Skip to content

Non-record: 10L MLP3x int6 baseline (MLX smoke test) #404

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
yashward001 wants to merge 1 commit into
base: main
Choose a base branch
from
Open

Non-record: 10L MLP3x int6 baseline (MLX smoke test) #404

Show file tree
Hide file tree
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
20 changes: 20 additions & 0 deletions records/track_10min_16mb/2026-03-22_10L_MLP3x_Int6_Baseline/README.md
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -0,0 +1,20 @@
# 10L MLP3x Int6 Baseline (non-record)

Non-record submission. Local MLX smoke test confirming pipeline works end-to-end.

## Config
- 10 layers, 512 dim, 8 heads, 4 KV heads
- MLP 3x expansion (hidden=1536), relu²
- int6 quantization, zlib-9 compression
- Trained on Apple Silicon (MLX), 200 iterations only

## Score
val_bpb: 2.3517 (200 iterations — not a competitive score)

## Planned improvements
- zstd-22 compression
- Sliding window eval (stride=64)
- Muon WD=0.04
- SmearGate + BigramHash
- SWA over last 40% of warmdown
- Full 10-min run on 8xH100
6 changes: 6 additions & 0 deletions records/track_10min_16mb/2026-03-22_10L_MLP3x_Int6_Baseline/submission.json
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -0,0 +1,6 @@
{
"name": "Your Name",
"github_id": "your_github_username",
"val_bpb": 2.3517,
"notes": "Non-record submission. 10-layer, 3x MLP, int6 quant baseline run on Apple Silicon MLX. 200 iterations smoke test only \u2014 full H100 run pending compute grant."
}
Loading