be595a27be
Split model name correctly
...
`String.split(..., 2)` will split the string at most 2 times, resulting in a 3-tuple. Presumably the intent here was to have a 2-tuple at most, meaning the string gets split at most once.
6 months ago
da516f48ad
Update CACHE_URL_BASE
1 year ago
2cf03aaf6e
change dribble to dribbble ( #69 )
2 years ago
ce9d271aa1
Support for many different caption models:
...
blip-base, blip-large, blip2-2.7b, blip2-flan-t5-xl, git-large-coco
2 years ago
ac74904908
Expose LabelTable and load_list and give example in README how they can be used to rank your own list of terms.
2 years ago
a0f278c4a9
Move definition of clip_model_name ( #52 )
2 years ago
08546eae22
Properly sync up all the version numbers
2 years ago
384e234ba2
Minor fix to BLIP offloading
2 years ago
c4e16359a7
More safetensor, download, and VRAM improvements
2 years ago
6a62ce73e8
0.5.1
2 years ago
bf7404d7fa
When blip_offload enabled keep BLIP model on CPU to start.
2 years ago
9f04c8550c
add the negative cache url
2 years ago
ae88b07a65
safetensors!
...
- store cached embeddings in safetensor format
- updated huggingface ci-preprocess repo
- bumped version to 0.5.0
2 years ago
bcf1833ae0
0.4.4 use blip-ci instead of blip-vit
2 years ago
93db86fa70
.
2 years ago
78287e17e1
0.4.2:
...
- upgrade chain to take a min_count parameter so it won't early out until it has considered at least min_count flavors
- interrogate method ("best" mode) also checks against classic and fast to use their output if it's better
- fix bug of config.download_cache option not being used!
- add notes on Config object to readme
2 years ago
71c77633a6
.
2 years ago
42b3cf4d9e
Bunch of updates! ( #40 )
...
- auto download the cache files from huggingface
- experimental negative prompt mode
- slight quality and performance improvement to best mode
- analyze tab in Colab and run_gradio to get table of ranked terms
2 years ago
99c8d45e86
Make the BLIP model configurable, can set config.blip_model_type now to 'base' or 'large'
2 years ago
65c560ffac
Reuse existing blip-vit package on pypi
2 years ago
55fe80c74c
Simplify install further
...
Big ups to @justindujardin for proper syntax to get git dependency into requirements in way that pip will accept it! :D
2 years ago
1ec6cd9d45
Update to nicer BLIP packaging
2 years ago
6f17fb09af
Fix for running on CPU
2 years ago
152d5f551f
0.3.1 fix for running on cpu, update readme usage instructions
2 years ago
0a0b3968d1
Bump version to 0.3.0 for pip
2 years ago
8123696883
Handle exception trying to load cached table
2 years ago
884aab1a26
Default to ViT-L, lower intermediate count for Colab with ViT-H
2 years ago
5aed16b011
Ability to swap CLIP models (takes about 5s for ViTL and 10s for ViTH), update Replicate cog
2 years ago
917b7c6c15
More fixes, improvement and cleanup.
2 years ago
1221871c1b
Handle differences in how open_clip does prompt truncation, run_gradio support for all the open_clip models and --share option.
2 years ago
4bef32e69b
Shuffle BLIP back to system RAM to help with 16GB Colab
2 years ago
429d490901
First test version with OpenCLIP and ViTH!
2 years ago
55c922a48a
Update notebook batch processing with option to rename files so can be used with [filewords] in Dreambooth!
...
- new `quiet` config option so CLIP Interrogator doesn't print and tqdm
- `max_flavors` option to each interrogate method
2 years ago
8f5ddce2b3
.
2 years ago
b62cca2097
Gradio version plus classic and fast modes
2 years ago
9ce0f68ab3
CLIPInterrogator -> Interrogator
2 years ago
7a2ac9aa57
Add to pip!
2 years ago
1b5f9437bd
Make into a reusable library
2 years ago
Doxin
pharmapsychotic
Harry Wang
starovoitovs
pharmapsychotic