Cannot run FLUX.2-klein under AMD Radeon 890M with ROCm build #1235
Description
Hi all,
I've tried the latest build sd-master-43e829f-bin-win-rocm-x64.zip under AMD Radeon 890M (Ryzen AI 9 HX370)
However, it failed.
Command-line arguments used
sd-cli.exe --diffusion-model ../models/Flux/flux-2-klein-4b.safetensors --vae ../models/Flux/flux2-vae.safetensors --llm ../models/text_encoders/qwen_3_4b.safetensors -p "a lovely cat" --cfg-scale 1.0 --steps 4 -v --offload-to-cpu --diffusion-fa --vae-conv-direct --clip-on-cpu
Logs / error messages / stack trace
[DEBUG] main.cpp:500 - version: stable-diffusion.cpp version unknown, commit 43e829f
[DEBUG] main.cpp:501 - 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:502 - 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:503 - SDContextParams {
n_threads: 12,
model_path: "",
clip_l_path: "",
clip_g_path: "",
clip_vision_path: "",
t5xxl_path: "",
llm_path: "../models/text_encoders/qwen_3_4b.safetensors",
llm_vision_path: "",
diffusion_model_path: "../models/Flux/flux-2-klein-4b.safetensors",
high_noise_diffusion_model_path: "",
vae_path: "../models/Flux/flux2-vae.safetensors",
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,
flow_shift: INF
offload_params_to_cpu: true,
enable_mmap: false,
control_net_cpu: false,
clip_on_cpu: true,
vae_on_cpu: false,
diffusion_flash_attn: true,
diffusion_conv_direct: false,
vae_conv_direct: true,
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:504 - 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: 1.00, img_cfg: 1.00, 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: 4, eta: 0.00, shifted_timestep: 0),
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),
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:164 - Using CUDA backend
[INFO ] stable-diffusion.cpp\ggml_extend.hpp:78 - ggml_cuda_init: found 1 ROCm devices:
[INFO ] stable-diffusion.cpp\ggml_extend.hpp:78 - Device 0: AMD Radeon(TM) 890M Graphics, gfx1150 (0x1150), VMM: no, Wave Size: 32
[INFO ] stable-diffusion.cpp:258 - loading diffusion model from '../models/Flux/flux-2-klein-4b.safetensors'
[INFO ] model.cpp:373 - load ../models/Flux/flux-2-klein-4b.safetensors using safetensors format
[DEBUG] model.cpp:507 - init from '../models/Flux/flux-2-klein-4b.safetensors', prefix = 'model.diffusion_model.'
[INFO ] stable-diffusion.cpp:305 - loading llm from '../models/text_encoders/qwen_3_4b.safetensors'
[INFO ] model.cpp:373 - load ../models/text_encoders/qwen_3_4b.safetensors using safetensors format
[DEBUG] model.cpp:507 - init from '../models/text_encoders/qwen_3_4b.safetensors', prefix = 'text_encoders.llm.'
[INFO ] stable-diffusion.cpp:319 - loading vae from '../models/Flux/flux2-vae.safetensors'
[INFO ] model.cpp:373 - load ../models/Flux/flux2-vae.safetensors using safetensors format
[DEBUG] model.cpp:507 - init from '../models/Flux/flux2-vae.safetensors', prefix = 'vae.'
[INFO ] stable-diffusion.cpp:335 - Version: Flux.2 klein
[INFO ] stable-diffusion.cpp:363 - Weight type stat: f32: 248 | bf16: 547
[INFO ] stable-diffusion.cpp:364 - Conditioner weight type stat: bf16: 398
[INFO ] stable-diffusion.cpp:365 - Diffusion model weight type stat: bf16: 149
[INFO ] stable-diffusion.cpp:366 - VAE weight type stat: f32: 248
[DEBUG] stable-diffusion.cpp:368 - ggml tensor size = 400 bytes
[INFO ] stable-diffusion.cpp:427 - CLIP: Using CPU backend
[DEBUG] stable-diffusion.cpp\llm.hpp:285 - merges size 151387
[DEBUG] stable-diffusion.cpp\llm.hpp:317 - vocab size: 151669
[DEBUG] stable-diffusion.cpp\llm.hpp:1139 - llm: num_layers = 36, vocab_size = 151936, hidden_size = 2560, intermediate_size = 9728
[INFO ] stable-diffusion.cpp\flux.hpp:1353 - flux: depth = 5, depth_single_blocks = 20, guidance_embed = false, context_in_dim = 7680, hidden_size = 3072, num_heads = 24
[INFO ] stable-diffusion.cpp:573 - Using flash attention in the diffusion model
[DEBUG] stable-diffusion.cpp\ggml_extend.hpp:1950 - qwen3 params backend buffer size = 8414.50 MB(RAM) (398 tensors)
[DEBUG] stable-diffusion.cpp\ggml_extend.hpp:1950 - flux params backend buffer size = 7392.03 MB(RAM) (149 tensors)
[INFO ] stable-diffusion.cpp:624 - Using Conv2d direct in the vae model
[DEBUG] stable-diffusion.cpp\ggml_extend.hpp:1950 - vae params backend buffer size = 96.72 MB(RAM) (140 tensors)
[DEBUG] stable-diffusion.cpp:752 - loading weights
[DEBUG] model.cpp:1381 - using 12 threads for model loading
[DEBUG] model.cpp:1403 - loading tensors from ../models/Flux/flux-2-klein-4b.safetensors
|=========> | 149/795 - 10.44it/s←[K
[DEBUG] model.cpp:1403 - loading tensors from ../models/text_encoders/qwen_3_4b.safetensors
|==================================> | 547/795 - 18.93it/s←[K
[DEBUG] model.cpp:1403 - loading tensors from ../models/Flux/flux2-vae.safetensors
|==================================================| 795/795 - 27.12it/s←[K
[INFO ] model.cpp:1629 - loading tensors completed, taking 29.32s (process: 0.00s, read: 27.90s, memcpy: 0.00s, convert: 0.04s, copy_to_backend: 0.00s)
[DEBUG] stable-diffusion.cpp:787 - finished loaded file
[INFO ] stable-diffusion.cpp:860 - total params memory size = 15903.24MB (VRAM 7488.74MB, RAM 8414.50MB): text_encoders 8414.50MB(RAM), diffusion_model 7392.03MB(VRAM), vae 96.72MB(VRAM), controlnet 0.00MB(VRAM), pmid 0.00MB(RAM)
[INFO ] stable-diffusion.cpp:939 - running in Flux2 FLOW mode
[DEBUG] stable-diffusion.cpp:3472 - generate_image 512x512
[INFO ] stable-diffusion.cpp:3506 - sampling using Euler method
[DEBUG] stable-diffusion.cpp\denoiser.hpp:703 - Flux2FlowDenoiser: set shift to 2.031
[INFO ] stable-diffusion.cpp\denoiser.hpp:403 - get_sigmas with discrete scheduler
[INFO ] stable-diffusion.cpp:3633 - TXT2IMG
[DEBUG] stable-diffusion.cpp\conditioner.hpp:1679 - parse '<|im_start|>user
a lovely cat<|im_end|>
<|im_start|>assistant
' to [['<|im_start|>user
', 1], ['a lovely cat', 1], ['<|im_end|>
<|im_start|>assistant
', 1], ]
[DEBUG] stable-diffusion.cpp\llm.hpp:259 - split prompt "<|im_start|>user
" to tokens ["<|im_start|>", "user", "Ċ", ]
[DEBUG] stable-diffusion.cpp\llm.hpp:259 - split prompt "a lovely cat" to tokens ["a", "Ġlovely", "Ġcat", ]
[DEBUG] stable-diffusion.cpp\llm.hpp:259 - split prompt "<|im_end|>
<|im_start|>assistant
" to tokens ["<|im_end|>", "Ċ", "<|im_start|>", "assistant", "Ċ", "", "ĊĊ", "", "ĊĊ", ]
[DEBUG] stable-diffusion.cpp\llm.hpp:203 - token length: 512
[DEBUG] stable-diffusion.cpp\ggml_extend.hpp:1762 - qwen3 compute buffer size: 74.00 MB(RAM)
[DEBUG] stable-diffusion.cpp\conditioner.hpp:1923 - computing condition graph completed, taking 11443 ms
[INFO ] stable-diffusion.cpp:3250 - get_learned_condition completed, taking 11446 ms
[INFO ] stable-diffusion.cpp:3361 - generating image: 1/1 - seed 42
[INFO ] stable-diffusion.cpp\ggml_extend.hpp:1862 - flux offload params (7392.03 MB, 149 tensors) to runtime backend (ROCm0), taking 3.23s
[DEBUG] stable-diffusion.cpp\ggml_extend.hpp:1762 - flux compute buffer size: 356.00 MB(VRAM)
[ERROR] stable-diffusion.cpp\ggml_extend.hpp:84 - ROCm error: invalid device function
[ERROR] stable-diffusion.cpp\ggml_extend.hpp:84 - current device: 0, in function ggml_cuda_op_mul_mat at D:/a/stable-diffusion.cpp/stable-diffusion.cpp/ggml/src/ggml-cuda/ggml-cuda.cu:1722
[ERROR] stable-diffusion.cpp\ggml_extend.hpp:84 - hipGetLastError()
D:/a/stable-diffusion.cpp/stable-diffusion.cpp/ggml/src/ggml-cuda/ggml-cuda.cu:96: ROCm error
workflows
/build.yml It does not show "gfx1150", is this issue related to it?
Do I need to install ROCm or HIP related packages?
Or Radeon 890M cannot work on ROCm ? (It seems gfx1150, which is over ROCm 7.0 ?)
Thanks