v2.17.0 — validate a --quantize-embeddings <mode> value. Accepts the
canonical "f32" / "int8" plus a few user-friendly aliases ("none"
for f32; "q8" for int8). Anything else throws with the exact list of
accepted values so the user can fix the typo immediately.
v2.17.0 — validate a
--quantize-embeddings <mode>value. Accepts the canonical"f32"/"int8"plus a few user-friendly aliases ("none"for f32;"q8"for int8). Anything else throws with the exact list of accepted values so the user can fix the typo immediately.