GitLab

The modification replaces this method with a token-based approach using a tokenizer. This tokenizes the entire prompt, identifies the token corresponding to the trigger word, removes it, and then decodes the modified tokens back into text.

Fixing the google colab demo notebook, ready to use now (on A100, the code is not V100 compatible) (#39)

version 3 blows up
l the latest PEFT and transformers packages in the future.
  deprecate("fuse_text_encoder_lora", "0.27", LORA_DEPRECATION_MESSAGE)
Traceback (most recent call last):
  File "/media/2TB/PhotoMaker/gradio_demo/app.py", line 285, in <module>
    demo.launch()
  File "/home/oem/miniconda3/envs/torch2/lib/python3.10/site-packages/gradio/blocks.py", line 1996, in launch
    self.validate_queue_settings()
  File "/home/oem/miniconda3/envs/torch2/lib/python3.10/site-packages/gradio/blocks.py", line 1821, in validate_queue_settings
    raise ValueError("Progress tracking requires queuing to be enabled.")
ValueError: Progress tracking requires queuing to be enabled.

fidelty -> fidelity

* support MPS

* Update README.md

* Make requirements more clearly

* resolve conflict

* Update: support MacOS

---------
Co-authored-by: Zhen Li <zhenli1031@gmail.com>

* feat: add cog configuration for replicate deployment

* fix: clean up

* fix: clean up the configuration

* fix: download the weights with pget

* fix: download with huggingface

* fix: clean up

* fix: add missing bits

* fix: fix loading the image

* fix: default num_outputs to 1