202611011530_FastH3文生图+Sparse Attention_1mp5s_omnixpu加速_B60_64G_冷启.txt · 2026-11-01 15:30
该工作流仅执行 1 次(冷启),没有热启对比数据。
| 阶段 | 耗时 | 步数 | s/it | staged MB | 其他指标 |
|---|---|---|---|---|---|
| 冷启 | 214.80 s | 4 | 33.50 | 19,995 | CUTE_H3_VAE_D64 ×3CUTE_BHLD_D128 ×2 |
[INFO] got prompt [INFO] Model storage policy: fast_disk=True paths=['<ComfyUI>/models/vae/minimax_h3_audio_vae_fp32.safetensors'] [INFO] VAE load device: xpu:0, offload device: cpu, dtype: torch.float32 [INFO] Model storage policy: fast_disk=True paths=['<ComfyUI>/models/vae/minimax_h3_video_vae_fp16.safetensors'] [INFO] VAE load device: xpu:0, offload device: cpu, dtype: torch.float16 [INFO] Found quantization metadata version 1 [INFO] Using MixedPrecisionOps for text encoder [INFO] Model storage policy: fast_disk=True paths=['<ComfyUI>/models/text_encoders/qwen3vl_32b_minimax_h3_nvfp4_awq.safetensors'] [INFO] CLIP/text encoder model load device: xpu:0, offload device: cpu, current: cpu, dtype: torch.float16 [INFO] Requested to load MiniMaxH3TEModel_ [INFO] Model MiniMaxH3TEModel_ prepared for dynamic VRAM loading. 14956MB Staged. 0 patches attached. Force pre-loaded 410 weights: 4572 KB. [INFO] [OmniXPU] norm first use: rms_norm_fn shape=torch.Size([1, 719, 5120]) [omni_xpu::device] warning: physical Device ID 0xe211 (physical_sku=b60) uses kernel_profile=b60, policy_id=b60-v1, support=experimental; functional support may be accepted independently, performance_claim=false [INFO] Found quantization metadata version 1 [INFO] Detected mixed precision quantization [INFO] Using mixed precision operations [INFO] Native ops: int8_tensorwise , emulated ops: float8_e5m2, w6a8_int8, nvfp4, float8_e4m3fn, mxfp8, convrot_w4a4, asym_w4a8_int8 [INFO] model weight dtype torch.bfloat16, manual cast: torch.bfloat16 [INFO] model_type FLOW_AV [INFO] Model storage policy: fast_disk=True paths=['<ComfyUI>/models/diffusion_models/minimax_h3_fused_refdelta_r1024_turbo8_mystic07_int8_convrot.safetensors'] [WARNING] unet unexpected: ['adaln_basis', 'adaln_mean'] [WARNING] Attention backend 'comfy kitchen attention' is unavailable; using PyTorch attention. [INFO] VSA: extra_tokens ignored (the trained sparse pattern is the target) [WARNING] VSA: the model has no to_gate_compress layers; running the fine stage without the coarse branch [INFO] [OmniXPU] AIMDO inference budget: devices=0 minimum=8323.1MiB [INFO] Requested to load MiniMaxH3 [INFO] Model MiniMaxH3 prepared for dynamic VRAM loading. 19995MB Staged. 0 patches attached. Force pre-loaded 210 weights: 1175 KB. [WARNING] [OmniXPU] experimental capability-driven BMG D128 CUTE route enabled; set OMNI_ATTN_BACKEND=torch to disable it if this workflow shows an issue [INFO] [OmniXPU] attention CUTE_BHLD_D128 #1: heads=56 q=719 kv=719 dtype=torch.bfloat16 [INFO] [OmniXPU] attention CUTE_BHLD_D128 #2: heads=56 q=719 kv=719 dtype=torch.bfloat16 100%|███████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████| 4/4 [02:13<00:00, 33.50s/it] [INFO] [OmniXPU] AIMDO inference budget: devices=0 minimum=221.7MiB [INFO] Requested to load MiniMaxH3AudioVAE [INFO] Model MiniMaxH3AudioVAE prepared for dynamic VRAM loading. 576MB Staged. 0 patches attached. Force pre-loaded 401 weights: 539 KB. [INFO] [OmniXPU] AIMDO inference budget: devices=0 minimum=1117.5MiB [INFO] Requested to load MiniMaxH3VideoVAE [INFO] Model MiniMaxH3VideoVAE prepared for dynamic VRAM loading. 4965MB Staged. 0 patches attached. Force pre-loaded 128 weights: 352 KB. [INFO] [OmniXPU] attention CUTE_H3_VAE_D64 #1: heads=32 seq=1797 dtype=torch.float16 [INFO] [OmniXPU] attention CUTE_H3_VAE_D64 #2: heads=32 seq=1797 dtype=torch.float16 [INFO] [OmniXPU] attention CUTE_H3_VAE_D64 #3: heads=32 seq=1797 dtype=torch.float16 [INFO] Prompt executed in 214.80 seconds