Brett Kuprel
b55bcba4c0 removed deepcopy, delete expendable parameters after use 2022-06-30 11:09:09 -04:00
Brett Kuprel
41a44068d0 keep params in expendable mode 2022-06-30 09:36:32 -04:00
Brett Kuprel
df9aa6f915 sort -> topk, prev_token_and_index -> prev_token, token_index 2022-06-30 09:04:11 -04:00
Brett Kuprel
fb97ba5e20 update readme, cleanup 2022-06-30 07:41:31 -04:00
Brett Kuprel
1e18ba0ffa is_expendable argument reduces memory usage for command line script 2022-06-30 06:43:10 -04:00
Brett Kuprel
d99828a239 simplified flax attention and matched torch attention 2022-06-29 14:56:28 -04:00
Brett Kuprel
61cc99c13c read tokenizer files with utf8 encoding 2022-06-29 14:18:23 -04:00
Brett Kuprel
661ec976ac simplified attention for torch model 2022-06-29 13:48:12 -04:00
Brett Kuprel
ed91ab4a30 refactored to load models once and run multiple times 2022-06-29 09:42:12 -04:00
Adam Novak
28c812c832 Use all logical cores in Torch mode 2022-06-28 22:26:51 -04:00
Brett Kuprel
1fbb209623 fixed bug with cuda in detokenizer 2022-06-28 22:02:35 -04:00
Brett Kuprel
764b0bc685 cuda in detokenizer from previous commit broke colab flax model, fixed 2022-06-28 21:36:48 -04:00
Brett Kuprel
17c96fe110 works with cuda 2022-06-28 21:28:36 -04:00
Brett Kuprel
9d6b6dcc92 previous commit broke flax model, fixed now 2022-06-28 12:54:58 -04:00