llama : reuse compute graphs #14482
ggerganov
merged
15 commits into
master
from
gg/llama-reuse-graphs
Base automatically changed from
gg/kv-cache-use-set-rows
to
master
1 year ago
esrakorkmz
approved these changes
on 2025-07-03
compilade
commented
on 2025-07-05
ggerganov
commented
on 2025-07-08
slaren
commented
on 2025-07-09
llama : reuse compute graphs
594b184e
llama-bench : add graph reuse parameter
ffc7634c
cont : remove the parameter and the sched resets
02cd9574
graph : rename update() to can_reuse()
3b8cef9f
params : remove is_same()
c82a5336
graph : set res->params in llm_graph_context constructor
3d28b3b1
slaren
commented
on 2025-07-16
graph : avoid set_max_nodes in llm_graph_result
b7b6caf3
kv-cache : reuse llama_context's graph result instance
84e0c1f3
Merge branch 'master' into gg/llama-reuse-graphs
0ef1f6bb
context : reset the previous graph result upon memory updates
3d7ec2b9
batch : llama_ubatch now carries its data instead of pointing to balloc
acaf4b77
Merge branch 'master' into gg/llama-reuse-graphs
04155f0b
merge : fix build
a872790b
ggerganov
commented
on 2025-07-17
graph : fix can_reuse() checks when flash-attention is disabled
41366a4d
graph : move llm_graph_result impl in source file + debug env
c7ccf383
slaren
commented
on 2025-07-17
Reviewers
Assignees
No one assigned
Labels
examples
hot
Milestone
No milestone
Login to write a write a comment.
Login via GitHub