Stable Diffusion txt2image.py prompt length > 75 tokens crash

[pjain]

Using a prompt longer than 75 tokens results in SD txt2image.py failing on a Mac mini w/ M4 Pro and 48GB RAM.

"Generate an image of an alien riding a missile, set in a futuristic cityscape with sleek skyscrapers and neon lights. The alien should be tall, slender, and have scaly blue skin with large green eyes. It's wearing a metallic exosuit with glowing blue accents, and holding onto the missile's handlebars with long, spindly fingers. Incorporate a sense of speed and motion, with blurred backgrounds and dynamic lighting effects.

Style: Hyper-realistic

Resolution: 512x512 pixels

Color palette: Neon blues, purples, and silvers."

0%| | 0/2 [00:00<?, ?it/s]
Traceback (most recent call last):
File "/Users/foo/projects/mlx/mlx-examples/stable_diffusion/txt2image.py", line 71, in
for x_t in tqdm(latents, total=args.steps):
File "/Users/foo/projects/mlx/lib/python3.11/site-packages/tqdm/std.py", line 1181, in iter
for obj in iterable:
File "/Users/foo/projects/mlx/mlx-examples/stable_diffusion/stable_diffusion/init.py", line 246, in generate_latents
conditioning, pooled_conditioning = self._get_text_conditioning(
^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/Users/foo/projects/mlx/mlx-examples/stable_diffusion/stable_diffusion/init.py", line 217, in _get_text_conditioning
conditioning_1 = self.text_encoder_1(tokens_1)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/Users/foo/projects/mlx/mlx-examples/stable_diffusion/stable_diffusion/clip.py", line 94, in call
x = x + self.position_embedding.weight[:N]
~~^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
ValueError: [broadcast_shapes] Shapes (1,116,768) and (77,768) cannot be broadcast.