[Bug] Illegal instruction (core dumped)

[Gzpch97]

Git commit

none
https://github.com/leejet/stable-diffusion.cpp/releases/download/master-535-84cbd88/sd-master-84cbd88-bin-Linux-Ubuntu-24.04-x86_64-vulkan.zip

Operating System & Version

Debian GNU/Linux forky/sid (forky) x86_64

GGML backends

Vulkan

Command-line arguments used

./sd-cli -v -m ./v1-5-pruned-emaonly.safetensors -p "a lovely cat"

Steps to reproduce

ran ./sd-cli -v -m ./v1-5-pruned-emaonly.safetensors -p "a lovely cat" in Konsole at the program directory

What you expected to happen

for an image to be generated

What actually happened

program returned

Illegal instruction (core dumped) ./sd-cli -v -m ./v1-5-pruned-emaonly.safetensors -p "a lovely cat"

after loading the model, nothing generated

Logs / error messages / stack trace

[DEBUG] main.cpp:515 - version: stable-diffusion.cpp version unknown, commit 5792c66
[DEBUG] main.cpp:516 - System Info:
SSE3 = 1 | AVX = 1 | AVX2 = 1 | AVX512 = 0 | AVX512_VBMI = 0 | AVX512_VNNI = 0 | FMA = 1 | NEON = 0 | ARM_FMA = 0 | F16C = 1 | FP16_VA = 0 | WASM_SIMD = 0 | VSX = 0 |
[DEBUG] main.cpp:517 - SDCliParams {
mode: img_gen,
output_path: "output.png",
verbose: true,
color: false,
canny_preprocess: false,
convert_name: false,
preview_method: none,
preview_interval: 1,
preview_path: "preview.png",
preview_fps: 16,
taesd_preview: false,
preview_noisy: false
}
[DEBUG] main.cpp:518 - SDContextParams {
n_threads: 4,
model_path: "./v1-5-pruned-emaonly.safetensors",
clip_l_path: "",
clip_g_path: "",
clip_vision_path: "",
t5xxl_path: "",
llm_path: "",
llm_vision_path: "",
diffusion_model_path: "",
high_noise_diffusion_model_path: "",
vae_path: "",
taesd_path: "",
esrgan_path: "",
control_net_path: "",
embedding_dir: "",
embeddings: {
}
wtype: NONE,
tensor_type_rules: "",
lora_model_dir: ".",
photo_maker_path: "",
rng_type: cuda,
sampler_rng_type: NONE,
offload_params_to_cpu: false,
enable_mmap: false,
control_net_cpu: false,
clip_on_cpu: false,
vae_on_cpu: false,
flash_attn: false,
diffusion_flash_attn: false,
diffusion_conv_direct: false,
vae_conv_direct: false,
circular: false,
circular_x: false,
circular_y: false,
chroma_use_dit_mask: true,
qwen_image_zero_cond_t: false,
chroma_use_t5_mask: false,
chroma_t5_mask_pad: 1,
prediction: NONE,
lora_apply_mode: auto,
vae_tiling_params: { 0, 0, 0, 0.5, 0, 0 },
force_sdxl_vae_conv_scale: false
}
[DEBUG] main.cpp:519 - SDGenerationParams {
loras: "{
}",
high_noise_loras: "{
}",
prompt: "a lovely cat",
negative_prompt: "",
clip_skip: -1,
width: -1,
height: -1,
batch_count: 1,
init_image_path: "",
end_image_path: "",
mask_image_path: "",
control_image_path: "",
ref_image_paths: [],
control_video_path: "",
auto_resize_ref_image: true,
increase_ref_index: false,
pm_id_images_dir: "",
pm_id_embed_path: "",
pm_style_strength: 20,
skip_layers: [7, 8, 9],
sample_params: (txt_cfg: 7.00, img_cfg: 7.00, distilled_guidance: 3.50, slg.layer_count: 3, slg.layer_start: 0.01, slg.layer_end: 0.20, slg.scale: 0.00, scheduler: NONE, sample_method: NONE, sample_steps: 20, eta: 0.00, shifted_timestep: 0, flow_shift: inf),
high_noise_skip_layers: [7, 8, 9],
high_noise_sample_params: (txt_cfg: 7.00, img_cfg: 7.00, distilled_guidance: 3.50, slg.layer_count: 3, slg.layer_start: 0.01, slg.layer_end: 0.20, slg.scale: 0.00, scheduler: NONE, sample_method: NONE, sample_steps: 20, eta: 0.00, shifted_timestep: 0, flow_shift: inf),
custom_sigmas: [],
cache_mode: "",
cache_option: "",
cache: disabled (threshold=1, start=0.15, end=0.95),
moe_boundary: 0.875,
video_frames: 1,
fps: 16,
vace_strength: 1,
strength: 0.75,
control_strength: 0.9,
seed: 42,
upscale_repeats: 1,
upscale_tile_size: 128,
}
[DEBUG] stable-diffusion.cpp:177 - Using Vulkan backend
[DEBUG] ggml_extend.hpp:75 - ggml_vulkan: Found 1 Vulkan devices:
[DEBUG] ggml_extend.hpp:75 - ggml_vulkan: 0 = AMD Radeon RX 580 Series (RADV POLARIS10) (radv) | uma: 0 | fp16: 0 | bf16: 0 | warp size: 64 | shared memory: 65536 | int dot: 0 | matrix cores: none
[INFO ] stable-diffusion.cpp:198 - Vulkan: Using device 0
[INFO ] stable-diffusion.cpp:256 - loading model from './v1-5-pruned-emaonly.safetensors'
[INFO ] model.cpp:369 - load ./v1-5-pruned-emaonly.safetensors using safetensors format
[DEBUG] model.cpp:503 - init from './v1-5-pruned-emaonly.safetensors', prefix = ''
[INFO ] stable-diffusion.cpp:341 - Version: SD 1.x
[INFO ] stable-diffusion.cpp:369 - Weight type stat: f32: 1131
[INFO ] stable-diffusion.cpp:370 - Conditioner weight type stat: f32: 196
[INFO ] stable-diffusion.cpp:371 - Diffusion model weight type stat: f32: 686
[INFO ] stable-diffusion.cpp:372 - VAE weight type stat: f32: 248
[DEBUG] stable-diffusion.cpp:374 - ggml tensor size = 400 bytes
[DEBUG] clip.hpp:161 - vocab size: 49408
[DEBUG] clip.hpp:172 - trigger word img already in vocab
[DEBUG] ggml_extend.hpp:1949 - clip params backend buffer size = 469.44 MB(VRAM) (196 tensors)
[DEBUG] ggml_extend.hpp:1949 - unet params backend buffer size = 2155.33 MB(VRAM) (686 tensors)
[DEBUG] ggml_extend.hpp:1949 - vae params backend buffer size = 94.47 MB(VRAM) (140 tensors)
[DEBUG] stable-diffusion.cpp:781 - loading weights
[DEBUG] model.cpp:1350 - using 4 threads for model loading
[DEBUG] model.cpp:1372 - loading tensors from ./v1-5-pruned-emaonly.safetensors
|==================================================| 1131/1131 - 95.11it/s
[INFO ] model.cpp:1592 - loading tensors completed, taking 11.90s (process: 0.00s, read: 8.09s, memcpy: 0.00s, convert: 1.00s, copy_to_backend: 1.17s)
[DEBUG] stable-diffusion.cpp:816 - finished loaded file
[INFO ] stable-diffusion.cpp:874 - total params memory size = 2719.24MB (VRAM 2719.24MB, RAM 0.00MB): text_encoders 469.44MB(VRAM), diffusion_model 2155.33MB(VRAM), vae 94.47MB(VRAM), controlnet 0.00MB(VRAM), pmid 0.00MB(VRAM)
[INFO ] stable-diffusion.cpp:944 - running in eps-prediction mode
[DEBUG] stable-diffusion.cpp:3528 - generate_image 512x512
[INFO ] stable-diffusion.cpp:3564 - sampling using Euler A method
[INFO ] denoiser.hpp:494 - get_sigmas with discrete scheduler
[INFO ] stable-diffusion.cpp:3691 - TXT2IMG
Illegal instruction (core dumped) ./sd-cli -v -m ./v1-5-pruned-emaonly.safetensors -p "a lovely cat"

Additional context / environment details

OS: Debian GNU/Linux forky/sid (forky) x86_64
Kernel: Linux 6.18.15+deb14-amd64
Shell: bash 5.3.9
DE: KDE Plasma 6.5.4
WM: KWin (X11)
CPU: AMD A10-7860K Radeon R7, 12 Compute Cores 4C+8G (4) @ 3.60 GHz
GPU: AMD Radeon RX 580 Series [Discrete]
Memory: 5.95 GiB / 7.71 GiB (77%)

mesa-vulkan-drivers/testing,now 26.0.0-1 amd64 [installed]
mesa-vulkan-drivers/testing,now 26.0.0-1 i386 [installed,automatic]

KoboldCpp does work, i tried it with the latest release

[wbruna]

This release (84cbd88):

https://github.com/leejet/stable-diffusion.cpp/releases/download/master-535-84cbd88/sd-master-84cbd88-bin-Linux-Ubuntu-24.04-x86_64-vulkan.zip

does not match the one on the log (5792c66):

[DEBUG] main.cpp:515 - version: stable-diffusion.cpp version unknown, commit 5792c66

Could you provide a log with the latest stable-diffusion.cpp release?

[Gzpch97]

does not match the one on the log
Could you provide a log with the latest stable-diffusion.cpp release?

sorry i tried a bunch and i got the log mixed up. here's 5265a5e since the latest has no linux build up

[DEBUG] main.cpp:515 - version: stable-diffusion.cpp version unknown, commit 5265a5e [DEBUG] main.cpp:516 - System Info: SSE3 = 1 | AVX = 1 | AVX2 = 1 | AVX512 = 0 | AVX512_VBMI = 0 | AVX512_VNNI = 0 | FMA = 1 | NEON = 0 | ARM_FMA = 0 | F16C = 1 | FP16_VA = 0 | WASM_SIMD = 0 | VSX = 0 | [DEBUG] main.cpp:517 - SDCliParams { mode: img_gen, output_path: "output.png", verbose: true, color: false, canny_preprocess: false, convert_name: false, preview_method: none, preview_interval: 1, preview_path: "preview.png", preview_fps: 16, taesd_preview: false, preview_noisy: false } [DEBUG] main.cpp:518 - SDContextParams { n_threads: 4, model_path: "./v1-5-pruned-emaonly.safetensors", clip_l_path: "", clip_g_path: "", clip_vision_path: "", t5xxl_path: "", llm_path: "", llm_vision_path: "", diffusion_model_path: "", high_noise_diffusion_model_path: "", vae_path: "", taesd_path: "", esrgan_path: "", control_net_path: "", embedding_dir: "", embeddings: { } wtype: NONE, tensor_type_rules: "", lora_model_dir: ".", photo_maker_path: "", rng_type: cuda, sampler_rng_type: NONE, offload_params_to_cpu: false, enable_mmap: false, control_net_cpu: false, clip_on_cpu: false, vae_on_cpu: false, flash_attn: false, diffusion_flash_attn: false, diffusion_conv_direct: false, vae_conv_direct: false, circular: false, circular_x: false, circular_y: false, chroma_use_dit_mask: true, qwen_image_zero_cond_t: false, chroma_use_t5_mask: false, chroma_t5_mask_pad: 1, prediction: NONE, lora_apply_mode: auto, vae_tiling_params: { 0, 0, 0, 0.5, 0, 0 }, force_sdxl_vae_conv_scale: false } [DEBUG] main.cpp:519 - SDGenerationParams { loras: "{ }", high_noise_loras: "{ }", prompt: "a lovely cat", negative_prompt: "", clip_skip: -1, width: -1, height: -1, batch_count: 1, init_image_path: "", end_image_path: "", mask_image_path: "", control_image_path: "", ref_image_paths: [], control_video_path: "", auto_resize_ref_image: true, increase_ref_index: false, pm_id_images_dir: "", pm_id_embed_path: "", pm_style_strength: 20, skip_layers: [7, 8, 9], sample_params: (txt_cfg: 7.00, img_cfg: 7.00, distilled_guidance: 3.50, slg.layer_count: 3, slg.layer_start: 0.01, slg.layer_end: 0.20, slg.scale: 0.00, scheduler: NONE, sample_method: NONE, sample_steps: 20, eta: 0.00, shifted_timestep: 0, flow_shift: inf), high_noise_skip_layers: [7, 8, 9], high_noise_sample_params: (txt_cfg: 7.00, img_cfg: 7.00, distilled_guidance: 3.50, slg.layer_count: 3, slg.layer_start: 0.01, slg.layer_end: 0.20, slg.scale: 0.00, scheduler: NONE, sample_method: NONE, sample_steps: 20, eta: 0.00, shifted_timestep: 0, flow_shift: inf), custom_sigmas: [], cache_mode: "", cache_option: "", cache: disabled (threshold=inf, start=0.15, end=0.95), moe_boundary: 0.875, video_frames: 1, fps: 16, vace_strength: 1, strength: 0.75, control_strength: 0.9, seed: 42, upscale_repeats: 1, upscale_tile_size: 128, } [DEBUG] stable-diffusion.cpp:183 - Using Vulkan backend [DEBUG] ggml_extend.hpp:75 - ggml_vulkan: Found 1 Vulkan devices: [DEBUG] ggml_extend.hpp:75 - ggml_vulkan: 0 = AMD Radeon RX 580 Series (RADV POLARIS10) (radv) | uma: 0 | fp16: 0 | bf16: 0 | warp size: 64 | shared memory: 65536 | int dot: 0 | matrix cores: none [INFO ] stable-diffusion.cpp:204 - Vulkan: Using device 0 [INFO ] stable-diffusion.cpp:262 - loading model from './v1-5-pruned-emaonly.safetensors' [INFO ] model.cpp:369 - load ./v1-5-pruned-emaonly.safetensors using safetensors format [DEBUG] model.cpp:503 - init from './v1-5-pruned-emaonly.safetensors', prefix = '' [INFO ] stable-diffusion.cpp:355 - Version: SD 1.x [INFO ] stable-diffusion.cpp:383 - Weight type stat: f32: 1131 [INFO ] stable-diffusion.cpp:384 - Conditioner weight type stat: f32: 196 [INFO ] stable-diffusion.cpp:385 - Diffusion model weight type stat: f32: 686 [INFO ] stable-diffusion.cpp:386 - VAE weight type stat: f32: 248 [DEBUG] stable-diffusion.cpp:388 - ggml tensor size = 400 bytes [DEBUG] clip.hpp:161 - vocab size: 49408 [DEBUG] clip.hpp:172 - trigger word img already in vocab [DEBUG] ggml_extend.hpp:2005 - clip params backend buffer size = 469.44 MB(VRAM) (196 tensors) [DEBUG] ggml_extend.hpp:2005 - unet params backend buffer size = 2155.33 MB(VRAM) (686 tensors) [INFO ] stable-diffusion.cpp:672 - using VAE for encoding / decoding [DEBUG] ggml_extend.hpp:2005 - vae params backend buffer size = 94.47 MB(VRAM) (140 tensors) [DEBUG] stable-diffusion.cpp:796 - loading weights [DEBUG] model.cpp:1352 - using 4 threads for model loading [DEBUG] model.cpp:1374 - loading tensors from ./v1-5-pruned-emaonly.safetensors |==================================================| 1131/1131 - 99.92it/s [INFO ] model.cpp:1594 - loading tensors completed, taking 11.32s (process: 0.00s, read: 7.90s, memcpy: 0.00s, convert: 1.00s, copy_to_backend: 1.39s) [DEBUG] stable-diffusion.cpp:832 - finished loaded file [INFO ] stable-diffusion.cpp:884 - total params memory size = 2719.24MB (VRAM 2719.24MB, RAM 0.00MB): text_encoders 469.44MB(VRAM), diffusion_model 2155.33MB(VRAM), vae 94.47MB(VRAM), controlnet 0.00MB(VRAM), pmid 0.00MB(VRAM) [INFO ] stable-diffusion.cpp:954 - running in eps-prediction mode [DEBUG] stable-diffusion.cpp:3252 - generate_image 512x512 [INFO ] stable-diffusion.cpp:3288 - sampling using Euler A method [INFO ] denoiser.hpp:494 - get_sigmas with discrete scheduler [INFO ] stable-diffusion.cpp:3415 - TXT2IMG Illegal instruction (core dumped) ./sd-cli -v -m ./v1-5-pruned-emaonly.safetensors -p "a lovely cat"

[wbruna]

I couldn't reproduce it on a forky chroot on my Debian 13 (kernel 6.18.9+deb13-amd64, mesa-vulkan-drivers 26.0.0-1); working both on a 3400G iGPU and on a RX 7600 XT.

But Koboldcpp ships with a much more recent ggml version, so it could be an already-fixed issue specific to your hardware. I made a test build with a more up-to-date version; could you check if it works for you?

https://github.com/wbruna/stable-diffusion.cpp/actions/runs/23177286855/job/67342222153 (link under "Upload artifacts"); it should also eventually show up under https://github.com/wbruna/stable-diffusion.cpp/releases .

[Gzpch97]

Doesn't work, i wish the error was more descriptive. In my very limited experience with C, i picture it's hell to pinpoint the problem in something this large

[DEBUG] main.cpp:515  - version: stable-diffusion.cpp version unknown, commit f69a1e1
[DEBUG] main.cpp:516  - System Info: 
    SSE3 = 1 |     AVX = 1 |     AVX2 = 1 |     AVX512 = 0 |     AVX512_VBMI = 0 |     AVX512_VNNI = 0 |     FMA = 1 |     NEON = 0 |     ARM_FMA = 0 |     F16C = 1 |     FP16_VA = 0 |     WASM_SIMD = 0 |     VSX = 0 | 
[DEBUG] main.cpp:517  - SDCliParams {
  mode: img_gen,
  output_path: "output.png",
  verbose: true,
  color: false,
  canny_preprocess: false,
  convert_name: false,
  preview_method: none,
  preview_interval: 1,
  preview_path: "preview.png",
  preview_fps: 16,
  taesd_preview: false,
  preview_noisy: false
}
[DEBUG] main.cpp:518  - SDContextParams {
  n_threads: 4,
  model_path: "./v1-5-pruned-emaonly.safetensors",
  clip_l_path: "",
  clip_g_path: "",
  clip_vision_path: "",
  t5xxl_path: "",
  llm_path: "",
  llm_vision_path: "",
  diffusion_model_path: "",
  high_noise_diffusion_model_path: "",
  vae_path: "",
  taesd_path: "",
  esrgan_path: "",
  control_net_path: "",
  embedding_dir: "",
  embeddings: {
  }
  wtype: NONE,
  tensor_type_rules: "",
  lora_model_dir: ".",
  photo_maker_path: "",
  rng_type: cuda,
  sampler_rng_type: NONE,
  offload_params_to_cpu: false,
  enable_mmap: false,
  control_net_cpu: false,
  clip_on_cpu: false,
  vae_on_cpu: false,
  flash_attn: false,
  diffusion_flash_attn: false,
  diffusion_conv_direct: false,
  vae_conv_direct: false,
  circular: false,
  circular_x: false,
  circular_y: false,
  chroma_use_dit_mask: true,
  qwen_image_zero_cond_t: false,
  chroma_use_t5_mask: false,
  chroma_t5_mask_pad: 1,
  prediction: NONE,
  lora_apply_mode: auto,
  vae_tiling_params: { 0, 0, 0, 0.5, 0, 0 },
  force_sdxl_vae_conv_scale: false
}
[DEBUG] main.cpp:519  - SDGenerationParams {
  loras: "{
  }",
  high_noise_loras: "{
  }",
  prompt: "a lovely cat",
  negative_prompt: "",
  clip_skip: -1,
  width: -1,
  height: -1,
  batch_count: 1,
  init_image_path: "",
  end_image_path: "",
  mask_image_path: "",
  control_image_path: "",
  ref_image_paths: [],
  control_video_path: "",
  auto_resize_ref_image: true,
  increase_ref_index: false,
  pm_id_images_dir: "",
  pm_id_embed_path: "",
  pm_style_strength: 20,
  skip_layers: [7, 8, 9],
  sample_params: (txt_cfg: 7.00, img_cfg: 7.00, distilled_guidance: 3.50, slg.layer_count: 3, slg.layer_start: 0.01, slg.layer_end: 0.20, slg.scale: 0.00, scheduler: NONE, sample_method: NONE, sample_steps: 20, eta: 0.00, shifted_timestep: 0, flow_shift: inf),
  high_noise_skip_layers: [7, 8, 9],
  high_noise_sample_params: (txt_cfg: 7.00, img_cfg: 7.00, distilled_guidance: 3.50, slg.layer_count: 3, slg.layer_start: 0.01, slg.layer_end: 0.20, slg.scale: 0.00, scheduler: NONE, sample_method: NONE, sample_steps: 20, eta: 0.00, shifted_timestep: 0, flow_shift: inf),
  custom_sigmas: [],
  cache_mode: "",
  cache_option: "",
  cache: disabled (threshold=inf, start=0.15, end=0.95),
  moe_boundary: 0.875,
  video_frames: 1,
  fps: 16,
  vace_strength: 1,
  strength: 0.75,
  control_strength: 0.9,
  seed: 42,
  upscale_repeats: 1,
  upscale_tile_size: 128,
}
[DEBUG] stable-diffusion.cpp:556  - Using Vulkan backend
[DEBUG] ggml_extend.hpp:75   - ggml_vulkan: Found 1 Vulkan devices:
[DEBUG] ggml_extend.hpp:75   - ggml_vulkan: 0 = AMD Radeon RX 580 Series (RADV POLARIS10) (radv) | uma: 0 | fp16: 0 | bf16: 0 | warp size: 64 | shared memory: 65536 | int dot: 0 | matrix cores: none
[INFO ] stable-diffusion.cpp:577  - Vulkan: Using device 0
[INFO ] stable-diffusion.cpp:635  - loading model from './v1-5-pruned-emaonly.safetensors'
[INFO ] model.cpp:369  - load ./v1-5-pruned-emaonly.safetensors using safetensors format
[DEBUG] model.cpp:503  - init from './v1-5-pruned-emaonly.safetensors', prefix = ''
[INFO ] stable-diffusion.cpp:728  - Version: SD 1.x 
[INFO ] stable-diffusion.cpp:756  - Weight type stat:                      f32: 1131 
[INFO ] stable-diffusion.cpp:757  - Conditioner weight type stat:          f32: 196  
[INFO ] stable-diffusion.cpp:758  - Diffusion model weight type stat:      f32: 686  
[INFO ] stable-diffusion.cpp:759  - VAE weight type stat:                  f32: 248  
[DEBUG] stable-diffusion.cpp:761  - ggml tensor size = 400 bytes
[DEBUG] clip.hpp:161  - vocab size: 49408
[DEBUG] clip.hpp:172  - trigger word img already in vocab
[DEBUG] ggml_extend.hpp:2005 - clip params backend buffer size =  469.44 MB(VRAM) (196 tensors)
[DEBUG] ggml_extend.hpp:2005 - unet params backend buffer size =  2155.33 MB(VRAM) (686 tensors)
[INFO ] stable-diffusion.cpp:1045 - using VAE for encoding / decoding
[DEBUG] ggml_extend.hpp:2005 - vae params backend buffer size =  94.47 MB(VRAM) (140 tensors)
[DEBUG] stable-diffusion.cpp:1169 - loading weights
[DEBUG] model.cpp:1352 - using 4 threads for model loading
[DEBUG] model.cpp:1374 - loading tensors from ./v1-5-pruned-emaonly.safetensors
  |==================================================| 1131/1131 - 103.70it/s
[INFO ] model.cpp:1594 - loading tensors completed, taking 10.91s (process: 0.00s, read: 7.84s, memcpy: 0.00s, convert: 0.96s, copy_to_backend: 1.09s)
[DEBUG] stable-diffusion.cpp:1205 - finished loaded file
[INFO ] stable-diffusion.cpp:1257 - total params memory size = 2719.24MB (VRAM 2719.24MB, RAM 0.00MB): text_encoders 469.44MB(VRAM), diffusion_model 2155.33MB(VRAM), vae 94.47MB(VRAM), controlnet 0.00MB(VRAM), pmid 0.00MB(VRAM)
[INFO ] stable-diffusion.cpp:1327 - running in eps-prediction mode
[DEBUG] stable-diffusion.cpp:3298 - generate_image 512x512
[INFO ] stable-diffusion.cpp:3334 - sampling using Euler A method
[INFO ] denoiser.hpp:494  - get_sigmas with discrete scheduler
[INFO ] stable-diffusion.cpp:3461 - TXT2IMG
Illegal instruction        (core dumped) ./sd-cli -v -m ./v1-5-pruned-emaonly.safetensors -p "a lovely cat"

[Gzpch97]

As a hail mary i compiled 537-545fac4 myself and it appears to work fine. i used the vulkan sdk tarball 1.4.341.1
i will test the official release once it publishes, tell me if there's anything i can do to help narrow whatever is going on.
here's the executable, github didn't let me upload so i hope this site i grabbed of google is acceptable

working log:
/sd-cli -v -m ./v1-5-pruned-emaonly.safetensors -p "a lovely cat"
[DEBUG] main.cpp:515 - version: stable-diffusion.cpp version master-537-545fac4, commit 545fac4
[DEBUG] main.cpp:516 - System Info:
SSE3 = 1 | AVX = 1 | AVX2 = 0 | AVX512 = 0 | AVX512_VBMI = 0 | AVX512_VNNI = 0 | FMA = 1 | NEON = 0 | ARM_FMA = 0 | F16C = 1 | FP16_VA = 0 | WASM_SIMD = 0 | VSX = 0 |
[DEBUG] main.cpp:517 - SDCliParams {
mode: img_gen,
output_path: "output.png",
verbose: true,
color: false,
canny_preprocess: false,
convert_name: false,
preview_method: none,
preview_interval: 1,
preview_path: "preview.png",
preview_fps: 16,
taesd_preview: false,
preview_noisy: false
}
[DEBUG] main.cpp:518 - SDContextParams {
n_threads: 4,
model_path: "./v1-5-pruned-emaonly.safetensors",
clip_l_path: "",
clip_g_path: "",
clip_vision_path: "",
t5xxl_path: "",
llm_path: "",
llm_vision_path: "",
diffusion_model_path: "",
high_noise_diffusion_model_path: "",
vae_path: "",
taesd_path: "",
esrgan_path: "",
control_net_path: "",
embedding_dir: "",
embeddings: {
}
wtype: NONE,
tensor_type_rules: "",
lora_model_dir: ".",
photo_maker_path: "",
rng_type: cuda,
sampler_rng_type: NONE,
offload_params_to_cpu: false,
enable_mmap: false,
control_net_cpu: false,
clip_on_cpu: false,
vae_on_cpu: false,
flash_attn: false,
diffusion_flash_attn: false,
diffusion_conv_direct: false,
vae_conv_direct: false,
circular: false,
circular_x: false,
circular_y: false,
chroma_use_dit_mask: true,
qwen_image_zero_cond_t: false,
chroma_use_t5_mask: false,
chroma_t5_mask_pad: 1,
prediction: NONE,
lora_apply_mode: auto,
vae_tiling_params: { 0, 0, 0, 0.5, 0, 0 },
force_sdxl_vae_conv_scale: false
}
[DEBUG] main.cpp:519 - SDGenerationParams {
loras: "{
}",
high_noise_loras: "{
}",
prompt: "a lovely cat",
negative_prompt: "",
clip_skip: -1,
width: -1,
height: -1,
batch_count: 1,
init_image_path: "",
end_image_path: "",
mask_image_path: "",
control_image_path: "",
ref_image_paths: [],
control_video_path: "",
auto_resize_ref_image: true,
increase_ref_index: false,
pm_id_images_dir: "",
pm_id_embed_path: "",
pm_style_strength: 20,
skip_layers: [7, 8, 9],
sample_params: (txt_cfg: 7.00, img_cfg: 7.00, distilled_guidance: 3.50, slg.layer_count: 3, slg.layer_start: 0.01, slg.layer_end: 0.20, slg.scale: 0.00, scheduler: NONE, sample_method: NONE, sample_steps: 20, eta: 0.00, shifted_timestep: 0, flow_shift: inf),
high_noise_skip_layers: [7, 8, 9],
high_noise_sample_params: (txt_cfg: 7.00, img_cfg: 7.00, distilled_guidance: 3.50, slg.layer_count: 3, slg.layer_start: 0.01, slg.layer_end: 0.20, slg.scale: 0.00, scheduler: NONE, sample_method: NONE, sample_steps: 20, eta: 0.00, shifted_timestep: 0, flow_shift: inf),
custom_sigmas: [],
cache_mode: "",
cache_option: "",
cache: disabled (threshold=inf, start=0.15, end=0.95),
moe_boundary: 0.875,
video_frames: 1,
fps: 16,
vace_strength: 1,
strength: 0.75,
control_strength: 0.9,
seed: 42,
upscale_repeats: 1,
upscale_tile_size: 128,
}
[DEBUG] stable-diffusion.cpp:556 - Using Vulkan backend
[DEBUG] ggml_extend.hpp:75 - ggml_vulkan: Found 1 Vulkan devices:
[DEBUG] ggml_extend.hpp:75 - ggml_vulkan: 0 = AMD Radeon RX 580 Series (RADV POLARIS10) (radv) | uma: 0 | fp16: 0 | bf16: 0 | warp size: 64 | shared memory: 65536 | int dot: 0 | matrix cores: none
[INFO ] stable-diffusion.cpp:577 - Vulkan: Using device 0
[INFO ] stable-diffusion.cpp:635 - loading model from './v1-5-pruned-emaonly.safetensors'
[INFO ] model.cpp:369 - load ./v1-5-pruned-emaonly.safetensors using safetensors format
[DEBUG] model.cpp:503 - init from './v1-5-pruned-emaonly.safetensors', prefix = ''
[INFO ] stable-diffusion.cpp:728 - Version: SD 1.x
[INFO ] stable-diffusion.cpp:756 - Weight type stat: f32: 1131
[INFO ] stable-diffusion.cpp:757 - Conditioner weight type stat: f32: 196
[INFO ] stable-diffusion.cpp:758 - Diffusion model weight type stat: f32: 686
[INFO ] stable-diffusion.cpp:759 - VAE weight type stat: f32: 248
[DEBUG] stable-diffusion.cpp:761 - ggml tensor size = 400 bytes
[DEBUG] clip.hpp:161 - vocab size: 49408
[DEBUG] clip.hpp:172 - trigger word img already in vocab
[DEBUG] ggml_extend.hpp:2005 - clip params backend buffer size = 469.44 MB(VRAM) (196 tensors)
[DEBUG] ggml_extend.hpp:2005 - unet params backend buffer size = 2155.33 MB(VRAM) (686 tensors)
[INFO ] stable-diffusion.cpp:1045 - using VAE for encoding / decoding
[DEBUG] ggml_extend.hpp:2005 - vae params backend buffer size = 94.47 MB(VRAM) (140 tensors)
[DEBUG] stable-diffusion.cpp:1169 - loading weights
[DEBUG] model.cpp:1352 - using 4 threads for model loading
[DEBUG] model.cpp:1374 - loading tensors from ./v1-5-pruned-emaonly.safetensors
|==================================================| 1131/1131 - 104.26it/s
[INFO ] model.cpp:1594 - loading tensors completed, taking 10.85s (process: 0.00s, read: 8.27s, memcpy: 0.00s, convert: 0.96s, copy_to_backend: 0.91s)
[DEBUG] stable-diffusion.cpp:1205 - finished loaded file
[INFO ] stable-diffusion.cpp:1257 - total params memory size = 2719.24MB (VRAM 2719.24MB, RAM 0.00MB): text_encoders 469.44MB(VRAM), diffusion_model 2155.33MB(VRAM), vae 94.47MB(VRAM), controlnet 0.00MB(VRAM), pmid 0.00MB(VRAM)
[INFO ] stable-diffusion.cpp:1327 - running in eps-prediction mode
[DEBUG] stable-diffusion.cpp:3298 - generate_image 512x512
[INFO ] stable-diffusion.cpp:3334 - sampling using Euler A method
[INFO ] denoiser.hpp:494 - get_sigmas with discrete scheduler
[INFO ] stable-diffusion.cpp:3461 - TXT2IMG
[DEBUG] conditioner.hpp:378 - parse 'a lovely cat' to [['a lovely cat', 1], ]
[DEBUG] clip.hpp:402 - token a lovely cat
[DEBUG] clip.hpp:305 - token length: 77
[DEBUG] ggml_extend.hpp:1818 - clip compute buffer size: 1.42 MB(VRAM)
[DEBUG] conditioner.hpp:516 - computing condition graph completed, taking 224 ms
[DEBUG] conditioner.hpp:378 - parse '' to [['', 1], ]
[DEBUG] clip.hpp:305 - token length: 77
[DEBUG] ggml_extend.hpp:1818 - clip compute buffer size: 1.42 MB(VRAM)
[DEBUG] conditioner.hpp:516 - computing condition graph completed, taking 26 ms
[INFO ] stable-diffusion.cpp:3040 - get_learned_condition completed, taking 254 ms
[INFO ] stable-diffusion.cpp:3151 - generating image: 1/1 - seed 42
[DEBUG] ggml_extend.hpp:1818 - unet compute buffer size: 559.90 MB(VRAM)
|==================================================| 20/20 - 1.21s/it
[INFO ] stable-diffusion.cpp:3193 - sampling completed, taking 24.14s
[INFO ] stable-diffusion.cpp:3204 - generating 1 latent images completed, taking 24.14s
[INFO ] stable-diffusion.cpp:3207 - decoding 1 latents
[DEBUG] ggml_extend.hpp:1818 - vae compute buffer size: 1984.06 MB(VRAM)
[DEBUG] vae.hpp:181 - computing vae decode graph completed, taking 8.56s
[INFO ] stable-diffusion.cpp:3217 - latent 1 decoded, taking 8.56s
[INFO ] stable-diffusion.cpp:3221 - decode_first_stage completed, taking 8.56s
[INFO ] stable-diffusion.cpp:3573 - generate_image completed in 32.96s
[INFO ] main.cpp:424 - save result image 0 to 'output.png' (success)
[INFO ] main.cpp:473 - 1/1 images saved

[wbruna]

Unfortunately, I don't know much details about the Vulkan build process. I'd suspect it could be some incompatibility between your hardware and the toolset used by our Github builds (SDK version, compiler version, etc). But it works for you with 1.4.341.1; stable-diffusion.cpp builds have the older 1.4.328.1 (https://github.com/leejet/stable-diffusion.cpp/blob/master-537-545fac4/.github/workflows/build.yml#L315); and Koboldcpp uses an even older embedded SDK, so that doesn't make much sense, either.

[wbruna]

Wait: I've noticed a difference between the non-working log:

SSE3 = 1 | AVX = 1 | AVX2 = 1 | AVX512 = 0 | AVX512_VBMI = 0 | AVX512_VNNI = 0 | FMA = 1 | NEON = 0 | ARM_FMA = 0 | F16C = 1 | FP16_VA = 0 | WASM_SIMD = 0 | VSX = 0 |

and the working one:

SSE3 = 1 | AVX = 1 | AVX2 = 0 | AVX512 = 0 | AVX512_VBMI = 0 | AVX512_VNNI = 0 | FMA = 1 | NEON = 0 | ARM_FMA = 0 | F16C = 1 | FP16_VA = 0 | WASM_SIMD = 0 | VSX = 0 |

I believe your CPU does not support AVX2?

[Gzpch97]

I've checked with this, and it seems it doesn't, only avx, could that be it?

[wbruna]

Yes, that's the issue.

ggml, our backend library, is usually supposed to be built natively, not distributed as a binary. So the default build, if you don't specify any flags, targets the "native" hardware - which happens to enable avx2 in our binary release. Koboldcpp has binary releases targeting broader hardware support, so the one you've downloaded likely had avx2 disabled.

IMHO this is still a bug; or maybe several:

• on stable-diffusion.cpp side, the "System Info:" message is a bit misleading: it's showing the features of the "system" where the binary was built, instead of where it is running (I believed otherwise myself, until I looked at the code). And maybe the project could consider targeting non-avx2 hardware for Vulkan binaries, since it tends to be the backend of choice for broader hardware support (but that's not without issues: just changing the flag could lower performance for avx2 hardware when part of the inference needs to run on the CPU; adding a separate binary target would work, but increase CI resources...);
• ggml should at least give a clear warning about having being built with features not supported by the current hardware, instead of just crashing. Ideally, it should support a way to select CPU support at runtime, but that's a whole other can of worms:
  • Bug: Processor features are determined at compile time ggml-org/llama.cpp#9147
  • Tried to run AVX2 instruction on AVX only CPU ggml-org/llama.cpp#1380
  • Compilation / distribution of a whisper.dll that 'just works' and is performant across all x86 processors, whether they support AVX2 extensions or not ggml-org/whisper.cpp#1939 (comment)

And the main workaround is the one you've already tested: just build stable-diffusion.cpp locally.