Commit Graph
136 Commits
Author SHA1 Message Date
kijai a549df730f better caption check 2024-09-16 10:10:35 +03:00
kijai 0419754233 Update train_network.py 2024-09-14 09:21:17 +03:00
kijai 3274cd0574 Update flux_lora_train_example01.json 2024-09-12 11:24:18 +03:00
kijai 608073d244 Update lora_flux.py 2024-09-11 02:19:20 +03:00
kijai 94ae756b5e add ability to set clip and T5 LR separately 2024-09-11 02:06:24 +03:00
kijai 753f617f26 save workflow as json first thing as well, allow block ranges for block select 2024-09-10 17:05:18 +03:00
kijai 47c62fd8c9 Update nodes.py 2024-09-10 01:06:54 +03:00
kijai 09a8d7ffca Update nodes.py 2024-09-09 22:52:01 +03:00
kijai c3aa4ea889 fix cpu_offload_checkpointing 2024-09-09 20:27:31 +03:00
kijai b52849ad9a Update nodes.py 2024-09-09 02:52:15 +03:00
kijai 4cf65f66e2 fix validation sampling when training clip_l and caching 2024-09-09 01:25:33 +03:00
kijai 1d2e4c225f bugfix 2024-09-09 00:36:21 +03:00
kijai 2c41f44ef0 Merge branch 'train_t5' 2024-09-08 23:31:29 +03:00
kijai 54d8a4b60d fix example 2024-09-08 23:08:57 +03:00
kijai c61fcf09cf update examples 2024-09-08 23:06:59 +03:00
kijai 66a5fa96c9 Update nodes.py 2024-09-08 17:48:02 +03:00
kijai 284bb6f112 add lora path to load network_weights in the model select node 2024-09-08 16:20:40 +03:00
kijai 95de0cbcc9 allow loading models saved in Comfy 2024-09-08 15:57:45 +03:00
kijai 280f2557b9 Update nodes.py 2024-09-08 13:11:01 +03:00
kijai 299a994197 Update nodes.py 2024-09-08 02:56:30 +03:00
kijai 7f486c8916 add seed widget to help refresh dataset between runs 2024-09-08 02:22:55 +03:00
kijai 712ffb7a1b fix test sampler 2024-09-07 23:59:13 +03:00
kijai 1c70cce963 support fp8 e5m2 2024-09-07 18:27:02 +03:00
kijai cd8e756103 pbar fixes 2024-09-07 02:41:09 +03:00
kijai c6de455381 block select 2024-09-07 02:15:42 +03:00
kijai 98ca556b23 Update nodes.py 2024-09-06 21:47:51 +03:00
Kijai 0a720115de update 2024-09-06 20:43:40 +03:00
kijai 77587e0193 cpu_offload_checkpointing 2024-09-05 16:17:11 +03:00
kijai 0d1d93931f cpu_offload_checkpointing 2024-09-05 15:58:17 +03:00
kijai ba58d55a6c create comfy lora dir if it doesn't exist 2024-09-05 04:19:34 +03:00
kijai 201891ef88 Update nodes.py 2024-09-05 04:18:56 +03:00
kijai a4d2f7545b Update flux_train_utils.py 2024-09-04 19:42:30 +03:00
kijai 721dfc94f3 Update lora_flux.py 2024-09-04 18:59:23 +03:00
kijai 38d3d9ddf6 initial T5 training 2024-09-04 18:50:27 +03:00
kijai 638076f337 Update requirements.txt 2024-09-04 13:57:53 +03:00
kijai 372787dfb6 Add Lion optimizers 2024-09-02 22:33:16 +03:00
kijai 16464ff1ba fix flux_shift 2024-09-01 18:46:45 +03:00
kijai 19a77d835f Use relative nework module import to avoid conflicts 2024-09-01 18:42:18 +03:00
kijai 8d19121517 possible resume fix 2024-09-01 16:19:17 +03:00
kijai 8f6a90e5a9 add flux_shift 2024-09-01 13:37:13 +03:00
kijai 7095626d6a fix the ability to disable fp8_base 2024-08-31 16:50:18 +03:00
kijai 2e745a2612 Update requirements.txt 2024-08-30 16:08:30 +03:00
Jukka Seppänen 72959ea936 Update README.md 2024-08-30 12:53:31 +03:00
kijai 816f0ec75f fix bug with dataset adding stacking
removing a dataset node still won't remove added dataset
2024-08-30 03:11:15 +03:00
kijai aba4a61024 Update train_util.py 2024-08-30 01:44:22 +03:00
kijai fc1f0e59e2 better error for invalid dataset path 2024-08-30 01:36:52 +03:00
kijai 4daa4181e7 Update nodes.py 2024-08-29 13:02:48 +03:00
kijai 66749c8772 clip training 2024-08-29 12:48:33 +03:00
Jukka Seppänen 4b48695ad4 Merge pull request #3 from ComfyNodePRs/publish
Add Github Action for Publishing to Comfy Registry
2024-08-28 19:29:05 +03:00
kijai e9beacd83c update workflow 2024-08-28 19:26:48 +03:00