ComfyUI-SolAttn_triton
Opt-in MiniMax H3 long-sequence Sol-Attn for AMD ROCm with conservative exact-row routing.
Nodes (3)
MiniMax H3 Sol-Attn for AMD ROCm
DBJ's conservative, opt-in long-sequence Sol-Attn patch for native ComfyUI MiniMax H3 on AMD ROCm.
comfy node install minimax-h3-sol-attn-rocm
What this release does
MiniMaxH3FastPatch follows ComfyUI's ModelAttentionBackend, leaving Comfy Kitchen attention as the dense backend and fallback. It activates only on eligible MiniMax H3 self-attention calls of at least 12,288 tokens.
Validated policy:
- 90% estimated attention-mass route coverage
- 50% minimum exact-block floor
- exact conditioning, reference, and target-audio rows
- INT8 QK/PV attention arithmetic
- dense fallback on unsupported or malformed calls
The patch is approximate and disabled unless the user explicitly enables it.
Validated environment: Linux, ComfyUI 0.32.0, comfy-kitchen 0.2.31, and RX 7900 XTX/gfx1100. Other GPUs and ComfyUI versions are unvalidated.
Placement and LoRA
MiniMaxH3INT8FastLoader
→ LoraLoaderBypassModelOnly (optional)
→ ModelAttentionBackend (comfy kitchen attention)
→ MiniMaxH3FastPatch
→ MiniMaxH3BlockCacheT8 (optional)
→ MiniMaxH3SigmaShift
→ scheduler / guider / sampler
A fixed-strength bypass LoRA is compatible because Sol-Attn sees the resulting current-call Q/K/V tensors. Match the LoRA variant and its 4-step or 8-step schedule to the FL2VA or Ref2VA base.
Companion packages
Provenance and license boundary
This release is derived from kijai/ComfyUI-SolAttn_triton, pinned at 0e334dc981cfe3b0ed926ee13ad43f64914b7f5b. See H3_COMMUNITY_R01.md for frozen release details.
The upstream repository does not currently declare a software license. This fork does not add or imply one.