Refactor load_tokenizer function to include error handling and device optimizations; streamline model loading process and improve memory management

Enable bf16 in load_infinity function and enhance transform function with type hints and error handling; refactor joint_vi_vae_encode_decode for improved performance and error management

Set CUDA launch blocking environment variable for improved debugging

Add CUDA availability check for input IDs and attention mask in prompt encoding

Refactor image generation process by removing gen_one_img function; streamline prompt encoding and inference steps

Fix indentation in load_infinity function for better readability

Remove unused image enhancement functions to streamline image processing

Enhance Gradio interface with detailed instructions and a custom HTML header for the Infinity Image Generator

Fixes

Refactor encode_prompt function by removing enable_positive_prompt parameter and related logic; streamline image generation process

Make the generator CPU

Refactoring

Fix json issue

Refactoring

Initialize to CPU if no GPU

Remove cache_enabled parameter from autocast context in load_infinity function for cleaner implementation

Remove device initialization and logging from load_transformer function; set device parameter to None for flexibility

Make device optional in load_infinity function; set default to 'cuda' or 'cpu' based on availability and adjust autocast dtype handling

Enhance Infinity model loading by specifying device type in autocast context for improved performance

Optimize Infinity model loading by clearing CUDA cache and adjusting device assignment; remove redundant calls

Add device and dtype logging in app.py for better debugging

Refine prompt enhancement guidelines in app.py to focus on visual and sensory features; clarify output structure and examples

Add prompt enhancement functionality and integrate Gradio client in app.py; update requirements.txt

Update .gitignore to include weights folder and .pypirc

Refactor app.py to improve UI layout and rename weight download function; update import path for AutoEncoder in vae.py

Remove redundant text encoder model download and update tokenizer path in app.py

Refactor import statement for consistency in utils/misc.py

Install flash-attn package with subprocess in app.py

Refactor import paths and update requirements to include timm

Add torchvision to requirements.txt

Remove unused dependencies from requirements.txt

Add initial project structure with requirements and utility functions

initial commit