Finding

Sign in with GitHub
← Publications

Finding · P155 · Author-curated

Qwen2.5-0.5B has 0.0744 bits per byte on GSM8K problem statements and at least 0.4046 bits per byte on each of the other nine evaluation texts.

Published by @stw2 · 2026-09-14 · Sources, measurements and interpretation are supplied by the author.

Structured assertion

Relation: Lowest text value

model
Qwen2.5-0.5BModel scored.
metric
Bits per byteMetric.
text
GSM8K problemsText with the lowest value.
value
0.0744 bits per byteMetric on text.
other texts minimum
0.4046 bits per byteLowest value on the other nine texts.

Experimental provenance

Method and evaluation protocol
Bits per byte of Qwen2.5-0.5B on the first 200 documents of each of ten texts.
Dataset
The first 200 documents of ten evaluation texts: Polish and English web holdouts, English arXiv abstracts, LaTeX method sections, Python code and GSM8K problems, Polish Wikipedia science articles, PES questions and reviews, math_clean.Version: unspecified · Access: restricted
Reported results
GSM8K problems 0.0744; lowest other text 0.4046 (English Python code).
Uncertainty and replication
Point values without intervals.
Limitations
A low value is consistent with the model having seen the GSM8K training split, but no contamination test was run; the GSM8K text serves as a diagnostic, not an endpoint. Dataset and base-model revisions were not recorded in the original run; the pinned revisions are inferred.

Paper citations

arXiv:2604.10799v1 →Revision supplied by author

Associated through an explicitly referenced cited claim.

Concept definitions

Reuse the defining version and key when the meaning fits your assertion.

Lowest text value

model's metric on text is value, and its metric on each of the other evaluated texts is at least other_texts_minimum.

Key lowest_text_value · version 72a9fbb8-0a40-4519-87c2-6b68769e15b1

Bits per byte

Summed next-token negative log-likelihood of a text in bits divided by the text's UTF-8 byte count.

Key bits_per_byte · version 9a53082a-65e8-4c6a-82be-33dda5810584

Qwen2.5-0.5B

The Qwen2.5 base language model with 0.5B parameters, Qwen/Qwen2.5-0.5B.

Key qwen2_5_0_5b · version f89e740e-7204-4148-8934-85fbc73f323c

GSM8K problems

The 7,473 questions of the GSM8K training split.

Key english_gsm8k_problems · version 4ab4fa55-c833-4de4-bec3-4d5b68245a5e

Exact references

related

36da68d9-97cf-4575-ad42-ed56e0e895e9

GSM8K is the benchmark of the regression; its problem statements are one of the evaluation texts.

extends

4ac9d8db-adb9-49c5-b0dd-e12ea68e549b

The same pattern for Qwen2.5-1.5B on the same ten texts.