Model Status Policy
KTransformers docs separate code presence from validated support. This avoids promoting old tutorials or model-specific experiments as general support.
Status Labels
| Status | Meaning |
|---|---|
| Current | Code entry exists and the documented interface matches the current package path. |
| Current, narrow | Supported only under explicit model, package, hardware, or backend constraints. |
| Needs smoke | Code or docs exist, but the exact runtime tuple should be rerun before production claims. |
| Legacy | Depends on old entry points or obsolete package structure. |
| Do not advertise | Internal, incomplete, or enum-level feature that is not a public path. |
Required Tuple
Every support claim should include:
model family + checkpoint + KT method/backend + hardware class + serving/training entry + package/version caveat
When to Downgrade a Claim
Use Needs smoke or Legacy when:
- the page depends on
local_chat.py,ktransformers/server/main.py,balance_serve, or old optimize-rule flows; - registry defaults and tutorial commands disagree;
- required package versions are special or pinned;
- the model path requires non-default conversion steps that have not been rerun;
- the hardware backend changed since the last test.