Update TRAINING.md

This commit is contained in:
Victor Hall 2023-01-10 21:09:23 -05:00 committed by GitHub
parent 8a528e9ab1
commit 6c03a6edc8
No known key found for this signature in database
GPG Key ID: 4AEE18F83AFDEB23
1 changed files with 3 additions and 1 deletions

View File

@ -18,6 +18,8 @@ You can edit the example `train.json` file to your liking, then run the followin
Be careful with editing the json file, as any syntax errors will cause the program to crash. You might want to use a json validator to check your file before running it. You can use an online validator such as https://jsonlint.com/ or look at it in VS Code. Be careful with editing the json file, as any syntax errors will cause the program to crash. You might want to use a json validator to check your file before running it. You can use an online validator such as https://jsonlint.com/ or look at it in VS Code.
One particular note is if your path to data_root or resume_ckpt has backslashes they need to use double \\ or single /. There is an example train.json in the repo root.
## Running from the command line with arguments ## Running from the command line with arguments
I recommend you copy one of the examples below and keep it in a text file for future reference. Your settings are logged in the logs folder, but you'll need to make a command to start training. I recommend you copy one of the examples below and keep it in a text file for future reference. Your settings are logged in the logs folder, but you'll need to make a command to start training.
@ -80,4 +82,4 @@ Or use relative pathing:
```--resume_ckpt "logs\myproj20221213-161620\ckpts\myproj-ep22-gs01099" ^``` ```--resume_ckpt "logs\myproj20221213-161620\ckpts\myproj-ep22-gs01099" ^```
You should point to the folder in the logs per above if you want to resume rather than running a conversion back on a 2.0GB or 2.5GB pruned file if possible. You should point to the folder in the logs per above if you want to resume rather than running a conversion back on a 2.0GB or 2.5GB pruned file if possible.