Skip to content

sched : fix reserve ignoring user tensor assignments#17232

Merged
slaren merged 1 commit intomasterfrom
sl/fix-sched-reserve-reset
Nov 13, 2025
Merged

sched : fix reserve ignoring user tensor assignments#17232
slaren merged 1 commit intomasterfrom
sl/fix-sched-reserve-reset

Conversation

@slaren
Copy link
Member

@slaren slaren commented Nov 13, 2025

Calling ggml_backend_sched_reset before splitting the graph was causing the user tensor backend assignments to be reset, which in turn causes future allocations to require a full re-allocation of the graph, which disables pipeline parallelism. I don't remember why I added this, but it is definitely wrong.

Fixes #17163

@slaren slaren merged commit dd091e5 into master Nov 13, 2025
66 of 67 checks passed
@slaren slaren deleted the sl/fix-sched-reserve-reset branch November 13, 2025 12:14
@github-actions github-actions bot added the ggml changes relating to the ggml tensor library for machine learning label Nov 13, 2025
Anico2 added a commit to Anico2/llama.cpp that referenced this pull request Jan 15, 2026
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

ggml changes relating to the ggml tensor library for machine learning

Projects

None yet

Development

Successfully merging this pull request may close these issues.

Eval bug: Much slower prompt processing (only using 1 GPU) on odd prompt sizes

2 participants