mtmd: Add Vision Support for Minimax-M3 (#25113)

• Add preliminary MiniMax-M3 support

Text-only port that re-uses existing components: MiniMax-M2 style GQA with per-head QK-norm and partial rotary, DeepSeek-V3 style leading-dense and routed/shared experts, and swigluoai activation. Sparse attention is not yet supported (dense fallback); vision tower and MTP heads are dropped.

• MiniMax-M3 vision tower (mmproj + clip graph)

• Delete m3_vision_ref.py

• Update clip.cpp

• MSA

• Update constants.py

• Update minimax.py

• Cache creation. Working withotu flash attention

• Added flash attention for sparse layers

• Decomposed slow cpu OP into GPU + CPU ops. Massive speedup over long ctx

• Rewrote indexer op to be cuda native. Modified flash attention to match per group block picking

• Implement sparse attention calc out of stock ops.

• Fix a cache allocation and cont issue

• Fixed -fa auto crash, flagged debug spots

• Delete vocab.json

• Delete model.safetensors.index.json

• Delete generation_config.json

• Delete Minimax directory

• Handled multi stream case to fall back on Dense Attention

• Development scaffolding cleanup. No functional change to the decode or 4-way paths. Full debug harness remains at <8136a9c68ed7a5eb009aa67bba3fda8062f4648f> for reproducing the selection-parity validation.

• Remove redundant comment from minimax-m3.cpp

• Changed 3 Gelu Ops for vision into Gelu_erf ops

• Assert that n_kv is multiple of 128

• Rename MSA index tensors to indexer convention

Note: All GGUFs generated before this change will need to be regenerated.

• Fix incorrect Assert

• Review driven changes (#3)

• Remove comment from conversion minimax.py

Co-authored-by: Sigbjørn Skjæret [email protected]

• Remove whitespaces from constants.py

Co-authored-by: Sigbjørn Skjæret [email protected]

• Tighten comment in minimax.py

Co-authored-by: Sigbjørn Skjæret [email protected]

• inherit MiniMax-M3 from MiniMax-M2

• drop dead text_config fallbacks

• Add indexer writer methods

• Reuse LLM_FFN_SWIGLU_OAI_MOE

• Remove duplicate indexer setters, add only block_size/local_blocks, follow value naming convention

• Fix conversion error /gguf_writer.py

Co-authored-by: Sigbjørn Skjæret [email protected]

• Update gguf-py/gguf/gguf_writer.py

Co-authored-by: Sigbjørn Skjæret [email protected]

• Update gguf-py/gguf/tensor_mapping.py

Co-authored-by: Sigbjørn Skjæret [email protected]

• Update conversion/minimax.py

Co-authored-by: Sigbjørn Skjæret [email protected]

• Update conversion/minimax.py

Co-authored-by: Sigbjørn Skjæret [email protected]

• Remove whitespace in src/llama-kv-cache.cpp

Co-authored-by: Sigbjørn Skjæret [email protected]

• Remove Whitespace in Update src/llama-model.h

Co-authored-by: Sigbjørn Skjæret [email protected]

• Remove whitespace in src/llama-hparams.h

Co-authored-by: Sigbjørn Skjæret [email protected]

• Update minimax_m3.cpp

Rewrite code comment based on feedback and to better reflect the actual architecture, and reuse existing build_vit

• Rename minimax_m3.cpp to minimax-m3.cpp

• Update CMakeLists.txt

• Remove debug code from clip.cpp

• Update clip.cpp

• Update comments in tools/mtmd/models/minimax-m3.cpp

• Permute Q/K at conversion, drop precomputed sin/cos

• Log cache size on launch, block ctx shift, support prompt caching

Log indexer cache size on launch

Disallow ctx shift

Support prompt caching

• Update minimax-m3.cpp

• Optimize implementation, add multi stream support.

Fully rewrote minimax-m3.cpp for speed and buffer size gains:

Unified the 4-way + decode, 1 FA call per layer instead of 4, with the groups mapped onto ne[3]

Custom CPU op now emits block-level mask, expanded on GPU, which causes CPU to GPU transfer to shrinks at prefill

Decode: ~25 nodes/layer vs ~50, no per-group concats/conts

Unified selection semantics, so both regimes rank bs + local bias (position-anchored local force), which means prefill/decode can no longer disagree on selection

can_reuse on the MSA bias input. Graph reuse at decode restored (was rebuilding the full graph every token)

In-place mask adds, shrinking compute buffer ~6.8 to ~4.2 GiB at ub2048/62k

Multi-stream: MSA now runs with -np N when kv_unified=false. Decode stays batched across streams (still 1 FA call), prefill loops per stream. dense fallback only for --kv-unified + multi-seq

Measured effect on expert offload bound setup: decode 6.2(4WAY)–7.15(MSA_decode) -> 7.7~7.8 t/s, flat from 5k to 60k+. prefill around 10% faster. buffer about 20% smaller, multi-user support.

• set default cache type to F32

• Fix potential DSA double indexer cache allocation bug, only allocate in-cache k_idx for archs that opt in

• remove F16 downcasts in MSA attention, force F32 indexer score accum

• Add Minimax eos to llama vocab

• Guard edge case where idx cache can become stale after a tail trim

• Update llama-kv-cache.h

• Update llama-kv-cache.cpp

• Update llama-kv-cache.cpp

• Update llama-kv-cache.h

• Change resize Pad to none, resize alg to Bicubic Pillow

• Review driven changes

• Update llama-kv-cache.cpp

• rm unrotated pos_t

• fused rope w + pad

• rename merge --> merger for consistency

• add review skill for mtmd

• graph should use hparams n_merge

• fix lint

Co-authored-by: Daniel Han [email protected] Co-authored-by: Sigbjørn Skjæret [email protected] Co-authored-by: Xuan Son Nguyen [email protected]