llama.cpp
model: GraniteSWAForCausalLM / GraniteMoeSWAForCausalLM
#25505
Merged

model: GraniteSWAForCausalLM / GraniteMoeSWAForCausalLM #25505

gabe-l-hart
github-actions github-actions added model
github-actions github-actions added conversion
pwilkin
gabe-l-hart
gabe-l-hart
gabe-l-hart feat(convert): Add conversion for GraniteSWAForCausalLM
4bbf9aba
gabe-l-hart feat(llama): Add granite_swa support
c74049a8
gabe-l-hart feat(conversion): Add conversion infra for rope_pattern array
1f3741a5
gabe-l-hart fix(conversion): Fix SWA pattern logic and support for non-rope layers
f7d057b7
gabe-l-hart feat(conversion): Add support for GraniteMoeSWA
44215bb1
gabe-l-hart feat: Add llama_hparams::has_rope and arch constants
cf315175
gabe-l-hart feat: Add support for per-layer rope determination
f5d4e9f4
gabe-l-hart style: Fix failing flake8 for extra newlines
9d75ea8a
gabe-l-hart test: Write out SLIDING_WINDOW_PATTERN in llama-model-saver
ef14e1c3
gabe-l-hart fix(convert): Fix missing registration for GraniteMoeSWAForCausalLM
c32fe0bf
gabe-l-hart fix: Load MoE params as optional
36aa2e02
gabe-l-hart feat: Handle MoE params in conversion
94d2643c
gabe-l-hart gabe-l-hart force pushed to 94d2643c 112 days ago
gabe-l-hart
gabe-l-hart Merge remote-tracking branch 'origin/master' into GraniteSWAForCausalLM
dc7f155f
gabe-l-hart style: Remove unnecessary newline
90c3d646
gabe-l-hart fix: Remove unnecessary tensor additions to GRANITE architecture
0b1df4c5
gabe-l-hart fix: Correctly handle naming for ffn gate inp
8865f1bd
gabe-l-hart
gabe-l-hart commented on 2026-08-12
gabe-l-hart
gabe-l-hart commented on 2026-08-12
gabe-l-hart fix: Always default hparams.rope_pattern to 1s
8e5bcadb
gabe-l-hart feat: Move to has_rope for all granite model architectures
977437b0
gabe-l-hart feat: No hacky rope_finetuned fallback in has_rope
e400df6f
gabe-l-hart
gabe-l-hart commented on 2026-08-12
gabe-l-hart fix: Fully remove rope hparam filling in granitemoe
e7695cf3
gabe-l-hart
gabe-l-hart commented on 2026-08-12
gabe-l-hart
gabe-l-hart commented on 2026-08-12
gabe-l-hart gabe-l-hart marked this pull request as ready for review 84 days ago
gabe-l-hart gabe-l-hart requested a review from CISC CISC 84 days ago
gabe-l-hart gabe-l-hart requested a review from ggerganov ggerganov 84 days ago
gabe-l-hart
gabe-l-hart fix: Save out rope_pattern in model-saver
03067047
gabe-l-hart
gabe-l-hart fix: Set hparams.rope_finetuned for round trip
93f6cb8c
gabe-l-hart
pwilkin
pwilkin approved these changes on 2026-08-12
CISC
CISC commented on 2026-08-13
gabe-l-hart fix: Code review cleanup
600615dc
gabe-l-hart refactor: Keep gate/up fused for MoE path
94fc69fd
gabe-l-hart Merge remote-tracking branch 'origin/master' into GraniteSWAForCausalLM
283c4ec4
gabe-l-hart fix: Skip GRANITE_SWA in model saver
c94d5ac9
gabe-l-hart
gabe-l-hart commented on 2026-08-17
CISC add sliding window pattern for model in test
b5b7d1d6
CISC CISC requested a review from JohannesGaessler JohannesGaessler 78 days ago
CISC CISC removed review request from JohannesGaessler JohannesGaessler 78 days ago
github-actions github-actions added testing
CISC
CISC approved these changes on 2026-08-18
gabe-l-hart style: Fix indentation
f7b13112
gabe-l-hart fix: Fix \r\n
eeb83f3e
gabe-l-hart feat: Keep shared expert fused
d7a74f1e
CISC
CISC approved these changes on 2026-08-19
CISC
CISC commented on 2026-08-19
CISC
CISC approved these changes on 2026-08-19
CISC
gabe-l-hart
CISC
gabe-l-hart style: More indentation fixes
de5f6fd5
gabe-l-hart gabe-l-hart force pushed to de5f6fd5 77 days ago
CISC
CISC approved these changes on 2026-08-19
gabe-l-hart
CISC
CISC CISC merged 7221e24f into master 77 days ago

Login to write a write a comment.

Login via GitHub

Reviewers
Assignees
No one assigned
Labels
Milestone