0e49211a11
Load the SD3 T5xxl model in the same dtype stored in the checkpoint.
9 months ago
742d5720d1
Support zeroing out text embeddings with the attention mask.
9 months ago
56333d4850
Use the end token for the text encoder attention mask.
9 months ago
65397ce601
Replace prints with logging and add --verbose argument.
12 months ago
03c47fc0f2
Add a min_length property to tokenizer class.
1 year ago
8ac69f62e5
Make return_projected_pooled setable from the __init__
1 year ago
c2cb8e889b
Always return unprojected pooled output for gligen.
1 year ago
1cb3f6a83b
Move text projection into the CLIP model code.
...
Fix issue with not loading the SSD1B clip correctly.
1 year ago
97d03ae04a
StableCascade CLIP model support.
1 year ago
4871a36458
Cleanup some unused imports.
1 year ago
57926635e8
Switch text encoder to manual cast.
...
Use fp16 text encoder weights for CPU inference to lower memory usage.
1 year ago
9ac0b487ac
Make --gpu-only put intermediate values in GPU memory instead of cpu.
1 year ago
fbdb14d4c4
Cleaner CLIP text encoder implementation.
...
Use a simple CLIP model implementation instead of the one from
transformers.
This will allow some interesting things that would too hackish to implement
using the transformers implementation.
1 year ago
be3468ddd5
Less useless downcasting.
1 year ago
728613bb3e
Fix last pr.
1 year ago
f2e49b1d57
fix: adaptation to older versions of pytroch
1 year ago
656c0b5d90
CLIP code refactor and improvements.
...
More generic clip model class that can be used on more types of text
encoders.
Don't apply weighting algorithm when weight is 1.0
Don't compute an empty token output when it's not needed.
1 year ago
b3fcd64c6c
Make SDTokenizer class work with more types of tokenizers.
1 year ago
2a134bfab9
Fix checkpoint loader with config.
1 year ago
e60ca6929a
SD1 and SD2 clip and tokenizer code is now more similar to the SDXL one.
1 year ago
434ce25ec0
Restrict loading embeddings from embedding folders.
1 year ago
44361f6344
Support for text encoder models that need attention_mask.
1 year ago
fb3b728203
Fix issue where autocast fp32 CLIP gave different results from regular.
1 year ago
ec96f6d03a
Move text_projection to base clip model.
2 years ago
e3d0a9a490
Fix potential issue with text projection matrix multiplication.
2 years ago
00c0b2c507
Initialize text encoder to target dtype.
2 years ago
f081017c1a
Save memory by storing text encoder weights in fp16 in most situations.
...
Do inference in fp32 to make sure quality stays the exact same.
2 years ago
c99d8002f8
Make sure the pooled output stays at the EOS token with added embeddings.
2 years ago
50b1180dde
Fix CLIPSetLastLayer not reverting when removed.
2 years ago
46dc050c9f
Fix potential tensors being on different devices issues.
2 years ago
606a537090
Support SDXL embedding format with 2 CLIP.
2 years ago
608fcc2591
Fix bug with weights when prompt is long.
2 years ago
ce35d8c659
Lower latency by batching some text encoder inputs.
2 years ago
b6a60fa696
Try to keep text encoders loaded and patched to increase speed.
...
load_model_gpu() is now used with the text encoder models instead of just
the unet.
2 years ago
97ee230682
Make highvram and normalvram shift the text encoders to vram and back.
...
This is faster on big text encoder models than running it on the CPU.
2 years ago
9920367d3c
Fix embeddings not working with --gpu-only
2 years ago
20f579d91d
Add DualClipLoader to load clip models for SDXL.
...
Update LoadClip to load clip models for SDXL refiner.
2 years ago
f87ec10a97
Support base SDXL and SDXL refiner models.
...
Large refactor of the model detection and loading code.
2 years ago
f7edcfd927
Add a --gpu-only argument to keep and run everything on the GPU.
...
Make the CLIP model work on the GPU.
2 years ago
bb1f45d6e8
Properly disable weight initialization in clip models.
2 years ago
0c7cad404c
Don't initialize clip weights to default values.
2 years ago
23cf8ca7c5
Fix bug when embedding gets ignored because of mismatched size.
2 years ago
af9cc1fb6a
Search recursively in subfolders for embeddings.
2 years ago
81d1f00df3
Some refactoring: from_tokens -> encode_from_tokens
2 years ago
719c26c3c9
Merge branch 'master' of https://github.com/BlenderNeko/ComfyUI
2 years ago
d0b1b6c6bf
fixed improper padding
2 years ago
04d9bc13af
Safely load pickled embeds that don't load with weights_only=True.
2 years ago
da115bd78d
ensure backwards compat with optional args
2 years ago
752f7a162b
align behavior with old tokenize function
2 years ago
334aab05e5
Don't stop workflow if loading embedding fails.
2 years ago
comfyanonymous
Jianqi Pan
BlenderNeko