countrycode-small-v1.2
Browse files
README.md
CHANGED
|
@@ -168,7 +168,8 @@ Browser: onnxruntime-web computes the 8-bit matrix products in fp32 where the CP
|
|
| 168 |
- One human reviewer labelled the register strings; the calibration strings were labelled twice by
|
| 169 |
that reviewer, the second time blind (consistency, not inter-annotator agreement).
|
| 170 |
- Certified error rates hold for the population they were fitted on; recalibrate before relying on
|
| 171 |
-
them for a different source.
|
|
|
|
| 172 |
model, and there a 35B LLM is more accurate.
|
| 173 |
- Ambiguous names are returned as candidate lists; downstream systems must not silently pick one.
|
| 174 |
- Known gaps: Canadian province codes in US jurisdiction fields (`ON`, `BC`) are not recognised;
|
|
|
|
| 168 |
- One human reviewer labelled the register strings; the calibration strings were labelled twice by
|
| 169 |
that reviewer, the second time blind (consistency, not inter-annotator agreement).
|
| 170 |
- Certified error rates hold for the population they were fitted on; recalibrate before relying on
|
| 171 |
+
them for a different source. Each target is certified on its own (with probability ≥ 90 % over the
|
| 172 |
+
calibration draw); the 2 % and 5 % guarantees are not simultaneous. The curated GBIF domain is much harder than registers for this
|
| 173 |
model, and there a 35B LLM is more accurate.
|
| 174 |
- Ambiguous names are returned as candidate lists; downstream systems must not silently pick one.
|
| 175 |
- Known gaps: Canadian province codes in US jurisdiction fields (`ON`, `BC`) are not recognised;
|