Multi-Domain Expert Learning (M*DEL) is an approach to training LLMs for expertise in knowledge domains.
This process involves branching from a base model and training the branch on specific domain data in order to establish an expert layer, which routing logic is able to activate in serving inference requests. The result is a framework for domain expertise that is easily-extensible, modular, and efficient.
...
| Info |
|---|
To use example data for training, see the |
Next, make a configuration file to use for this training run with a meaningful filename; here, it is called "experiment1":
...
| Code Block | ||
|---|---|---|
| ||
$ huggingface-cli upload [repo-id] [local-path] [path-in-repo] --token=[read/write tokentoken with WRITE permission] |
For example:
| Code Block | ||
|---|---|---|
| ||
$ huggingface-cli upload stillerman/aurora-mathematica ./lora-out/ . --token=... |
This uploads the model to the Hub at HuggingFace and is public, by default. Add the --private flag to make it private.