books/LLMs-from-scratch
1
0
Fork 0
mirror of https://github.com/rasbt/LLMs-from-scratch.git synced 2026-04-10 12:33:42 +00:00
Code Issues Packages Projects Releases Wiki Activity

Fix encoding of multiple preceding spaces in BPE tokenizer. (#945)

Browse Source 
* Fix encoding of multiple preceding spaces in BPE tokenizer.

* Add test

---------

Co-authored-by: rasbt <mail@sebastianraschka.com>
This commit is contained in:
Maxwell De Jong
2026-01-10 11:27:23 -05:00
committed by GitHub
parent 90e0f3cc15
commit e0dbec3331
2 changed files with 13 additions and 2 deletions
Download Patch File Download Diff File Expand all files Collapse all files

2
ch02/05_bpe-from-scratch/bpe-from-scratch.ipynb
View File

@@ -609,9 +609,9 @@
" else:\n", " else:\n",
" word = m.group(2)\n", " word = m.group(2)\n",
" if pending_spaces > 0:\n", " if pending_spaces > 0:\n",
" tokens.append(\"Ġ\" + word) # one leading space\n",
" for _ in range(pending_spaces - 1):\n", " for _ in range(pending_spaces - 1):\n",
" tokens.append(\"Ġ\") # remaining spaces as standalone\n", " tokens.append(\"Ġ\") # remaining spaces as standalone\n",
" tokens.append(\"Ġ\" + word) # one leading space\n",
" pending_spaces = 0\n", " pending_spaces = 0\n",
" else:\n", " else:\n",
" tokens.append(word)\n", " tokens.append(word)\n",

11
ch02/05_bpe-from-scratch/tests.py
View File

@@ -239,3 +239,14 @@ def test_space_newline_space_patterns(imported_module, gpt2_files):
] ]
for s in samples: for s in samples:
assert tok.encode(s) == tik.encode(s), f"Mismatch vs tiktoken: {repr(s)}" assert tok.encode(s) == tik.encode(s), f"Mismatch vs tiktoken: {repr(s)}"
def test_multiple_leading_spaces_roundtrip(imported_module, gpt2_files):
BPETokenizerSimple = getattr(imported_module, "BPETokenizerSimple", None)
tok = BPETokenizerSimple()
tok.load_vocab_and_merges_from_openai(
vocab_path=gpt2_files["encoder.json"], bpe_merges_path=gpt2_files["vocab.bpe"]
)
text = " Hello World."
assert tok.decode(tok.encode(text)) == text