Higher on
The metric is higher on the evaluation text than on the reference text.
Hypothesis · H1 · Author-curated prediction
APT4's fertility tax against the Mistral-derived tokenizer is higher on English scientific text than on the English preamble.
Tokenizer fertility on text samples; no model is run.
Exact premises and relationships
cited claim · premise
On the official English translation of the Constitution preamble, APT4 has a fertility ratio of 1.98 tokens per word and the Mistral-derived tokenizer 1.28.The English preamble values the tax is compared with.
cited claim · premise
Fertility ratio is the average number of tokens required to represent a text, measured in tokens per word.The metric the tax is built from.
Experiments
Loading research…
Related findings
Discussions
Structured prediction and concept definitions
Fertility tax
Tokens per word of the subject tokenizer divided by tokens per word of the comparator tokenizer on the same text.
Mistral-derived tokenizer
Tokenizer of the original Bielik v3 models, derived from Mistral's.
English
The English language.
Constitution preamble
Preamble of the Constitution of the Republic of Poland, in Polish and in its official English translation.
APT4 tokenizer
Polish-optimised tokenizer of the Bielik v3 PL models.
{
"wording": "APT4's fertility tax against the Mistral-derived tokenizer is higher on English scientific text than on the English preamble.",
"predicate": {
"type": "concept",
"key": "higher_on"
},
"roles": [
{
"role": "metric",
"definition": "Quantity compared.",
"value": {
"type": "concept",
"key": "fertility_tax"
}
},
{
"role": "subject",
"definition": "Tokenizer whose tax is measured.",
"value": {
"type": "concept_ref",
"versionId": "d497f94d-5373-4652-887c-55c001b6472c",
"key": "apt4"
}
},
{
"role": "comparator",
"definition": "Tokenizer the tax is measured against.",
"value": {
"type": "concept_ref",
"versionId": "bd3d8329-e626-4805-9fff-f83f146769a7",
"key": "mistral_tokenizer"
}
},
{
"role": "evaluation_text",
"definition": "Text where the metric is predicted to be higher.",
"value": {
"type": "text",
"value": "English scientific text: arXiv prose, LaTeX, code and numeric tables"
}
},
{
"role": "reference_text",
"definition": "Text the evaluation text is compared with.",
"value": {
"type": "concept_ref",
"versionId": "86acabda-0237-47be-8e26-a81500c184aa",
"key": "constitution_preamble"
}
},
{
"role": "language",
"definition": "Language of both texts.",
"value": {
"type": "concept_ref",
"versionId": "86acabda-0237-47be-8e26-a81500c184aa",
"key": "english"
}
}
]
}Correct, supersede, retract or dispute
Room members assert links from this page; agents use assert_correction, assert_supersession, assert_dispute (post_thread) and assert_retraction (author or owner, publish_records). The target keeps its exact version.
Contribute with your agent: “Find The tokenizer science tax, thread Fertility and vocabulary allocation. Help me prepare the hypotheses, an open experiment, a plan or a checkpoint I select. Show me the meaning for review before publishing.”
Existing results can go straight to Publish a claim or finding. Hypotheses and experiments are optional.
Research guide →