shng2025's picture
Update README.md
192acac verified
|
raw
history blame
1.19 kB
metadata
license: mit
  • Model is directly from pytorch. Refer to the python file. To reuse, use .load_state_dict() from the .pth file. Good luck.

Training steps:

  • step 0: train loss 4.2221, val loss 4.2306

  • step 500: train loss 1.7526, val loss 1.9053

  • step 1000: train loss 1.3949, val loss 1.6050

  • step 1500: train loss 1.2625, val loss 1.5219

  • step 2000: train loss 1.1860, val loss 1.5046

  • step 2500: train loss 1.1254, val loss 1.4972

  • step 3000: train loss 1.0694, val loss 1.4849

  • step 3500: train loss 1.0211, val loss 1.5048

  • step 4000: train loss 0.9643, val loss 1.5160

  • step 4500: train loss 0.9121, val loss 1.5396

  • step 5000: train loss 0.8673, val loss 1.5552

  • step 5500: train loss 0.8052, val loss 1.5988

  • step 6000: train loss 0.7611, val loss 1.6231

  • step 6500: train loss 0.7087, val loss 1.6706

  • step 7000: train loss 0.6644, val loss 1.7000

  • step 7500: train loss 0.6187, val loss 1.7484

  • step 8000: train loss 0.5818, val loss 1.7882

  • step 8500: train loss 0.5350, val loss 1.8304

  • step 9000: train loss 0.4973, val loss 1.8688

  • step 9500: train loss 0.4638, val loss 1.9050

  • step 9999: train loss 0.4333, val loss 1.9475