microsoft / microsoft/winml-cli
Honor cache controls for GenAI runtime compilation
Nobody has claimed this yet.
- Dominant language
- Python
- Stars
- 40
- Forks
- 11
- Avg merge
- 1d 8h
- Merged PRs (30d)
- 50
Description
Problem
eval and perf --runtime winml-genai can compile pre-built GenAI bundles into a persistent _compiled/ directory, but the shared model cache controls do not govern that runtime compilation cache.
--rebuildcannot currently force regeneration of_compiled/.--no-use-cachecannot currently compile into a disposable directory without touching the bundle cache.- The behavior differs from the model-artifact cache controlled by the same flags.
Scope
- Define cache/rebuild semantics for
GenaiSessionruntime compilation. - Thread the resolved policy through
WinMLGenaiCausalLM,winml eval, andwinml perf --runtime winml-genai. - Ensure
--rebuildreplaces persistent compiled artifacts. - Ensure
--no-use-cacheuses disposable compiled artifacts and leaves_compiled/untouched. - Add focused tests for cache reuse, forced rebuild, temporary compilation, and pre-built bundles.
- Keep model build caching and runtime compilation caching clearly distinguished in warnings and documentation.
Contributor guide
First steps
- Read the whole issue, then the project's contributing guide.
- Comment on the issue to say you are picking it up — it saves two people doing the same work.
- Fork the repository and make your change on a branch.
- Open a pull request that references the issue number.
Research direction
Start by tracing cache policy through GenaiSession and WinMLGenaiCausalLM, then follow how winml eval and winml perf --runtime winml-genai invoke runtime compilation. Add focused tests for cache reuse, forced rebuild, temporary compilation, and pre-built bundles. Done means --rebuild replaces persistent artifacts, --no-use-cache leaves _compiled/ untouched, and warnings and documentation distinguish the two caches.
Written by the indexing model from the issue text.
Assessment
- Tech stack
- python
- Domain
- build-system, cli
- Issue type
- Feature
- Difficulty
- 4/5
- Estimated time
- 3-5 days
- Activity status
- Quiet
- Clarity
- Mostly clear
- Newbie friendliness
- 48/100