PR #20340 llama : enable chunked fused GDN path

llama : enable chunked fused GDN path #20340

ggerganov merged 9 commits into master from gg/llama-allow-gdn-ch

llama : enable chunked fused GDN path

ec2443a9

ggerganov requested a review from

CISC 8 days ago

ggerganov requested a review from

am17an 8 days ago

github-actions added model

github-actions added Nvidia GPU

github-actions added ggml

am17an approved these changes on 2026-03-10

models : avoid Q and K repeats when using fused GDA

39b6f5a7

ggerganov force pushed from 444eeed6 to 39b6f5a7 8 days ago

am17an commented on 2026-03-10

cont : fix comment

79541c0a

am17an commented on 2026-03-10

cont : fix the fix

46c693dc

cont : fix

c6b76caf

CISC commented on 2026-03-10

metal : add GDN kernel (#20361)

f6a0c16e

github-actions added Apple Metal

CUDA: AR gated delta net improvements (#20391)

d1b2301f

github-actions added testing

Merge branch 'master' into gg/llama-allow-gdn-ch

1baffbcb

llama : refactor llm_build_delta_net_base API

bac4f0f8

ggerganov merged d28961d8 into master 7 days ago

ggerganov deleted the gg/llama-allow-gdn-ch branch 7 days ago

ggml-org locked as too heated and limited conversation to collaborators 6 days ago

ggml-org unlocked this conversation 6 days ago

Reviewers

am17an

CISC

Assignees

No one assigned

Labels

model testing Nvidia GPU ggml Apple Metal

Milestone

No milestone

llama.cpp llama : enable chunked fused GDN path #20340 Merged

llama : enable chunked fused GDN path #20340

llama.cpp
llama : enable chunked fused GDN path
#20340

Merged