MCPcopy Create free account
hub / github.com/DeusData/codebase-memory-mcp / is_code_relevant

Function is_code_relevant

scripts/extract_nomic_vectors.py:58–94  ·  view source on GitHub ↗

Filter vocabulary to code-relevant tokens. Goal: keep tokens that our runtime camelCase/snake_case splitter would produce from identifiers. Reject BPE noise, punctuation combos, and non-Latin scripts.

(token_str: str)

Source from the content-addressed store, hash-verified

56# ── Token filtering ───────────────────────────────────────────────────
57
58def is_code_relevant(token_str: str) -> bool:
59 """Filter vocabulary to code-relevant tokens.
60
61 Goal: keep tokens that our runtime camelCase/snake_case splitter would
62 produce from identifiers. Reject BPE noise, punctuation combos, and
63 non-Latin scripts.
64 """
65 s = token_str.strip()
66 if not s:
67 return False
68
69 # Remove BPE markers (Ġ = space prefix, ▁ = sentencepiece, Ċ/ċ = newline in Qwen)
70 clean = s.lstrip("\u0120\u2581") # Ġ, ▁
71 if not clean:
72 return False
73
74 # Skip special tokens
75 if clean.startswith("<") and clean.endswith(">"):
76 return False
77 if clean.startswith("[") and clean.endswith("]"):
78 return False
79
80 # Strip leading/trailing underscores (common in BPE) but keep content
81 inner = clean.strip("_")
82 if not inner:
83 return False
84
85 # STRICT: must be purely alphanumeric + underscores (identifier-shaped)
86 # This rejects BPE noise like "!");ċ", "!!!!ċċ", etc.
87 if not re.match(r'^[a-zA-Z][a-zA-Z0-9_]*$', inner):
88 return False
89
90 # Must be at least 2 chars of actual content
91 if len(inner) < 2:
92 return False
93
94 return True
95
96
97def clean_token(token_str: str) -> str:

Callers 1

mainFunction · 0.85

Calls

no outgoing calls

Tested by

no test coverage detected