Move server-specific config args out of ModelParams and into new ServerParams in config_struct.py #877
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Problem
Currently, prefix sharing is read as a config key from the exported-model's config.json.
There is no reason for it to be there - - this is not a property of the model, but rather a property of the server instance that's being launched.
This overcomplicates the tests: rather than just inserting an additional cmdline arg, the tests have to make an edited copy of the exported config files in order to specify the prefix-sharing algorithm.
Solution
This PR puts in a ServerParam class for storing config options that don't require re-exporting the model. Right now it just affects the prefix-sharing algorithm but this makes space for things like:
Why this?
Deepseek v3 will have many run-time options that aren't relevant at export/compile, e.g. MTP. These changes will make it easier to set those up.