Skip to content

Fix metal kernel_norm broken by ca82cf7 #3060

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

Merged
merged 2 commits into from
Sep 7, 2023
Merged

Fix metal kernel_norm broken by ca82cf7 #3060

Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
f9c8ccc
Fix kernel_norm broken by ca82cf7
Kawrakow Sep 7, 2023
7d6fac3
Merge branch 'master' into ik/fix_kernel_norm
ggerganov Sep 7, 2023
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
30 changes: 5 additions & 25 deletions ggml-metal.metal
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -220,29 +220,14 @@ kernel void kernel_norm(
}
threadgroup_barrier(mem_flags::mem_threadgroup);
}
// broadcast
if (tpitg == 0) {
sum[0] /= ne00;
}
threadgroup_barrier(mem_flags::mem_threadgroup);
const float mean = sum[0];
const float mean = sum[0] / ne00;

// recenter
// recenter and VARIANCE
threadgroup_barrier(mem_flags::mem_threadgroup);
device float * y = dst + tgpig*ne00;
for (int i00 = tpitg; i00 < ne00; i00 += ntg) {
y[i00] = x[i00] - mean;
}

// VARIANCE
// parallel sum
//
// WARNING: combining this loop with the one above will give you wrong results for nth == 256
// I have no idea why, so for now I am keeping them separate. But this behavior is very concerning.
// Tested with:
// ./perplexity -m ./falcon-7b/ggml-model-q4_0.gguf -f wiki.test.raw -ngl 1 -t 4
//
sum[tpitg] = 0.0f;
for (int i00 = tpitg; i00 < ne00; i00 += ntg) {
y[i00] = x[i00] - mean;
sum[tpitg] += y[i00] * y[i00];
}

Expand All @@ -254,12 +239,7 @@ kernel void kernel_norm(
}
threadgroup_barrier(mem_flags::mem_threadgroup);
}
// broadcast
if (tpitg == 0) {
sum[0] /= ne00;
}
threadgroup_barrier(mem_flags::mem_threadgroup);
const float variance = sum[0];
const float variance = sum[0] / ne00;

const float scale = 1.0f/sqrt(variance + eps);
for (int i00 = tpitg; i00 < ne00; i00 += ntg) {
Expand Down