]> git.djapps.eu Git - pkg/ggml/sources/llama.cpp/commit
context : reserve new scheduler when graph topology changes (#18547)
authorGeorgi Gerganov <redacted>
Thu, 15 Jan 2026 14:39:17 +0000 (16:39 +0200)
committerGitHub <redacted>
Thu, 15 Jan 2026 14:39:17 +0000 (16:39 +0200)
commit39173bcacb67329850b9ff3108dd036eafb680f0
treea1294726ade6e9183f92ca4bbf6f280852a92003
parent5c662d21a3a1c6a41d8abe401f5791712a5c02ee
context : reserve new scheduler when graph topology changes (#18547)

* context : reserve new scheduler when graph topology changes

* cont : fix

* cont : fix reserve

* cont : reserve only when changes occur + timing

* context : add comments

* llama : reserve on sampler changes

* common : allow null common_sampler

* server : task declares needs (embd, logits, sampling)

* server : do not init sampler if not needed

* llama : fix need_reserve when unsetting a sampler

* server : consolidate slot reset/clear logic
common/common.cpp
common/sampling.cpp
examples/batched/batched.cpp
include/llama.h
src/llama-context.cpp
src/llama-context.h
src/llama-cparams.h
tools/server/server-context.cpp
tools/server/server-task.h