Jukka Seppänen be95871adc feat: Gemma4 text generation support (CORE-30) (#13376) ... * initial gemma4 support * parity with reference implementation outputs can 100% match transformers with same sdpa flags, checkpoint this and then optimize * Cleanup, video fixes * cleanup, enable fused rms norm by default * update comment * Cleanup * Update sd.py * Various fixes * Add fp8 scaled embedding support * small fixes * Translate think tokens * Fix image encoder attention mask type So it works with basic attention * Handle thinking tokens different only for Gemma4 * Code cleanup * Update nodes_textgen.py * Use embed scale class instead of buffer Slight difference to HF, but technically more accurate and simpler code * Default to fused rms_norm * Update gemma4.py		2026-05-02 22:46:15 -04:00
..
ace_lyrics_tokenizer	Initial ACE-Step model implementation. (#7972)	2025-05-07 08:33:34 -04:00
byt5_tokenizer	Support hunyuan image 2.1 regular model. (#9792)	2025-09-10 02:05:07 -04:00
hydit_clip_tokenizer	Basic hunyuan dit implementation. (#4102)	2024-07-25 18:21:08 -04:00
llama_tokenizer	Basic Hunyuan Video model support.	2024-12-16 19:35:40 -05:00
qwen25_tokenizer	Update qwen tokenizer to add qwen 3 tokens. (#11029)	2025-12-01 17:13:48 -05:00
qwen35_tokenizer	feat: Support Qwen3.5 text generation models (#12771)	2026-03-25 22:48:28 -04:00
t5_pile_tokenizer	Better tokenizing code for AuraFlow.	2024-07-12 01:15:25 -04:00
t5_tokenizer	Refactor: Move some code to the comfy/text_encoders folder.	2024-07-15 17:36:24 -04:00
ace15.py	fix(ace15): handle missing lm_metadata in memory estimation during checkpoint export #12669 (#12686)	2026-02-28 01:18:40 -05:00
ace_text_cleaners.py	Make japanese hiragana and katakana characters work with ACE. (#7997)	2025-05-08 03:32:36 -04:00
ace.py	Make japanese hiragana and katakana characters work with ACE. (#7997)	2025-05-08 03:32:36 -04:00
anima.py	Small cleanup and try to get qwen 3 work with the text gen. (#12537)	2026-02-19 22:42:28 -05:00
aura_t5.py	More flexible long clip support.	2025-04-15 10:32:21 -04:00
bert.py	P2 of qwen edit model. (#9412)	2025-08-18 22:38:34 -04:00
byt5_config_small_glyph.json	Support hunyuan image 2.1 regular model. (#9792)	2025-09-10 02:05:07 -04:00
cogvideo.py	Cogvideox (#13402)	2026-04-29 19:30:08 -04:00
cosmos.py	Fix chroma fp8 te being treated as fp16. (#11795)	2026-01-10 14:40:42 -08:00
ernie.py	Use ErnieTEModel_ not ErnieTEModel. (#13431)	2026-04-16 10:11:58 -04:00
flux.py	Implement Ernie Image model. (#13369)	2026-04-11 22:29:31 -04:00
gemma4.py	feat: Gemma4 text generation support (CORE-30) (#13376)	2026-05-02 22:46:15 -04:00
genmo.py	Fix chroma fp8 te being treated as fp16. (#11795)	2026-01-10 14:40:42 -08:00
hidream.py	Make old scaled fp8 format use the new mixed quant ops system. (#11000)	2025-12-05 14:35:42 -05:00
hunyuan_image.py	Make old scaled fp8 format use the new mixed quant ops system. (#11000)	2025-12-05 14:35:42 -05:00
hunyuan_video.py	Support loading flux 2 klein checkpoints saved with SaveCheckpoint. (#12033)	2026-01-22 18:20:48 -05:00
hydit_clip.json	Basic hunyuan dit implementation. (#4102)	2024-07-25 18:21:08 -04:00
hydit.py	Add a T5TokenizerOptions node to set options for the T5 tokenizer. (#7803)	2025-04-25 19:36:00 -04:00
jina_clip_2.py	Implement Jina CLIP v2 and NewBie dual CLIP (#11415)	2025-12-20 00:57:22 -05:00
kandinsky5.py	Fix qwen scaled fp8 not working with kandinsky. Make basic t2i wf work. (#11162)	2025-12-06 17:50:10 -08:00
llama.py	feat: Gemma4 text generation support (CORE-30) (#13376)	2026-05-02 22:46:15 -04:00
long_clipl.py	Cleanup.	2025-04-15 12:13:28 -04:00
longcat_image.py	LongCat-Image edit (#13003)	2026-03-21 23:51:05 -04:00
lt.py	feat: Gemma4 text generation support (CORE-30) (#13376)	2026-05-02 22:46:15 -04:00
lumina2.py	feat: Gemma4 text generation support (CORE-30) (#13376)	2026-05-02 22:46:15 -04:00
mt5_config_xl.json	Basic hunyuan dit implementation. (#4102)	2024-07-25 18:21:08 -04:00
newbie.py	Only apply gemma quant config to gemma model for newbie. (#11436)	2025-12-20 01:02:43 -05:00
omnigen2.py	Make old scaled fp8 format use the new mixed quant ops system. (#11000)	2025-12-05 14:35:42 -05:00
ovis.py	Fix #11963 (#11982)	2026-01-19 22:32:40 -05:00
pixart_t5.py	Fix chroma fp8 te being treated as fp16. (#11795)	2026-01-10 14:40:42 -08:00
qwen35.py	feat: Gemma4 text generation support (CORE-30) (#13376)	2026-05-02 22:46:15 -04:00
qwen_image.py	Make old scaled fp8 format use the new mixed quant ops system. (#11000)	2025-12-05 14:35:42 -05:00
qwen_vl.py	LongCat-Image edit (#13003)	2026-03-21 23:51:05 -04:00
sa_t5.py	More flexible long clip support.	2025-04-15 10:32:21 -04:00
sam3_clip.py	feat: SAM (segment anything) 3.1 support (CORE-34) (#13408)	2026-04-23 00:07:43 -04:00
sd2_clip_config.json	Fix potential issue with non clip text embeddings.	2024-07-30 14:41:13 -04:00
sd2_clip.py	More flexible long clip support.	2025-04-15 10:32:21 -04:00
sd3_clip.py	Make old scaled fp8 format use the new mixed quant ops system. (#11000)	2025-12-05 14:35:42 -05:00
spiece_tokenizer.py	feat: Add basic text generation support with native models, initially supporting Gemma3 (#12392)	2026-02-18 20:49:43 -05:00
t5_config_base.json	Refactor: Move some code to the comfy/text_encoders folder.	2024-07-15 17:36:24 -04:00
t5_config_xxl.json	Refactor: Move some code to the comfy/text_encoders folder.	2024-07-15 17:36:24 -04:00
t5_old_config_xxl.json	WIP support for Nvidia Cosmos 7B and 14B text to world (video) models.	2025-01-10 09:14:16 -05:00
t5_pile_config_xl.json	AuraFlow model implementation.	2024-07-11 16:52:26 -04:00
t5.py	P2 of qwen edit model. (#9412)	2025-08-18 22:38:34 -04:00
umt5_config_base.json	Initial ACE-Step model implementation. (#7972)	2025-05-07 08:33:34 -04:00
umt5_config_xxl.json	WIP support for Wan t2v model.	2025-02-25 17:20:35 -05:00
wan.py	Make old scaled fp8 format use the new mixed quant ops system. (#11000)	2025-12-05 14:35:42 -05:00
z_image.py	Enable embeddings for some qwen 3 models. (#12218)	2026-02-02 03:51:09 -05:00