Skip to content

gemma4_31b (MLX): don't allocate a second KV-cache copy for single-session runs #20560

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Open
metascroy wants to merge 1 commit into
base: main
Choose a base branch
from
+9 −1
Open

gemma4_31b (MLX): don't allocate a second KV-cache copy for single-session runs #20560

Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
10 changes: 9 additions & 1 deletion examples/models/gemma4_31b/gemma4_31b_engine.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -684,7 +684,15 @@ Result<std::unique_ptr<Gemma4_31BEngine>> Gemma4_31BEngine::create(
}
}
#elif defined(EXECUTORCH_BUILD_MLX)
mutable_state = std::make_unique<GemmaMutableStateContextOwner>();
// Only enable the per-session mutable-buffer path when actually serving more
// than one session. For a single session (the CLI runner) the rebind would
// allocate a second copy of the KV-cache buffers on top of the program's
// default buffers — doubling KV-cache memory and adding a one-time
// session-buffer allocation during the first prefill — for no isolation
// benefit. Leaving mutable_state null keeps the program's default buffers.
if (config.max_sessions > 1) {
mutable_state = std::make_unique<GemmaMutableStateContextOwner>();
}
#endif

auto module_res = mutable_state != nullptr
Expand Down
Loading