✓ Initialized. View run at https://modal.com/apps/seth-45699/main/ap-RsMkG2HydYTyWwERlflduX ✓ Created objects. ├── 🔨 Created mount │ /private/tmp/claude-501/-Users-sethstafford-dev-research-sutro/aabe6221-ffef │ -4b0f-bafe-aa419c95eb0e/scratchpad/c6/run_modal.py ├── 🔨 Created mount │ /private/tmp/claude-501/-Users-sethstafford-dev-research-sutro/aabe6221-ffef │ -4b0f-bafe-aa419c95eb0e/scratchpad/c6/mnist.py └── 🔨 Created function remote_score. ========== == CUDA == ========== CUDA Version 13.3.0 Container image Copyright (c) 2016-2023, NVIDIA CORPORATION & AFFILIATES. All rights reserved. ========== == CUDA == ========== CUDA Version 13.3.0 Container image Copyright (c) 2016-2023, NVIDIA CORPORATION & AFFILIATES. All rights reserved. ========== == CUDA == ========== CUDA Version 13.3.0 Container image Copyright (c) 2016-2023, NVIDIA CORPORATION & AFFILIATES. All rights reserved. --- run 1: ladder_d3.py:ladder at 2.70% error: 15 timed calls, each in a fresh process NVIDIA A100-SXM4-80GB, torch 2.12.0+cu130, sandbox on call 1/15: 15,811.740 ms, scorer's clock 15,812.277 ms call 2/15: 15,811.103 ms, scorer's clock 15,812.991 ms call 3/15: 15,832.995 ms, scorer's clock 15,860.192 ms call 4/15: 15,838.030 ms, scorer's clock 15,838.448 ms call 5/15: 15,837.302 ms, scorer's clock 15,837.801 ms call 6/15: 15,835.291 ms, scorer's clock 15,837.819 ms call 7/15: 15,812.384 ms, scorer's clock 15,815.099 ms call 8/15: 15,812.129 ms, scorer's clock 15,814.619 ms call 9/15: 15,824.431 ms, scorer's clock 15,827.229 ms call 10/15: 15,826.587 ms, scorer's clock 15,828.971 ms call 11/15: 15,821.231 ms, scorer's clock 15,825.747 ms call 12/15: 15,836.590 ms, scorer's clock 15,841.971 ms call 13/15: 15,821.257 ms, scorer's clock 15,824.169 ms call 14/15: 15,818.741 ms, scorer's clock 15,821.317 ms call 15/15: 15,826.569 ms, scorer's clock 15,829.428 ms MNIST 97.48% (107,233/110,000), 15822.358 ms/call; hold-out (fashion) 86.78%, 15830.109 ms/call; score 15830.109 ms energy: one more fresh process runs the method back to back on fresh MNIST draws for 20 s, between idle windows in which it is frozen NVIDIA A100-SXM4-80GB: idle 68.5 W with the method frozen; telemetry reference 18.2 TFLOP/s at 8.14 J/TFLOP above idle (a healthy NVIDIA A100-SXM4-80GB reads 6-11) the round trip alone: 15.3 mJ per call over 1,808 empty calls, subtracted the method: 3 calls in 47.5 s, MNIST 97.49%, 15,813.391 ms per call, 141.8 W on average energy 1159909.313 mJ per call above idle --- run 2: ladder_d3.py:ladder at 2.70% error: 15 timed calls, each in a fresh process NVIDIA A100-SXM4-80GB, torch 2.12.0+cu130, sandbox on call 1/15: 15,775.480 ms, scorer's clock 15,778.989 ms call 2/15: 15,773.674 ms, scorer's clock 15,777.151 ms call 3/15: 15,769.851 ms, scorer's clock 15,773.517 ms call 4/15: 15,771.887 ms, scorer's clock 15,776.181 ms call 5/15: 15,772.979 ms, scorer's clock 15,782.494 ms call 6/15: 15,771.438 ms, scorer's clock 15,783.579 ms call 7/15: 15,772.114 ms, scorer's clock 15,785.143 ms call 8/15: 15,775.615 ms, scorer's clock 15,787.690 ms call 9/15: 15,773.681 ms, scorer's clock 15,786.210 ms call 10/15: 15,772.441 ms, scorer's clock 15,783.449 ms call 11/15: 15,770.346 ms, scorer's clock 15,783.012 ms call 12/15: 15,771.514 ms, scorer's clock 15,784.393 ms call 13/15: 15,771.284 ms, scorer's clock 15,783.635 ms call 14/15: 15,773.214 ms, scorer's clock 15,785.264 ms call 15/15: 15,769.290 ms, scorer's clock 15,782.119 ms MNIST 97.34% (107,070/110,000), 15771.693 ms/call; hold-out (kmnist) 96.24%, 15774.046 ms/call; score 15774.046 ms energy: one more fresh process runs the method back to back on fresh MNIST draws for 20 s, between idle windows in which it is frozen NVIDIA A100-SXM4-80GB: idle 70.3 W with the method frozen; telemetry reference 17.8 TFLOP/s at 8.02 J/TFLOP above idle (a healthy NVIDIA A100-SXM4-80GB reads 6-11) the round trip alone: 16.4 mJ per call over 2,255 empty calls, subtracted the method: 3 calls in 47.3 s, MNIST 97.37%, 15,738.514 ms per call, 142.8 W on average energy 1142580.269 mJ per call above idle ========== == CUDA == ========== CUDA Version 13.3.0 Container image Copyright (c) 2016-2023, NVIDIA CORPORATION & AFFILIATES. All rights reserved. ========== == CUDA == ========== CUDA Version 13.3.0 Container image Copyright (c) 2016-2023, NVIDIA CORPORATION & AFFILIATES. All rights reserved. --- run 3: ladder_d3.py:ladder at 2.70% error: 15 timed calls, each in a fresh process NVIDIA A100-SXM4-80GB, torch 2.12.0+cu130, sandbox on call 1/15: 15,803.351 ms, scorer's clock 15,807.218 ms call 2/15: 15,806.382 ms, scorer's clock 15,809.662 ms call 3/15: 15,806.217 ms, scorer's clock 15,809.745 ms call 4/15: 15,810.177 ms, scorer's clock 15,813.969 ms call 5/15: 15,812.559 ms, scorer's clock 15,815.890 ms call 6/15: 15,816.638 ms, scorer's clock 15,820.136 ms call 7/15: 15,812.764 ms, scorer's clock 15,816.099 ms call 8/15: 15,817.057 ms, scorer's clock 15,820.909 ms call 9/15: 15,825.322 ms, scorer's clock 15,828.916 ms call 10/15: 15,828.729 ms, scorer's clock 15,832.605 ms call 11/15: 15,825.248 ms, scorer's clock 15,829.254 ms call 12/15: 15,825.034 ms, scorer's clock 15,828.742 ms call 13/15: 15,824.457 ms, scorer's clock 15,828.251 ms call 14/15: 15,825.256 ms, scorer's clock 15,828.544 ms call 15/15: 15,824.022 ms, scorer's clock 15,827.307 ms MNIST 97.50% (107,250/110,000), 15818.770 ms/call; hold-out (kmnist) 96.00%, 15814.185 ms/call; score 15818.770 ms energy: one more fresh process runs the method back to back on fresh MNIST draws for 20 s, between idle windows in which it is frozen NVIDIA A100-SXM4-80GB: idle 63.7 W with the method frozen; telemetry reference 18.1 TFLOP/s at 8.43 J/TFLOP above idle (a healthy NVIDIA A100-SXM4-80GB reads 6-11) the round trip alone: 13.9 mJ per call over 2,089 empty calls, subtracted the method: 3 calls in 47.6 s, MNIST 97.51%, 15,830.407 ms per call, 137.1 W on average energy 1162985.570 mJ per call above idle Stopping app - local entrypoint completed. ✓ App completed. View run at https://modal.com/apps/seth-45699/main/ap-RsMkG2HydYTyWwERlflduX passed 3 of 3 runs; median 15818.770 ms; energy median 1,159,909.3 mJ per call over 3 runs EXIT 0