Segmentation fault for 3.5-large-turbo

[DanaNikifor]

Danilka-Netbook:/mnt/e/stable-diffusion.cpp/build/bin$ ./sd -m /mnt/e/Models/sd3.5_large_turbo-Q4_0.gguf --cfg-scale 1 -
-steps 1 -p "masterpiece, best quality, 1girl, aqua eyes, baseball cap, blonde hair, closed mouth, earrings, green backg
round, hat, hoop earrings, jewellery, looking at viewer, shirt, short hair, simple background, solo, upper body, yellow
shirt" -n "low quality, logo on cap" -v
Option:
n_threads: 4
mode: txt2img
model_path: /mnt/e/Models/sd3.5_large_turbo-Q4_0.gguf
wtype: unspecified
clip_l_path:
clip_g_path:
t5xxl_path:
diffusion_model_path:
vae_path:
taesd_path:
esrgan_path:
controlnet_path:
embeddings_path:
stacked_id_embeddings_path:
input_id_images_path:
style ratio: 20.00
normalize input image : false
output_path: output.png
init_img:
control_image:
clip on cpu: false
controlnet cpu: false
vae decoder on cpu:false
strength(control): 0.90
prompt: masterpiece, best quality, 1girl, aqua eyes, baseball cap, blonde hair, closed mouth, earrings, green background, hat, hoop earrings, jewellery, looking at viewer, shirt, short hair, simple background, solo, upper body, yellow shirt
negative_prompt: low quality, logo on cap
min_cfg: 1.00
cfg_scale: 1.00
guidance: 3.50
clip_skip: -1
width: 512
height: 512
sample_method: euler_a
schedule: default
sample_steps: 1
strength(img2img): 0.75
rng: cuda
seed: 42
batch_count: 1
vae_tiling: false
upscale_repeats: 1
System Info:
BLAS = 0
SSE3 = 1
AVX = 0
AVX2 = 0
AVX512 = 0
AVX512_VBMI = 0
AVX512_VNNI = 0
FMA = 0
NEON = 0
ARM_FMA = 0
F16C = 0
FP16_VA = 0
WASM_SIMD = 0
VSX = 0
[DEBUG] stable-diffusion.cpp:182 - Using CPU backend
[INFO ] stable-diffusion.cpp:197 - loading model from '/mnt/e/Models/sd3.5_large_turbo-Q4_0.gguf'
[INFO ] model.cpp:801 - load /mnt/e/Models/sd3.5_large_turbo-Q4_0.gguf using gguf format
[DEBUG] model.cpp:818 - init from '/mnt/e/Models/sd3.5_large_turbo-Q4_0.gguf'
[INFO ] stable-diffusion.cpp:244 - Version: SD3.5 8B
[INFO ] stable-diffusion.cpp:275 - Weight type: q4_0
Segmentation fault

Help fix it, pls

[stduhpf]

This look oddly similar to this #442 (comment).
If you're using city96's quants, I think they are incompatible with sdcpp for some obscure reason. Try quantizing yourself, or use these instead: https://huggingface.co/stduhpf/SD3.5-Large-Turbo-GGUF-mixed-sdcpp

[DanaNikifor]

OK, I'll try to do it soon

[DanaNikifor]

No. I couldn't run your model. I also tried to run the model from ND911 https://huggingface.co/ND911/stable-diffusion-3.5-large-Turbo-GGUF/blob/main/sd3.5_large_turbo-Q4_0.gguf and also did not start

[stduhpf]

That's odd, I couldn't run City96's quant either, but I tested all of mines at least 3 times each....
Does the full model work?

[SkutteOleg]

You have to specify t5xxl, clip_l and clip_g

[DanaNikifor]

Why, if the model itself has its own clip and there is no need to add t5xxl?

[DanaNikifor]

If so, where can I download them?

[DanaNikifor]

I don't think my laptop can do this. I only have 8GB of RAM.

[stduhpf]

@DanaNikifor It should be fine, both models don't need to be in memory at the same time.

[DanaNikifor]

Danilka-Netbook:/mnt/e/Models$ ./sd -m sd3.5_large_turbo.safetensors --steps 1 -p "masterpiece, best quality, 1girl, aqu
a eyes, baseball cap, blonde hair, closed mouth, earrings, green backg round, hat, hoop earrings, jewellery, looking at
viewer, shirt, short hair, simple background, solo, upper body, yellow shirt" -n "low quality, logo on cap" -v --cfg-sca
le 1
Option:
n_threads: 4
mode: txt2img
model_path: sd3.5_large_turbo.safetensors
wtype: unspecified
clip_l_path:
clip_g_path:
t5xxl_path:
diffusion_model_path:
vae_path:
taesd_path:
esrgan_path:
controlnet_path:
embeddings_path:
stacked_id_embeddings_path:
input_id_images_path:
style ratio: 20.00
normalize input image : false
output_path: output.png
init_img:
control_image:
clip on cpu: false
controlnet cpu: false
vae decoder on cpu:false
strength(control): 0.90
prompt: masterpiece, best quality, 1girl, aqua eyes, baseball cap, blonde hair, closed mouth, earrings, green backg round, hat, hoop earrings, jewellery, looking at viewer, shirt, short hair, simple background, solo, upper body, yellow shirt
negative_prompt: low quality, logo on cap
min_cfg: 1.00
cfg_scale: 1.00
guidance: 3.50
clip_skip: -1
width: 512
height: 512
sample_method: euler_a
schedule: default
sample_steps: 1
strength(img2img): 0.75
rng: cuda
seed: 42
batch_count: 1
vae_tiling: false
upscale_repeats: 1
System Info:
BLAS = 0
SSE3 = 1
AVX = 0
AVX2 = 0
AVX512 = 0
AVX512_VBMI = 0
AVX512_VNNI = 0
FMA = 0
NEON = 0
ARM_FMA = 0
F16C = 0
FP16_VA = 0
WASM_SIMD = 0
VSX = 0
[DEBUG] stable-diffusion.cpp:182 - Using CPU backend
[INFO ] stable-diffusion.cpp:197 - loading model from 'sd3.5_large_turbo.safetensors'
[INFO ] model.cpp:804 - load sd3.5_large_turbo.safetensors using safetensors format
[DEBUG] model.cpp:872 - init from 'sd3.5_large_turbo.safetensors'
[INFO ] stable-diffusion.cpp:244 - Version: SD3.5 8B
[INFO ] stable-diffusion.cpp:275 - Weight type: f32
Segmentation fault

Why?

[SkutteOleg]

You have to specify t5xxl, clip_l and clip_g

Take a look at #445

[DanaNikifor]

Danilka-Netbook:/mnt/e/stable-diffusion.cpp/build/bin$ ./sd -m ../../../Models/sd3.5_large_turbo.safetensors --clip_l .
./../../Models/clip/clip_l.safetensors --clip_g ../../../Models/clip/clip_g.safetensors --t5xxl ../../../Models/clip/t5x
xl_fp8_e4m3fn.safetensors --steps 1 --cfg-scale 1 -p 'a lovely cat holding a sign says "Stable diffusion 3.5 Large"'
[INFO ] stable-diffusion.cpp:197 - loading model from '../../../Models/sd3.5_large_turbo.safetensors'
[INFO ] model.cpp:804 - load ../../../Models/sd3.5_large_turbo.safetensors using safetensors format
[INFO ] stable-diffusion.cpp:204 - loading clip_l from '../../../Models/clip/clip_l.safetensors'
[INFO ] model.cpp:804 - load ../../../Models/clip/clip_l.safetensors using safetensors format
[INFO ] stable-diffusion.cpp:211 - loading clip_g from '../../../Models/clip/clip_g.safetensors'
[INFO ] model.cpp:804 - load ../../../Models/clip/clip_g.safetensors using safetensors format
[INFO ] stable-diffusion.cpp:218 - loading t5xxl from '../../../Models/clip/t5xxl_fp8_e4m3fn.safetensors'
[INFO ] model.cpp:804 - load ../../../Models/clip/t5xxl_fp8_e4m3fn.safetensors using safetensors format
[INFO ] stable-diffusion.cpp:244 - Version: SD3.5 8B
[INFO ] stable-diffusion.cpp:275 - Weight type: f32
[INFO ] stable-diffusion.cpp:276 - Conditioner weight type: f16
[INFO ] stable-diffusion.cpp:277 - Diffusion model weight type: f16
[INFO ] stable-diffusion.cpp:278 - VAE weight type: f32
ggml_backend_cpu_buffer_type_alloc_buffer: failed to allocate buffer of size 9525022752
[ERROR] ggml_extend.hpp:1039 - t5 alloc params backend buffer failed, num_tensors = 219
ggml_backend_cpu_buffer_type_alloc_buffer: failed to allocate buffer of size 16532522272
[ERROR] ggml_extend.hpp:1039 - mmdit alloc params backend buffer failed, num_tensors = 923
Segmentation fault
I don't understand why this happened. There is enough RAM

[Green-Sky]

Are you running a 32bit system?

[dsignarius]

try using '--diffusion-model' instead of '-m' on your commandline if using a gguf quant of model

also if using quants, make sure you have a vae somewhere.

the two above cured my segmentation fault at that point... which was line 277/278 in current sd.cpp