Feat/add modelscope -ms args support

[yrk111222]

Overview

This PR adds ModelScope (-ms) option support. Based on the feedback #22279 about keeping responsibilities separated and following the hf-cache.cpp pattern

Additional information

Key Design Decisions

1. Downloading, quantization selection, and filtering logic stays in download.cpp

Following the reviewer's request, we did not duplicate these functions in ms-cache.cpp. Instead, the existing functions in download.cpp were templatized so they can be shared by both HF and MS. This keeps ms-cache.cpp focused on caching only — mirroring hf-cache.cpp's responsibilities. The templates are defined entirely within download.cpp, with no header changes, so HF behavior is unaffected.

2. ms-cache.cpp follows the pattern of hf-cache.cpp closely

ms_file struct mirrors hf_file (identical fields)
Public API matches hf-cache.h (get_repo_files, get_cached_files, finalize_file)
Same blob/snapshot/symlink cache structure
The only differences are platform-specific (unavoidable):
Auth: Cookie vs Bearer
Endpoint: dedicated ModelScope endpoint
Snapshot key: "master" (MS API has no single repo-wide commit hash)
oid: Sha256 from API

3. Cache path aligned with HF + token-less loading

MS uses refs/master as the snapshot key. Since MS files use regular filenames requiring ETag validation (unlike HF's content-addressed blobs), cached MS models need extra handling to load without a token — just like HF. When all files are cached, download tasks are skipped and cached paths are returned directly.

Usage

# Auto-select best model
llama-cli -ms Qwen/Qwen3-0.6B-GGUF -p "hello"

# Private repo
export MS_TOKEN=<token>
llama-cli -ms <owner>/<private-model>

Requirements

• I have read and agree with the contributing guidelines
• AI usage disclosure: YES

I used an LLM to translate the PR description and do some code review.

[ggml-gh-bot]

Hi @yrk111222, thanks for your contribution!

Per our contribution guidelines, the automated PR checker found the following issue(s) that need your attention:

• AI-generated content: This project does not accept PRs, descriptions or commit messages that are fully or predominantly AI-generated. If you have used AI to assist you in writing code, please make sure to disclose that explicitly.

---

Please note that maintainers reserve the right to make final decisions on PRs. If you believe there is a mistake, please comment below.

[yrk111222]

HI @angt:
Rebased onto the latest master and resolved all conflicts. PR is now ready for review — I'd really appreciate your feedback or any suggestions!😊

[angt]

@yrk111222 That looks much better indeed, well done!
I'll try to do the the final review this week 👍

[yrk111222]

Hi @angt,

I noticed the recent large refactor to the download code, and this PR has fallen a bit behind. To minimize rework, would you rather I rebase it against the current state now, or hold off until things stabilize? Happy to follow your lead on this—just let me know what works best.

Thanks!😊

[angt]

Hi @angt,

I noticed the recent large refactor to the download code, and this PR has fallen a bit behind. To minimize rework, would you rather I rebase it against the current state now, or hold off until things stabilize? Happy to follow your lead on this—just let me know what works best.

Thanks!😊

Yes we are sorry for that, and I think we still need some refactoring.
But i believe that rebasing at each step might be easier to follow the process for you 🤔

[yrk111222]

Hi @angt,

Rebased onto the latest master and synced the MS code with the download refactor. Build + tests pass.
Whenever you have time, could you take another look and let me know if there's anything that still needs changing?

Thanks!😊

[yrk111222]

Hi @angt, rebased on latest master and aligned MS with the HF download pipeline changes. Key updates since last review:

• Aligned dflash/eagle3 sidecar support for MS
• Fixed server.cpp to include ms_repo in models_handler init (was HF-only)
• Fixed app/download.cpp to include ms_repo in has_source check
• Added ms_cache::remove_cached_repo and extended cache list/remove to MS

Would appreciate another review when you have a moment.😊

[yingdachen]

@angt could you please take another look and (hopefully) get this merged soon? this has been outstanding for a (unfortunate) long period of time...

thanks!

[yingdachen]

a gentle ping again @angt

hope we can get this merged before we need to resolve conflict again ... :(

[yrk111222]

Hi @angt, gentle ping again. Rebased on latest master and resolved conflicts. Is the refactoring stabilized enough for another PR review? Happy to address any feedback promptly... : )