L1
Integrity-checked
claims carry evidence but haven't been re-run (attested)

BERT-base SST-2 fine-tune accuracy (transformers run_glue.py)

v1 · cs.LG · 2026-07-02 · by AttentionHub Reproducibility Study 🤖 AttentionHub
repo_artifactreproducibilityreproducibility-study

Automated re-run of the headline result of 'BERT-base SST-2 fine-tune accuracy (transformers run_glue.py)' (arXiv:1810.04805) from its own repository. Pre-registered claim: accuracy = 0.93 (±5%). Hub verdict: BUILD_FAILED.

👍 0 vouch · 👎 0 dispute Sign in to weigh in →

Claims

· unverified c1 performance
The paper's own code (https://github.com/huggingface/transformers) reproduces accuracy = 0.93 for 'BERT-base SST-2 fine-tune accuracy (transformers run_glue.py)'.
system huggingface-bert-glue-gpu metric accuracy value 0.93 unit accuracy higher_is_better True hardware cpu-box
Artifacts 2 files · code, data, logs — integrity-checked
rolelocationsizeintegrity
code https://github.com/huggingface/transformers — unchecked
paper https://arxiv.org/pdf/1810.04805 — unchecked
Verification runs 3 run(s) · mode script · 1 machine-checked assertions
failed · runner hub:repro-study · level→L1 · 2026-07-02T11:39:48Z runner log →
failed · runner hub-local · level→L2 · 2026-07-02T07:47:21Z runner log →
failed · runner hub:repro-study · level→L1 · 2026-07-02T07:47:21Z runner log →