[DEBUG] main.cpp:689  - version: stable-diffusion.cpp version master-813-bfbef5b-u13b9d92, commit 13b9d92
ggml_cuda_init: found 1 CUDA devices (Total VRAM: 97249 MiB):
  Device 0: NVIDIA RTX PRO 6000 Blackwell Server Edition, compute capability 12.0, VMM: yes, VRAM: 97249 MiB
[DEBUG] main.cpp:690  - System Info: 
   SSE3 = 1 |    SSSE3 = 1 |    AVX = 1 |    AVX_VNNI = 1 |    AVX2 = 1 |    F16C = 1 |    FMA = 1 |    BMI2 = 1 |    AVX512 = 1 |    AVX512_VBMI = 1 |    AVX512_VNNI = 1 |    AVX512_BF16 = 1 |    AMX_INT8 = 1 |    OPENMP = 1 |    REPACK = 1 | 
[DEBUG] main.cpp:691  - SDCliParams {
  mode: vid_gen,
  output_path: "/tmp/h3-iq1/outputs/smoke-IQ1_M.webm",
  image_path: "",
  metadata_format: "text",
  verbose: true,
  color: false,
  canny_preprocess: false,
  convert_name: false,
  preview_method: none,
  preview_interval: 1,
  preview_path: "preview.avi",
  preview_fps: 16,
  taesd_preview: false,
  preview_noisy: false,
  imatrix_out: "",
  metadata_raw: false,
  metadata_brief: false,
  metadata_all: false
}
[DEBUG] main.cpp:692  - SDContextParams {
  n_threads: 96,
  model_path: "",
  clip_l_path: "",
  clip_g_path: "",
  clip_vision_path: "",
  t5xxl_path: "",
  llm_path: "/tmp/h3-iq1/models/qwen3vl_32b_minimax_h3-Q2_K_M.gguf",
  llm_vision_path: "",
  diffusion_model_path: "/tmp/h3-iq1/outputs/minimax_h3_fl2va_pruned-IQ1_M.gguf",
  high_noise_diffusion_model_path: "",
  uncond_diffusion_model_path: "",
  embeddings_connectors_path: "",
  vae_path: "/tmp/h3-iq1/models/vae/minimax_h3_video_vae_fp16.safetensors",
  vae_format: "auto",
  audio_vae_path: "",
  taesd_path: "",
  esrgan_path: "",
  control_net_path: "",
  embedding_dir: "",
  embeddings: {
  }
  wtype: NONE,
  tensor_type_rules: "",
  lora_model_dir: ".",
  hires_upscalers_dir: "",
  photo_maker_path: "",
  rng_type: cpu,
  sampler_rng_type: NONE,
  offload_params_to_cpu: false,
  max_vram: "0",
  stream_layers: false,
  eager_load: false,
  backend: "te=cpu",
  params_backend: "",
  split_mode: "",
  model_args: "",
  auto_fit: false,
  enable_mmap: false,
  control_net_cpu: false,
  clip_on_cpu: false,
  vae_on_cpu: false,
  flash_attn: false,
  diffusion_flash_attn: true,
  diffusion_conv_direct: false,
  vae_conv_direct: false,
  prediction: NONE,
  lora_apply_mode: auto,
  force_sdxl_vae_conv_scale: false
}
[DEBUG] main.cpp:693  - SDGenerationParams {
  loras: "{
  }",
  high_noise_loras: "{
  }",
  prompt: "A red fox trots through falling snow, cinematic.",
  negative_prompt: "",
  ad_model_path: "",
  ad_prompt: "",
  ad_negative_prompt: "",
  extra_ad_args: "",
  clip_skip: -1,
  width: 320,
  height: 192,
  batch_count: 1,
  qwen_image_layers: 3,
  init_image_path: "",
  end_image_path: "",
  mask_image_path: "",
  control_image_path: "",
  ref_image_paths: [],
  ref_video_paths: [],
  ref_video_audio_paths: [],
  ref_audio_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: 0, slg.layer_start: 0.01, slg.layer_end: 0.20, slg.scale: 0.00, scheduler: NONE, sample_method: NONE, sample_steps: 1, eta: inf, shifted_timestep: 0, flow_shift: inf, extra_sample_args: ),
  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: 0, slg.layer_start: 0.01, slg.layer_end: 0.20, slg.scale: 0.00, scheduler: NONE, sample_method: NONE, sample_steps: 20, eta: inf, shifted_timestep: 0, flow_shift: inf, extra_sample_args: ),
  custom_sigmas: [],
  cache_mode: "",
  cache_option: "",
  cache: disabled (threshold=inf, start=0.15, end=0.95),
  moe_boundary: 0.875,
  video_frames: 5,
  fps: 16,
  vace_strength: 1,
  strength: 0.75,
  control_strength: 0.9,
  seed: 11,
  upscale_repeats: 1,
  upscale_tile_size: 128,
  hires: { enabled: false, upscaler: "Latent", model_path: "", scale: 2, target_width: 0, target_height: 0, steps: 0, denoising_strength: 0.7, custom_sigmas: [], upscale_tile_size: 128 },
  vae_tiling_params: { 0, 0, 0, 0, 0.5, 0, 0, "" },
}
[DEBUG] model_loader.cpp:228  - using 96 threads for model loading
[INFO ] stable-diffusion.cpp:717  - loading diffusion model from '/tmp/h3-iq1/outputs/minimax_h3_fl2va_pruned-IQ1_M.gguf'
[INFO ] model_loader.cpp:236  - load /tmp/h3-iq1/outputs/minimax_h3_fl2va_pruned-IQ1_M.gguf using gguf format
[DEBUG] model_loader.cpp:288  - init from '/tmp/h3-iq1/outputs/minimax_h3_fl2va_pruned-IQ1_M.gguf'
[INFO ] stable-diffusion.cpp:774  - loading llm from '/tmp/h3-iq1/models/qwen3vl_32b_minimax_h3-Q2_K_M.gguf'
[INFO ] model_loader.cpp:236  - load /tmp/h3-iq1/models/qwen3vl_32b_minimax_h3-Q2_K_M.gguf using gguf format
[DEBUG] model_loader.cpp:288  - init from '/tmp/h3-iq1/models/qwen3vl_32b_minimax_h3-Q2_K_M.gguf'
[INFO ] stable-diffusion.cpp:788  - loading vae from '/tmp/h3-iq1/models/vae/minimax_h3_video_vae_fp16.safetensors'
[INFO ] model_loader.cpp:242  - load /tmp/h3-iq1/models/vae/minimax_h3_video_vae_fp16.safetensors using safetensors format
[DEBUG] model_loader.cpp:316  - init from '/tmp/h3-iq1/models/vae/minimax_h3_video_vae_fp16.safetensors', prefix = 'vae.'
[INFO ] stable-diffusion.cpp:905  - Version: MiniMax-H3 
[DEBUG] ggml_extend_backend.cpp:545  - Found 2 backend devices:
[DEBUG] ggml_extend_backend.cpp:548  - #0: CUDA0
[DEBUG] ggml_extend_backend.cpp:548  - #1: CPU
[DEBUG] ggml_extend_backend.cpp:395  - Initializing backend: CUDA0
[INFO ] stable-diffusion.cpp:956  - Weight type stat:                      f32: 10   |     f16: 664  |    q2_K: 300  |    q5_K: 50   |   iq1_m: 208  |    bf16: 764  
[INFO ] stable-diffusion.cpp:957  - Conditioner weight type stat:         q2_K: 300  |    q5_K: 50   |    bf16: 552  
[INFO ] stable-diffusion.cpp:958  - Diffusion model weight type stat:      f32: 10   |     f16: 102  |   iq1_m: 208  |    bf16: 212  
[INFO ] stable-diffusion.cpp:959  - VAE weight type stat:                  f16: 562  
[DEBUG] stable-diffusion.cpp:961  - ggml tensor size = 432 bytes
[DEBUG] ggml_extend_backend.cpp:395  - Initializing backend: CPU
[DEBUG] qwen2_tokenizer.cpp:14   - merges size 151387
[DEBUG] qwen2_tokenizer.cpp:39   - vocab size: 151674
[DEBUG] llm.hpp:286  - llm: num_layers = 50, vocab_size = 151936, hidden_size = 5120, intermediate_size = 25600
[DEBUG] llm.hpp:1768 - enable llm vision
[DEBUG] minimax_h3.hpp:109  - minimax_h3: layers=50, hidden=5376, heads=56, head_dim=128, ffn=14336, adaln_curve=1025
[INFO ] stable-diffusion.cpp:1494 - using VAE for encoding / decoding
[INFO ] stable-diffusion.cpp:1619 - Using flash attention in the diffusion model
[DEBUG] stable-diffusion.cpp:1627 - validating model metadata
[DEBUG] stable-diffusion.cpp:1686 - model metadata validated; weights will be prepared lazily
[INFO ] stable-diffusion.cpp:1727 - total params memory size = 23373.68MB (VRAM 9387.85MB, RAM 13985.83MB): text_encoders 13985.83MB(RAM), diffusion_model 4406.86MB(VRAM), vae 4980.99MB(VRAM), controlnet 0.00MB(N/A), extensions 0.00MB(N/A)
[INFO ] stable-diffusion.cpp:1846 - running in FLOW mode
[WARN ] stable-diffusion.cpp:4092 - MiniMax-H3 uses 24 fps; overriding requested fps 16
[INFO ] denoiser.hpp:1051 - get_sigmas with discrete scheduler
[INFO ] stable-diffusion.cpp:4365 - sampling using Euler method
[DEBUG] conditioner.hpp:1987 - parse 'A red fox trots through falling snow, cinematic.' to [['A red fox trots through falling snow, cinematic.', 1], ]
[DEBUG] bpe_tokenizer.cpp:208  - split prompt "A red fox trots through falling snow, cinematic." to 11 tokens ["A", "Ġred", "Ġfox", "Ġtro", "ts", "Ġthrough", "Ġfalling", "Ġsnow", ",", "Ġcinematic", ".", ]
[DEBUG] model_loader.cpp:1042 - loading 551/902 tensors from /tmp/h3-iq1/models/qwen3vl_32b_minimax_h3-Q2_K_M.gguf

  |##########                                        | 103/551 - 825.59MB/s
  |#####################################             | 398/551 - 10.09GB/s
  |##################################################| 551/551 - 4.09GB/s
[INFO ] model_loader.cpp:1304 - loading tensors completed, taking 2.71s (read: 0.40s, memcpy: 0.00s, convert: 0.01s, copy_to_backend: 0.00s)
[DEBUG] model_manager.cpp:406  - model manager prepared params backend buffer (12844.50 MB, 551 tensors, RAM)
[DEBUG] ggml_extend.hpp:2193 - qwen3vl compute buffer size: 2.58 MB(RAM)
[DEBUG] conditioner.hpp:2882 - computing condition graph completed, taking 3608 ms
[INFO ] stable-diffusion.cpp:6537 - get_learned_condition completed, taking 3.61s
[INFO ] stable-diffusion.cpp:6899 - generate_video 320x192x5
[DEBUG] stable-diffusion.cpp:6964 - sample 20x12x2
[DEBUG] model_loader.cpp:1042 - loading 532/532 tensors from /tmp/h3-iq1/outputs/minimax_h3_fl2va_pruned-IQ1_M.gguf

  |##################                                | 191/532 - 4.17GB/s
  |################################                  | 335/532 - 8.03GB/s
  |##################################################| 532/532 - 8.79GB/s
[INFO ] model_loader.cpp:1304 - loading tensors completed, taking 0.48s (read: 0.01s, memcpy: 0.00s, convert: 0.00s, copy_to_backend: 0.39s)
[DEBUG] model_manager.cpp:406  - model manager prepared params backend buffer (4406.88 MB, 532 tensors, VRAM)
[DEBUG] ggml_extend.hpp:2193 - minimax_h3 compute buffer size: 40.91 MB(VRAM)

  |==================================================| 1/1 - 1.04it/s
[INFO ] stable-diffusion.cpp:6997 - sampling completed, taking 0.96s
[INFO ] stable-diffusion.cpp:7140 - generating latent video completed, taking 0.96s
[DEBUG] stable-diffusion.cpp:6559 - decode_video_outputs latent 20x12x2x24
[DEBUG] vae.hpp:190  - VAE Tile size: 16x12
[DEBUG] ggml_extend.hpp:884  - num tiles : 2, 1 
[DEBUG] ggml_extend.hpp:885  - optimal overlap : 0.750000, 0.000000 (targeting 0.250000)
[DEBUG] ggml_extend.hpp:886  - processing 2 tiles
[DEBUG] model_loader.cpp:1042 - loading 441/562 tensors from /tmp/h3-iq1/models/vae/minimax_h3_video_vae_fp16.safetensors

  |############################                      | 240/441 - 5.27GB/s
  |##########################################        | 369/441 - 7.57GB/s
  |##################################################| 441/441 - 8.66GB/s
[INFO ] model_loader.cpp:1304 - loading tensors completed, taking 0.52s (read: 0.01s, memcpy: 0.00s, convert: 0.00s, copy_to_backend: 0.42s)
[DEBUG] model_manager.cpp:406  - model manager prepared params backend buffer (4636.94 MB, 441 tensors, VRAM)
[DEBUG] ggml_extend.hpp:2193 - minimax_h3_video_vae compute buffer size: 307.03 MB(VRAM)

  |=========================>                        | 1/2 - 1.39it/s
  |==================================================| 2/2 - 5.71it/s
[DEBUG] vae.hpp:220  - computing vae decode graph completed, taking 0.90s
[INFO ] stable-diffusion.cpp:6568 - decode_first_stage completed, taking 0.90s
[DEBUG] stable-diffusion.cpp:6573 - decode_video_outputs decoded 320x192x5x3
[INFO ] stable-diffusion.cpp:7200 - generate_video completed in 5.47s
[DEBUG] model_manager.cpp:981  - model manager releasing params backend buffer (12844.50 MB, 551 tensors, RAM)
[DEBUG] model_manager.cpp:981  - model manager releasing params backend buffer (4406.88 MB, 532 tensors, VRAM)
[DEBUG] model_manager.cpp:981  - model manager releasing params backend buffer (4636.94 MB, 441 tensors, VRAM)
[INFO ] main.cpp:536  - save result video to '/tmp/h3-iq1/outputs/smoke-IQ1_M.webm'