Nerogar
[Bug]: ambiguity in embedding docs
What happened?
Trying to follow https://github.com/Nerogar/OneTrainer/blob/master/docs/EmbeddingTraining.md
It is unclear whether to really use the literal string '<embedding>', or whether you are expected to replace that tag with the actual name of your embedding.
If it is supposed to be literal.. then where does the name of the embedding get set?
What did you expect would happen?
More explicit wording
Relevant log output
No response
Output of pip freeze
No response
You need to add the literal <embedding> to the prompt. This text will be replaced by the trained embedding. This will soon change, you will be able to set a custom text. The name of the final embedding is just the file name.
If it is supposed to be literal.. then where does the name of the embedding get set?
I don't understand this question.
So, looking at the embeddings template file, it is filled with stuff like
"A photo of a <embedding>"
In other contexts, with other programs, if you are creating a (something) with name "mycoolthing", then you might expect
"A photo of a <embedding>" to be replaced with "A photo of a mycoolthing"
So if you are coming from an outsiders perspective, this is confusing, unless you explicitly say something like, " 'embedding' will automatically get replaced by a mathematical representation of your embedding"
If that is actually correct? :)
It is better explained in the Wiki, check the concept, embedding and FAQ pages, you also can jump in the Discord server for further questions.
Just that <embedding> works as a placeholder for training and sampling, the trigger word like "mycoolclothing" is the output model name (mycoolclothing.safetensors).
From the wiki FAQ:
- For embeddings, the trigger word is the embedding name, if it is TomCruise:
- Do you add <TomCruise> to captions or
? Answer: - Do you add <TomCruise> or TomCruise to embedding-tab -> Initial embedding text? Answer: a brief description of your subject to help your embedding to train faster, so just "*" or "man" or "short man" ...
- Do you add <TomCruise> or TomCruise or
to sampling-tab? Answer: - Note: TomCruise is only set as the output model name: TomCruise.safetensors.
erm.. thanks for replying...but.. your cutnpaste is really hard to read and understand
Then please go to the FAQ on the wiki for the original information. https://github.com/Nerogar/OneTrainer/wiki/F.A.Q.
Sigh. The auto-linker in guthub got confused by your reference (Maybe you guys wanna consider changing the name of the page to just "FAQ", no trailing "."