llama.cpp
Add MiniMax-M3 (MSA: MiniMax Sparse Attention) support
#24908
Merged
Overview
Commits
64
Changes
View On GitHub

Add MiniMax-M3 (MSA: MiniMax Sparse Attention) support #24908

ngxson merged 64 commits into ggml-org:master from timkhronos:MSA
timkhronos
danielhanchen Add preliminary MiniMax-M3 support
87cc29bc
timkhronos MiniMax-M3 vision tower (mmproj + clip graph)
53c81ddb
timkhronos Delete m3_vision_ref.py
f07f1d43
timkhronos Update clip.cpp
4a3206fd
timkhronos MSA
bbf1a808
timkhronos Update constants.py
09657dd7
timkhronos Update minimax.py
8fe2e015
timkhronos Cache creation. Working withotu flash attention
8c953a9c
timkhronos Added flash attention for sparse layers
ea6fbd6f
timkhronos Decomposed slow cpu OP into GPU + CPU ops. Massive speedup over long ctx
2e827590
timkhronos Rewrote indexer op to be cuda native. Modified flash attention to mat…
0152226b
timkhronos Implement sparse attention calc out of stock ops.
d1a04f7c
timkhronos Fix a cache allocation and cont issue
b1b174ee
timkhronos Fixed -fa auto crash, flagged debug spots
cea714a8
timkhronos Delete vocab.json
69c958d9
timkhronos Delete model.safetensors.index.json
afc09f1f
timkhronos Delete generation_config.json
f40d0f58
timkhronos Delete Minimax directory
714bbe93
timkhronos timkhronos requested a review 69 days ago
timkhronos timkhronos requested a review from JohannesGaessler JohannesGaessler 69 days ago
timkhronos timkhronos requested a review from CISC CISC 69 days ago
timkhronos timkhronos requested a review from ggerganov ggerganov 69 days ago
github-actions github-actions added model
github-actions github-actions added testing
github-actions github-actions added examples
github-actions github-actions added python
ggml-gh-bot
timkhronos
timkhronos Handled multi stream case to fall back on Dense Attention
8136a9c6
timkhronos Development scaffolding cleanup. No functional change to the decode or
3ed9b18b
timkhronos
timkhronos Remove redundant comment from minimax-m3.cpp
79a6eecf
USBhost
juj
timkhronos
timkhronos
CISC
juj
timkhronos
CISC
timkhronos Changed 3 Gelu Ops for vision into Gelu_erf ops
35990be6
timkhronos Assert that n_kv is multiple of 128
fa158507
github-actions github-actions added mtmd
github-actions github-actions added conversion
timkhronos
am17an
CISC
timkhronos
timkhronos Rename MSA index tensors to indexer convention
eae55e2c
timkhronos
detain
detain
timkhronos Fix incorrect Assert
2bb7eeb3
timkhronos
CISC
ngxson
Ganju0
CISC
timkhronos Review driven changes (#3)
d6f9426d
CISC
CISC commented on 2026-06-28
timkhronos Remove comment from conversion minimax.py
0a7b2dc5
timkhronos Remove whitespaces from constants.py
636143a0
timkhronos Tighten comment in minimax.py
7b7ff65e
timkhronos inherit MiniMax-M3 from MiniMax-M2
1cd03ef8
timkhronos drop dead text_config fallbacks
c70c8a93
timkhronos Add indexer writer methods
25199fa2
timkhronos Reuse LLM_FFN_SWIGLU_OAI_MOE
d54b4ebd
CISC
CISC commented on 2026-06-28
timkhronos Remove duplicate indexer setters, add only block_size/local_blocks, …
f57efcee
CISC
CISC commented on 2026-06-28
timkhronos Fix conversion error /gguf_writer.py
618e1454
timkhronos Update gguf-py/gguf/gguf_writer.py
5a24782d
timkhronos Update gguf-py/gguf/tensor_mapping.py
5dfe8387
timkhronos Update conversion/minimax.py
b99a8d5e
timkhronos Update conversion/minimax.py
56ba5416
CISC
CISC commented on 2026-06-28
timkhronos Remove whitespace in src/llama-kv-cache.cpp
044391ac
timkhronos Remove Whitespace in Update src/llama-model.h
c314fafe
timkhronos Remove whitespace in src/llama-hparams.h
5015a6bb
usrlocalben
Ganju0
timkhronos
ngxson
timkhronos remove multimodal code upon maintainer request. Will be made as a sep…
c97a2ac8
timkhronos
timkhronos Whitespace clean in tensor_mapping.py
fcb9ed2d
fairydreaming
CISC CISC removed review request 62 days ago
CISC CISC removed mtmd
ngxson
smalinin
CISC
timkhronos Merge branch 'master' into MSA
51e7bf80
timkhronos Merge branch 'ggml-org:master' into MSA
6b14feaa
timkhronos
smalinin
timkhronos
timkhronos Log cache size on launch, block ctx shift, support prompt caching
d2fe995d
neuromaniacMD
timkhronos Merge branch 'ggml-org:master' into MSA
9ea2aa12
timkhronos Update minimax-m3.cpp
aa73ec39
Sciguy429
timkhronos Optimize implementation, add multi stream support.
cacc42f7
timkhronos
Sciguy429
timkhronos
Sciguy429
Sciguy429
konradzamojski
timkhronos set default cache type to F32
6b61dd48
timkhronos Fix potential DSA double indexer cache allocation bug, only allocate…
5cd1c193
timkhronos remove F16 downcasts in MSA attention, force F32 indexer score accum
e99545c1
drrros
timkhronos
drrros
erm14254
timkhronos
timkhronos Add Minimax eos to llama vocab
f0201211
erm14254
timkhronos Guard edge case where idx cache can become stale after a tail trim
da2c1685
timkhronos
timkhronos Merge branch 'ggml-org:master' into MSA
297a179c
ngxson
ngxson commented on 2026-07-24
timkhronos Update llama-kv-cache.cpp
e53544b9
timkhronos Merge branch 'master' into MSA
954598b9
timkhronos
timkhronos Review driven changes
d6cc3375
timkhronos
timkhronos Merge branch 'ggml-org:master' into MSA
8bfbead0
ngxson style fix
285df2dd
ngxson indexer hparams are required
a56fd294
ngxson
ngxson approved these changes on 2026-07-26
ngxson
ngxson fix tests
296e8e40
ngxson fix lint
5d5ed01c
timkhronos
ngxson
timkhronos
bartowski1182
bartowski1182
bartowski1182
bartowski1182 approved these changes on 2026-07-26
ngxson ngxson merged b1d4c655 into master 35 days ago
gschukat
timkhronos
heapsoftware
heapsoftware
AesSedai
heapsoftware
AesSedai
aldehir
heapsoftware
aldehir
heapsoftware
bartowski1182
ggerganov
ggerganov
timkhronos
timkhronos timkhronos deleted the MSA branch 30 days ago
EugeoSynthesisThirtyTwo
timkhronos

Login to write a write a comment.

Login via GitHub

Reviewers
Assignees
No one assigned
Labels
model testing examples python conversion
Milestone
No milestone
Terms Of Use Privacy Policy Refund Policy Impressum