This is a fast, minimal port of Boris Dayma's [DALL·E Mega](https://github.com/borisdayma/dalle-mini). It has been stripped down for inference and converted to PyTorch. The only third party dependencies are numpy, requests, pillow and torch.
The required models will be downloaded to `models_root` if they are not already there. Set the `dtype` to `torch.float16` to save GPU memory. If you have an Ampere architecture GPU you can use `torch.bfloat16`. Once everything has finished initializing, call `generate_image` with some text as many times as you want. Use a positive `seed` for reproducible results. Higher values for `log2_supercondition_factor` result in better agreement with the text but a narrower variety of generated images. Every image token is sampled from the top-$k$ most probable tokens.
The images can also be generated as a `FloatTensor` in case you want to process them manually (e.g. save individual images).
```python
images = model.generate_images(
text='Nuclear explosion broccoli',
seed=-1,
image_count=7,
log2_k=6,
log2_supercondition_factor=5,
is_verbose=False
)
```
Note: you will have to move the images to the cpu and convert to numpy, e.g. `images = images.to('cpu').detach().numpy()`. Then image $i$ can be coverted to a PIL.Image `image = Image.fromarray(images[i])`, and saved with its `save` method `image.save('image.png')`.
If the model is being used interactively (e.g. in a notebook) `generate_image_stream` can be used to generate a stream of images as the model is decoding. The detokenizer adds a slight delay for each image. Setting `log2_mid_count` to 3 results in a total of `2 ** 3 = 8` generated images. The only valid values for `log2_mid_count` are 0, 1, 2, 3, and 4. This is implemented in the colab.
