kubeflow / mcp-server / 39483202716
87%

DEFAULT BRANCH: main
Ran
Jobs 3
Files 38
Run time 1min
Badge
Embed ▾
README BADGES
x

If you need to use a raster PNG badge, change the '.svg' to '.png' in the link

Markdown

Textile

RDoc

HTML

Rst
coverage: 86.901% (+0.4%) from 86.533%
39483202716

push

github

web-flow 
fix(trainer): validate Hugging Face model and dataset references in fine_tune (#320)

* fix(trainer): validate Hugging Face model and dataset references in fine_tune

fine_tune() accepted any hf:// model and dataset. A nonexistent repo passed
preview, and with an unresolved runtime the tool returned an unrelated
"Runtime '' does not support BuiltinTrainer" error.

Check hf:// reference format in preview (no whitespace, valid repo segments,
models may be unscoped like hf://gpt2, datasets need org/name). On confirmed
submission, a repo the Hub confirms missing (404, not gated) is a
VALIDATION_ERROR with suggestions for models; private, gated or unreachable
repos do not block. The existence check needs HF_TOKEN on the MCP host, which
the fine_tune docstring and troubleshooting guide now say.

Rebased on main; squashed the earlier review rounds.

Fixes #259

Signed-off-by: pratik naik <naikp0241@gmail.com>

* fix(trainer): validate fine_tune input before resolving the namespace

Review feedback on #320: _validate_fine_tune_params called
check_namespace_allowed() first. With an allowlist and no namespace, that
resolves the default through the Trainer SDK before malformed hf:// input was
rejected, against #259's validation order.

Run all local checks (name, hf:// format, dtype, loss, dataset_source, bounds)
first, then the namespace check, then the GPU preflight, then the Hub check on
submission. Test that bad hf:// references never resolve the namespace, in
preview and on submission.

Signed-off-by: pratik naik <naikp0241@gmail.com>

---------

Signed-off-by: pratik naik <naikp0241@gmail.com>

2866 of 3298 relevant lines covered (86.9%)

2.61 hits per line

Coverage Regressions

Lines Coverage ∆ File
87
49.65
2.51% trainer/api/planning.py
42
87.15
1.51% trainer/api/training.py
Jobs
ID Job ID Ran Files Coverage
1 python-3.11 - 39483202716.1 38
86.9
GitHub Action Run
2 python-3.12 - 39483202716.2 38
86.9
GitHub Action Run
3 python-3.10 - 39483202716.3 38
86.9
GitHub Action Run
Source Files on build 39483202716