model : fix k-pool indexer softmax gridDim.y overflow - #28730

Closed

AIalliAI wants to merge 48 commits into

ggml-org:masterfrom

AIalliAI:fix/glm5-kpool-softmax-grid-y

Commits on Aug 27, 2026

Rebase GLM-Next support onto master, and migrate to llama-memory-hybrid-idx
timkhronos
committed
9370c82
Add initial MTP support
timkhronos
committed
d4ce825

Commits on Aug 28, 2026

Merge branch optimizations. Reduce allocated compute buffer size, speed up long context decode, fla, and slight MTP improvements.
timkhronos
committed
d2fc716
Review driven changes, remove env vars, protect tensors
timkhronos
committed
9855711
Strip MTP for initial PR
timkhronos
committed
b93ed51
Clean up after mtp strip
timkhronos
authored
ba436fd
Clean up after mtp strip
timkhronos
authored
3ff8d92
Update speculative.cpp
timkhronos
authored
f176c8a
Update llama-context.h
timkhronos
authored
c34db58
Clean up after mtp strip
timkhronos
authored
0aa327b

Commits on Aug 29, 2026

Fix tokenizer ignore merges
timkhronos
committed
ec1cdbf
Improve quantization protection selection
timkhronos
committed
7152e9b

Commits on Aug 31, 2026

Refactor mhc helpers, graph base
timkhronos
committed
fdc54be
Merge branch 'master' into GLM5.3-Flash
timkhronos
authored
8543941
Lint Fixes
timkhronos
committed
74bb0e3
Apply suggestions from code review

timkhronos
and
CISC
authored
a771613
Skip glm5-next in model saver, fix CRLF
timkhronos
committed
5728a4b
Skip glm5-next in sweep
timkhronos
committed
c35bddd
Remove T4 fallback
timkhronos
committed
3bdb2d8

Commits on Sep 1, 2026

Review cleanup
timkhronos
committed
81f95af
Merge branch 'master' into GLM5.3-Flash
timkhronos
authored
3498cc7
Merge branch 'ggml-org:master' into GLM5.3-Flash
timkhronos
authored
611e407
Review suggestions
timkhronos
committed
1eca274
Defer separate MTP gguf handling to MTP PR, drop filter
timkhronos
committed
7de5a8e
Repad n_head_kv
timkhronos
committed
9fe9fd7
kpool init apply
timkhronos
committed
a386cd7
Merge branch 'ggml-org:master' into GLM5.3-Flash
timkhronos
authored
5b05fcc
Order by descending score
timkhronos
committed
120eb9e

Commits on Sep 2, 2026

Commits on Sep 3, 2026

Add multi stream support
timkhronos
committed
ff6be95
Merge branch 'ggml-org:master' into GLM5.3-Flash
timkhronos
authored
ad74ec3
Finish Rebase
timkhronos
committed
1b564d2
Sparse FA fir DSA prefill
timkhronos
committed
99fdaab
Merge branch 'ggml-org:master' into GLM5.3-Flash
timkhronos
authored
db54719
Const
timkhronos
committed
5c4bd50

Commits on Sep 4, 2026

Merge branch 'master' into GLM5.3-Flash
timkhronos
authored
cc05bfe
Update llama-model.cpp to fix rebase error
timkhronos
authored
8134115

Commits on Sep 10, 2026

Commits on Sep 11, 2026

model : fix k-pool indexer softmax gridDim.y overflow
AIalliAI
committed
79de1f5
model : comment k-pool softmax gridDim.y reshape
AIalliAI
committed
798d6ce