Segmentation fault during prompt expansion on AMD GPU

[sourajit02]

Describe the problem
Trying to generate an image leads to a segmentation fault.
GPU: AMD 7900XTX
Installation method: using python venv
Running a debugger shows that the segmentation fault occurs in modules/expansion.py when trying to execute model.generate() on the AutoModelForCausalLM model:

Full Console Log
(fooocus_env) ~ > python .apps/Fooocus/entry_with_update.py
Already up-to-date
Update succeeded.
[System ARGV] ['.apps/Fooocus/entry_with_update.py']
Python 3.11.6 (main, Nov 14 2023, 09:36:21) [GCC 13.2.1 20230801]
Fooocus version: 2.1.824
Running on local URL: http://127.0.0.1:7865

To create a public link, set share=True in launch().
Total VRAM 24560 MB, total RAM 63484 MB
Set vram state to: NORMAL_VRAM
Disabling smart memory management
Device: cuda:0 AMD Radeon RX 7900 XTX : native
VAE dtype: torch.float32
Using sub quadratic optimization for cross attention, if you have memory or speed issues try using: --use-split-cross-attention
13:56:59 INFO: Opening in existing instance
Refiner unloaded.
model_type EPS
adm 2816
Using split attention in VAE
Working with z of shape (1, 4, 32, 32) = 4096 dimensions.
Using split attention in VAE
extra keys {'cond_stage_model.clip_l.text_projection', 'cond_stage_model.clip_g.transformer.text_model.embeddings.position_ids', 'cond_stage_model.clip_l.logit_scale'}
Base model loaded: /home/s/.apps/Fooocus/models/checkpoints/juggernautXL_version6Rundiffusion.safetensors
Request to load LoRAs [['sd_xl_offset_example-lora_1.0.safetensors', 0.1], ['None', 1.0], ['None', 1.0], ['None', 1.0], ['None', 1.0]] for model [/home/s/.apps/Fooocus/models/checkpoints/juggernautXL_version6Rundiffusion.safetensors].
Loaded LoRA [/home/s/.apps/Fooocus/models/loras/sd_xl_offset_example-lora_1.0.safetensors] for UNet [/home/s/.apps/Fooocus/models/checkpoints/juggernautXL_version6Rundiffusion.safetensors] with 788 keys at weight 0.1.
Fooocus V2 Expansion: Vocab with 642 words.
Fooocus Expansion engine loaded for cuda:0, use_fp16 = True.
Requested to load SDXLClipModel
Requested to load GPT2LMHeadModel
Loading 2 new models
[Fooocus Model Management] Moving model(s) has taken 0.47 seconds
App started successful. Use the app with http://127.0.0.1:7865/ or 127.0.0.1:7865
[Parameters] Adaptive CFG = 7
[Parameters] Sharpness = 2
[Parameters] ADM Scale = 1.5 : 0.8 : 0.3
[Parameters] CFG = 4.0
[Parameters] Seed = 752634906692628473
[Parameters] Sampler = dpmpp_2m_sde_gpu - karras
[Parameters] Steps = 30 - 15
[Fooocus] Initializing ...
[Fooocus] Loading models ...
Refiner unloaded.
[Fooocus] Processing prompts ...
[Fooocus] Preparing Fooocus text #1 ...
zsh: segmentation fault (core dumped) python .apps/Fooocus/entry_with_update.py

[sourajit02]

EDIT: I got it to work. Steps taken:

Install latest versions of torch, torchvision and torchaudio from rocm:
pip install --pre torch torchvision torchaudio --index-url https://download.pytorch.org/whl/nightly/rocm5.7

Run with environment variable HSA_OVERRIDE_GFX_VERSION=11.0.0

Most likely image generation will still work with rocm5.6, the envvar should be the actual fix.