Closed sgerke-ST closed 2 years ago
Hi, I ran this on my Gigabyte Aero 15 with an RTX 3060 (6 GB VRAM) and Intel i7=10870H. I had to set the batch size to 8 due to memory constraints on that GPU.
device cuda Files already downloaded and verified Using cache found in /home/gerke/.cache/torch/hub/pytorch_vision_v0.11.0 Epoch: 001/001 | Batch 0000/5625 | Loss: 2.9805 Epoch: 001/001 | Batch 0100/5625 | Loss: 2.4223 Epoch: 001/001 | Batch 0200/5625 | Loss: 2.2558 Epoch: 001/001 | Batch 0300/5625 | Loss: 2.3670 Epoch: 001/001 | Batch 0400/5625 | Loss: 2.4191 Epoch: 001/001 | Batch 0500/5625 | Loss: 2.2816 Epoch: 001/001 | Batch 0600/5625 | Loss: 2.4255 Epoch: 001/001 | Batch 0700/5625 | Loss: 2.2249 Epoch: 001/001 | Batch 0800/5625 | Loss: 2.3222 Epoch: 001/001 | Batch 0900/5625 | Loss: 2.2943 Epoch: 001/001 | Batch 1000/5625 | Loss: 2.2927 Epoch: 001/001 | Batch 1100/5625 | Loss: 2.3051 Epoch: 001/001 | Batch 1200/5625 | Loss: 2.0943 Epoch: 001/001 | Batch 1300/5625 | Loss: 2.3638 Epoch: 001/001 | Batch 1400/5625 | Loss: 2.2915 Epoch: 001/001 | Batch 1500/5625 | Loss: 2.3452 Epoch: 001/001 | Batch 1600/5625 | Loss: 2.3329 Epoch: 001/001 | Batch 1700/5625 | Loss: 2.3025 Epoch: 001/001 | Batch 1800/5625 | Loss: 2.2956 Epoch: 001/001 | Batch 1900/5625 | Loss: 2.2994 Epoch: 001/001 | Batch 2000/5625 | Loss: 2.2964 Epoch: 001/001 | Batch 2100/5625 | Loss: 2.3315 Epoch: 001/001 | Batch 2200/5625 | Loss: 2.3060 Epoch: 001/001 | Batch 2300/5625 | Loss: 2.3055 Epoch: 001/001 | Batch 2400/5625 | Loss: 2.2876 Epoch: 001/001 | Batch 2500/5625 | Loss: 2.2982 Epoch: 001/001 | Batch 2600/5625 | Loss: 2.3192 Epoch: 001/001 | Batch 2700/5625 | Loss: 2.3176 Epoch: 001/001 | Batch 2800/5625 | Loss: 2.3281 Epoch: 001/001 | Batch 2900/5625 | Loss: 2.3009 Epoch: 001/001 | Batch 3000/5625 | Loss: 2.3031 Epoch: 001/001 | Batch 3100/5625 | Loss: 2.3171 Epoch: 001/001 | Batch 3200/5625 | Loss: 2.3002 Epoch: 001/001 | Batch 3300/5625 | Loss: 2.2863 Epoch: 001/001 | Batch 3400/5625 | Loss: 2.3257 Epoch: 001/001 | Batch 3500/5625 | Loss: 2.3163 Epoch: 001/001 | Batch 3600/5625 | Loss: 2.3172 Epoch: 001/001 | Batch 3700/5625 | Loss: 2.3052 Epoch: 001/001 | Batch 3800/5625 | Loss: 2.3109 Epoch: 001/001 | Batch 3900/5625 | Loss: 2.3230 Epoch: 001/001 | Batch 4000/5625 | Loss: 2.2976 Epoch: 001/001 | Batch 4100/5625 | Loss: 2.3131 Epoch: 001/001 | Batch 4200/5625 | Loss: 2.2884 Epoch: 001/001 | Batch 4300/5625 | Loss: 2.3075 Epoch: 001/001 | Batch 4400/5625 | Loss: 2.3074 Epoch: 001/001 | Batch 4500/5625 | Loss: 2.2968 Epoch: 001/001 | Batch 4600/5625 | Loss: 2.3173 Epoch: 001/001 | Batch 4700/5625 | Loss: 2.2763 Epoch: 001/001 | Batch 4800/5625 | Loss: 2.3128 Epoch: 001/001 | Batch 4900/5625 | Loss: 2.3160 Epoch: 001/001 | Batch 5000/5625 | Loss: 2.3110 Epoch: 001/001 | Batch 5100/5625 | Loss: 2.3077 Epoch: 001/001 | Batch 5200/5625 | Loss: 2.3121 Epoch: 001/001 | Batch 5300/5625 | Loss: 2.2908 Epoch: 001/001 | Batch 5400/5625 | Loss: 2.3096 Epoch: 001/001 | Batch 5500/5625 | Loss: 2.2987 Epoch: 001/001 | Batch 5600/5625 | Loss: 2.2882 Time / epoch without evaluation: 14.53 min Epoch: 001/001 | Train: 10.03% | Validation: 9.76% | Best Validation (Ep. 001): 9.76% Time elapsed: 18.63 min Total Training Time: 18.63 min Test accuracy 10.00% Total Time: 19.46 min
Thanks, I included it in the updated results at https://sebastianraschka.com/blog/2022/pytorch-m1-gpu.html
Hi, I ran this on my Gigabyte Aero 15 with an RTX 3060 (6 GB VRAM) and Intel i7=10870H. I had to set the batch size to 8 due to memory constraints on that GPU.