Best speedup
3.06x
9 accepted edits
Run detail
Live multi-trial optimizer demo
modal-b200-live on NVIDIA B200 with Fireworks + 1M controller. Git c8e96db.
runningOpen HUD
Best latency
0.0630 ms
Lower is better
Trials
12
3 rejected
Artifacts
1
Blob-backed source and logs
Improvement curve
Preparing chart...
GPU samples
Preparing chart...
Reward
raw reward preservedPreparing chart...
Trial audit log
Best source| Trial | Op | Candidate | Verdict | Reward | Speedup | Latency | Cost |
|---|---|---|---|---|---|---|---|
| #12 | fused_mlp | persistent_blocks_v3 | accepted | 3.060 | 3.06x | 0.1340 ms | $0.0024 |
| #11 | fused_mlp | fuse_gelu_matmul | accepted | 2.410 | 2.41x | 0.1700 ms | $0.0022 |
| #10 | fused_mlp | default_seed | accepted | 1.000 | 1.00x | 0.4100 ms | $0.0020 |
| #9 | matmul_tile | overfit_tile_bad_heldout held-out correctness failed | incorrect | 0.000 | n/a | n/a | $0.0018 |
| #8 | matmul_tile | tensor_core_tile_64 | accepted | 2.080 | 2.08x | 0.1180 ms | $0.0016 |
| #7 | matmul_tile | default_seed | accepted | 1.000 | 1.00x | 0.2450 ms | $0.0014 |
| #6 | softmax_rows | warp_reduce_v2 | accepted | 1.720 | 1.72x | 0.0690 ms | $0.0012 |
| #5 | softmax_rows | torch_passthrough anti-hack passthrough blocked | incorrect | 0.000 | n/a | n/a | $0.0010 |
| #4 | softmax_rows | default_seed | accepted | 1.000 | 1.00x | 0.1180 ms | $0.0008 |
| #3 | rmsnorm | block256_vectorized | accepted | 1.340 | 1.34x | 0.0630 ms | $0.0006 |
| #2 | rmsnorm | bad_shape_edit shape mismatch | incorrect | 0.000 | n/a | n/a | $0.0004 |
| #1 | rmsnorm | default_seed | accepted | 1.000 | 1.00x | 0.0840 ms | $0.0002 |