Initial upload: s4_jobbert_haiku (paper release 2026-04)

README.md

@@ -25,7 +25,7 @@ Distilled Named Entity Recognition model for English-language job postings. One
 - **Student identifier:** `s4_jobbert_haiku`
 - **Artefact size:** ~820 MB
 
-Latency is materially lower than the spaCy students (~15 ms vs ~44 ms per document on the gold set). F1 is materially lower because the 512-token window cannot reach the trailing part of most postings.
+Latency is materially lower than the spaCy students (~15 ms vs ~44 ms per document on the gold set). F1 is materially lower on the full distribution because the 512-token window cannot reach the trailing part of most postings; on shorter text the model recovers the full entity set.
 
 ## Intended use
 
@@ -70,6 +70,8 @@ Sonnet-trained students evaluate on all 516 gold postings; Haiku-trained student
 
 ### Per-entity-type
 
+Per-entity numbers below reflect the coverage constraint as much as the model's per-type quality. Entities that appear only in the trailing part of a long posting (typically `CERT`, `EDUCATION`, `EXPERIENCE_LEVEL`, and `COMPENSATION` in many English templates) are systematically outside the model's input window and therefore missed at the recall metric even when the model would classify them correctly on shorter text. For full-text coverage, use the spaCy variant.
+
 | Entity type | P | R | F1 |
 |---|---|---|---|
 | COMPANY | 0.636 | 0.470 | 0.540 |
@@ -99,8 +101,22 @@ text = 'Senior Machine Learning Engineer at Acme Corp in Berlin. Requires 5+ yea
 for ent in ner(text):
     print(ent["word"], "->", ent["entity_group"])
 
-# Note: the BERT base tokeniser has a 512-token window; very long postings
-# are truncated. For full-text coverage, use the spaCy variant.
+# Produces (verified on this release; note the BERT wordpiece tokenisation
+# artefacts in numeric spans):
+# Senior Machine Learning Engineer     -> JOB_TITLE         (0.70)
+# Acme Corp                            -> COMPANY           (0.76)
+# Berlin                               -> LOCATION          (0.63)
+# 5 + years of experience              -> EXPERIENCE_LEVEL  (0.76)
+# PyTorch                              -> SKILL             (0.60)
+# AWS                                  -> SKILL             (0.63)
+# Kubernetes                           -> SKILL             (0.67)
+# MSc in Computer Science              -> EDUCATION         (0.62)
+# $ 140, 000 – $ 180, 000              -> COMPENSATION      (0.93)
+
+# Note: the BERT base tokeniser has a 512-token window. The example text here
+# is short (well under the limit); on a full job posting of ~4,000 characters,
+# the tail is truncated and entities there are systematically missed. For
+# full-text coverage, use the spaCy variant.
 ```
 
 ## Ethical considerations