20 Commits
Author SHA1 Message Date
wassname 6b46f97191 fix 2019-12-25 08:25:37 +08:00
wassname 11d54e21cf fixes 2019-12-25 08:09:35 +08:00
wassname 2cc4b96807 fixes 2019-12-25 08:08:48 +08:00
wassname 291f4c0dd0 use less GPU ram on startup 2019-12-25 08:02:01 +08:00
wassname 7b95fa7e79 better truncation of context 2019-12-25 07:39:59 +08:00
wassname 90fdedcfc2 avoid loop 2019-12-25 07:10:57 +08:00
wassname 57432290ea add stop token to speed up 2019-12-25 07:10:23 +08:00
wassname 69d1151e86 remove endoftext etc 2019-12-25 06:39:20 +08:00
AccidentallyOnPurpose 4050e5b2a9 comments 2019-12-24 23:41:51 +08:00
AccidentallyOnPurpose d43836b83d maybe fix the hangs 2019-12-24 23:32:41 +08:00
AccidentallyOnPurpose d86ced8a0e fix 2019-12-24 20:51:35 +08:00
AccidentallyOnPurpose 9dfeca88c1 crop tokens when too long 2019-12-24 19:34:23 +08:00
AccidentallyOnPurpose 22a0025320 oops 2019-12-24 02:20:47 +00:00
AccidentallyOnPurpose 5daf905329 use past hidden states, potentially faster on cpu 2019-12-24 08:07:17 +08:00
AccidentallyOnPurpose 8c7ac35313 this is right 2019-12-23 22:45:53 +08:00
AccidentallyOnPurpose 9201ec5cb6 missing a not, was forcing cpu 2019-12-23 22:35:30 +08:00
wassname 7489b69955 more informative message if file is missing 2019-12-23 22:29:49 +08:00
cloveranon afd520f9a9 uses gpu by default 2019-12-23 04:15:20 -05:00
cloveranon 2a4efc3d88 Finally merged changes to a working state, move configparser and logging code to new file, and added config option to switch between cpu 2019-12-23 03:21:26 -05:00
cloveranon 646b589bee Manually merged in most of the changes from AccidentallyOnPurpose who changed the model to pytorch added suggested actions, a repetition penalty, and a number of other things. 2019-12-23 00:17:56 -05:00