Add MiniMax H3 architecture detection for metadata-less GGUF files - #480
Open
uppet wants to merge 1 commit into
Open
Add MiniMax H3 architecture detection for metadata-less GGUF files#480uppet wants to merge 1 commit into
uppet wants to merge 1 commit into
Conversation
Community GGUF conversions of the MiniMax H3 video DiT (e.g. unsloth/MiniMax-H3-GGUF) ship without metadata fields, so loading falls back to tensor-name detection, which fails with 'This model is not currently supported - (Unknown model architecture!)'. Add a ModelMiniMaxH3 template detected by adaln_t_table, blocks.0.adaln_proj.linear.weight and blocks.0.attn.qkv_proj.weight, and register minimax_h3 in IMG_ARCH_LIST. Tensor names verified to match the official Comfy-Org safetensors state dict exactly.
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
PR/Issue draft — city96/ComfyUI-GGUF
Title: Add MiniMax H3 architecture detection for metadata-less GGUF files
Problem
Community GGUF conversions of the MiniMax H3 video DiT (e.g.
unsloth/MiniMax-H3-GGUF) ship without any metadata fields (kv_count=0, nogeneral.architecture). Loading them falls back to tensor-name detection, which fails with:Change
Adds a
ModelMiniMaxH3template detected by three keys unique to the H3 state dict layout —adaln_t_table,blocks.0.adaln_proj.linear.weight,blocks.0.attn.qkv_proj.weight— and registersminimax_h3inIMG_ARCH_LIST.The detected names were verified to match the official Comfy-Org safetensors state dict exactly, so once detection passes, core ComfyUI model construction (
model_type FLOW_AV) works unchanged. No collisions with existing templates (closest arecosmos'sblocks.0.adaln_modulation_cross_attn.1.weightandwan'sblocks.0.self_attn.*).Tested
UnetLoaderGGUFloadingminimax_h3_ref2va_pruned-Q4_K.gguf(11.4 GB, 532 tensors): detection succeeds (arch:minimax_h3, loaded in sd.cpp compat mode), state dict keys match the native safetensors one-to-one.arch_list/IMG_ARCH_LIST.