aboutsummaryrefslogtreecommitdiff
path: root/tests/check_bert_parity.py
diff options
context:
space:
mode:
authorOnur Gündüz <ogunduz@gmail.com>2026-06-27 00:58:05 +0200
committerOnur Gündüz <ogunduz@gmail.com>2026-06-27 01:03:41 +0200
commitaf55803d0dff9eb692c760c8d486e736c945061d (patch)
tree0875a1112bf98304e445e0544c4309670d6f4769 /tests/check_bert_parity.py
parent0f861b7f3098750ddee4f6570f3be228d1da5320 (diff)
downloadffwd-af55803d0dff9eb692c760c8d486e736c945061d.tar.gz
ffwd-af55803d0dff9eb692c760c8d486e736c945061d.zip
chore: clean up comment formatting and wording
Diffstat (limited to 'tests/check_bert_parity.py')
-rwxr-xr-xtests/check_bert_parity.py5
1 files changed, 2 insertions, 3 deletions
diff --git a/tests/check_bert_parity.py b/tests/check_bert_parity.py
index 2500a5f..93a51f4 100755
--- a/tests/check_bert_parity.py
+++ b/tests/check_bert_parity.py
@@ -1,5 +1,5 @@
#!/usr/bin/env python3
-"""Compare the C BERT-family forward against SentenceTransformers.
+"""Compare the BERT-family forward against SentenceTransformers.
Builds a small C driver that loads the WordPiece tokenizer and the BERT model
from a sentence-transformers checkpoint, wraps each text with [CLS]/[SEP],
@@ -7,8 +7,7 @@ encodes it (BERT forward + mean pool + L2 normalize), and prints the vector.
The reference is SentenceTransformer(model).encode(normalize_embeddings=True),
which runs the same pipeline. Defaults to all-MiniLM-L6-v2.
-This one needs the real model weights to run the C forward, so it stays a
-manual --model-dir check rather than a hermetic stored-vector test.
+Needs the real model weights to run the C forward.
uv run --python 3.12 --with-requirements requirements-reference.txt \
tests/check_bert_parity.py --model-dir DIR