Skip to content

fix(cuda): preserve first-token handoff through native host #1031

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

Closed
jrb-ai wants to merge 1 commit into from
Closed

fix(cuda): preserve first-token handoff through native host #1031

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
7 changes: 5 additions & 2 deletions zig/src/native/cuda.zig
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -176,7 +176,7 @@ pub const Host = struct {
/// The family's backend, each call made with the context current on the calling thread.
fn backend(h: *Host) lanes.backend.Backend {
h.vtable = lane_backend.vtable(Host, h.inner.vtable.*);
return .{ .ptr = h, .vtable = &h.vtable };
return .{ .ptr = h, .vtable = &h.vtable, .first_before_draft = h.inner.first_before_draft };
}
};

Expand Down Expand Up @@ -487,12 +487,15 @@ test {
_ = windows;
}

test "the lane core sees a family's prefill_step through the lane backend, so a served prompt can step" {
test "CUDA host preserves prefill_step and first-before-draft capability" {
var target: lanes.fake.Fake = .{ .gpa = std.testing.allocator, .prefill_chunks = 3 };
defer target.deinit();
var h: Host = undefined;
h.inner = target.stepped();
try std.testing.expect(h.backend().vtable.prefill_step != null);
try std.testing.expect(!h.backend().first_before_draft);
h.inner.first_before_draft = true;
try std.testing.expect(h.backend().first_before_draft);
}

test "every registered family and d1's decision engine are listed for capabilities" {
Expand Down