[bug]: Certain FLUX LoRAs install, but fail to run

[rikublock]

Is there an existing issue for this problem?

• I have searched the existing issues

Operating system

Linux, Ubuntu 22.04

GPU vendor

Nvidia (CUDA)

Version number

v5.1.1

Browser

Chromium 127.0.6533.119

What happened

Successfully installed the LoRA with the model manager. Added the LoRA when generating an image with flux (quantized). The generation fails.

What you expected to happen

Successfully generate an image.

Additional context

• https://civitai.com/models/840275/d4n-oil-painting?modelVersionId=940107
  ValueError: Unsupported Linear LoRA layer type: <class 'invokeai.backend.lora.layers.lokr_layer.LoKRLayer'>
• https://civitai.com/models/850806/flux-digital-harmony-or-rendered-painting-style?modelVersionId=951900
  AssertionError in add_qkv_lora_layer_if_present: assert all(keys_present) or not any(keys_present)

Discord username

riku_block

[freelancer2000]

Seems to be the same issue due to Quantized version. Have you tried the LORA with a non quantized checkpoint?

[RyanJDick]

AssertionError in add_qkv_lora_layer_if_present: assert all(keys_present) or not any(keys_present)

This issue was fixed a couple months ago in #7313

ValueError: Unsupported Linear LoRA layer type: <class 'invokeai.backend.lora.layers.lokr_layer.LoKRLayer'>

As of v5.6.0rc4, LoKR layers work on most base FLUX models. After #7577 is merged and deployed, LoKR layers should work on all supported FLUX models.