Optimum/Neuron: RuntimeError: forward() is missing value for argument 'argument_4'

🤗Optimum
1

I’m trying to render an image on AWS Inferentia2 using Optimum NeuronX. I have converted juggernautXL_v8Rundiffusion model to the Diffusers format using sdxl-to-diffusers-v2 and then to Neuron format with:

optimum-cli export neuron \
    --model juggernautXL_v8Rundiffusion \
    --task stable-diffusion-xl \
    --batch_size 1 --num_images_per_prompt 1 \
    --height 1024 --width 1024 \
    juggernautXL_v8Rundiffusion_neuron

Then I’m trying to generate an image with this Python code:

from optimum.neuron import NeuronStableDiffusionPipeline
stable_diffusion = NeuronStableDiffusionPipeline.from_pretrained(model_path)
image = stable_diffusion(prompt).images[0]

But it fails with:

RuntimeError: forward() is missing value for argument 'argument_4'. 
Declaration: forward(__torch__.torch_neuronx.xla_impl.trace.___torch_mangle_11.NeuronModule self, 
Tensor argument_1, Tensor argument_2, Tensor argument_3, Tensor argument_4, Tensor argument_5)
-> ((Tensor))

Any idea why? Do I need any other parameters other than prompt in the inference step?

2

Figured it out. It’s a SDXL model so needed to use NeuronStableDiffusionXLPipeline instead of NeuronStableDiffusionPipeline.

Just in case someone comes across this too.

3

This topic was automatically closed 12 hours after the last reply. New replies are no longer allowed.