Skip to content

Conversation

@p1-0tr
Copy link
Contributor

@p1-0tr p1-0tr commented Aug 14, 2025

In case a segmented GGUF file does not follow the specified naming convention, it will not be possible to load it. So, modify the argument parser to allow repeated --model args to be specified on the CLI, and in such case assume those are GGUF splits given in order.

In case a segmented GGUF file does not follow the specified naming
convention, it will not be possible to load it. So, modify the argument
parser to allow repeated --model args to be specified on the CLI, and in
such case assume those are GGUF splits given in order.

Signed-off-by: Piotr Stankiewicz <[email protected]>
@p1-0tr p1-0tr requested a review from ngxson as a code owner August 14, 2025 15:28
@CISC
Copy link
Collaborator

CISC commented Aug 14, 2025

TBH this feels like an anti-feature, in which scenario would the files be incorrectly named, and why can't they just be renamed?

@github-actions github-actions bot added testing Everything test related examples server labels Aug 14, 2025
@ngxson
Copy link
Collaborator

ngxson commented Aug 14, 2025

Same thought with @CISC , I think we should not risk polluting the code base with a feature that less than 1% of users use.

If you cannot rename the file, you can also symlink it to make correct the file name.

The API for loading shards with custom naming was for downstream apps that use llama.cpp as a lib, they can manage files by naming them as SHA256 string for example. It is NOT intended to use by llama.cpp CLI

@p1-0tr
Copy link
Contributor Author

p1-0tr commented Aug 15, 2025

I think we should not risk polluting the code base with a feature that less than 1% of users use.

Fair enough.

@p1-0tr p1-0tr closed this Aug 15, 2025
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

examples server testing Everything test related

Projects

None yet

Development

Successfully merging this pull request may close these issues.

3 participants