binary-tokenizer-001-16k

A cross-platform BPE tokenizer for binary executables and machine code. Trained on 13 GB of diverse binaries spanning Linux, Windows, macOS, and Android platforms.

🔗 Model: mjbommar/binary-tokenizer-001-16k 📊 Dataset: mjbommar/binary-30k-tokenized 📄 Paper: Binary BPE: Cross-Platform Tokenization for Binary Analysis (arXiv preprint coming soon)

Overview

• Vocabulary Size: 16,384 tokens (2^14)
• Token Composition: 256 base bytes + 16,121 learned merges + 7 special tokens
• Average Token Length: 3.498 bytes
• 3-byte Instructions: 20.5% of vocabulary (3,360 tokens)
• Compression Ratio: ~2.4 bytes/token on typical binaries

---

Training Configuration

Training Corpus:

• Source: mjbommar/binary-30k-tokenized
• Size: ~13 GB
• Files: 30,738 binary files
• Platforms: Linux (ELF), Windows (PE), macOS (Mach-O), Android (APK)
• Architectures: x86-64, x86, ARM64, ARM, MIPS, RISC-V

Training Parameters:

• Vocabulary size: 16,384 (including 7 special tokens)
• Min frequency: 10
• Chunk size: 8,192 bytes
• Allowed lengths: DEFAULT (1-16 bytes)
• Training duration: ~3-4 hours

---

Vocabulary Statistics

Composition:

• Base bytes (0-255): 256 tokens
• Learned merges: 16,121 tokens
• Special tokens: 7 tokens (<|start|>, <|end|>, <|pad|>, <|unk|>, <|cls|>, <|sep|>, <|mask|>)
• Total: 16,384 tokens

Quality Metrics:

• All tokens reachable: ✓ Yes
• Valid merges: 16,121 / 16,121
• Power-of-2 size: ✓ Yes (2^14)

---

Token Length Distribution

Length	Count	Percentage	Description
1 byte	256	1.6%	Base bytes
2 bytes	7,149	43.7%	Byte pairs
3 bytes	3,360	20.5%	Complete x86-64 instructions
4 bytes	3,082	18.8%	Instructions with operands
5 bytes	719	4.4%	Complex patterns
6 bytes	606	3.7%	Complex patterns
7 bytes	228	1.4%	Complex patterns
8 bytes	377	2.3%	Complex patterns
9+ bytes	607	3.7%	Long patterns

Average Token Length: 3.498 bytes

---

Byte Content Analysis

Content Categories:

• Contains NULL byte (0x00): 4,128 tokens (25.2%)
• ASCII printable (0x20-0x7E): 3,513 tokens (21.5%)
• All ASCII (<0x80): 7,256 tokens (44.3%)
• High bytes (≥0x80): 9,121 tokens (55.7%)

Most Common Bytes in Tokens:

• 0x00 (NULL): 9,741 occurrences - Padding and alignment
• 0xFF: 1,718 occurrences - Sentinel values
• 0x48 (REX.W): 1,352 occurrences - x86-64 REX prefix
• 0x8B (MOV): 955 occurrences - x86-64 MOV opcode
• 0xCC (INT3): 751 occurrences - Debug breakpoint padding

---

Sequence Coverage

N-byte Sequence Diversity:

Length	Learned Tokens	Possible Sequences	Coverage
1-byte	256	256	100.00%
2-byte	7,149	65,536	10.91%
3-byte	3,360	16,777,216	0.020%
4-byte	3,082	4,294,967,296	0.000072%

---

Files

• tokenizer-16384.json - Trained tokenizer model (1.2 MB)
• analysis_results.json - Detailed analysis statistics
• training.log - Training output log (if available)
• training_stats.txt - Training summary (if available)

---

Usage

Load from HuggingFace Hub:

from tokenizers import Tokenizer

# Load directly from HuggingFace
tokenizer = Tokenizer.from_pretrained("mjbommar/binary-tokenizer-001-16k")

Load from local file:

# With bbpe CLI
bbpe encode --tokenizer tokenizer-16384.json /path/to/binary
bbpe info tokenizer-16384.json

Complete Python Example:

from tokenizers import Tokenizer

# Load from HuggingFace or local file
tokenizer = Tokenizer.from_pretrained("mjbommar/binary-tokenizer-001-16k")
# OR: tokenizer = Tokenizer.from_file("tokenizer-16384.json")

# Read binary file and decode as latin-1 (preserves all byte values 0-255)
with open("/usr/bin/ls", "rb") as f:
    data = f.read()
    data_str = data.decode("latin-1")

# Encode the binary data
encoding = tokenizer.encode(data_str)
print(f"File size: {len(data)} bytes")
print(f"Total tokens: {len(encoding.ids)}")
print(f"Compression: {len(data) / len(encoding.ids):.3f} bytes/token")

# First 10 tokens
for i, (token_id, token) in enumerate(zip(encoding.ids[:10], encoding.tokens[:10])):
    token_bytes = token.encode("latin-1")
    print(f"  Token {i}: ID={token_id:5d} hex={token_bytes.hex():20s} ({len(token_bytes)} bytes)")

# Decode tokens back to bytes
decoded_str = tokenizer.decode(encoding.ids)
decoded_bytes = decoded_str.encode("latin-1")
assert decoded_bytes == data  # Perfect reconstruction

Example output for /usr/bin/ls (142,312 bytes):

File size: 142312 bytes
Total tokens: 59531
Compression: 2.391 bytes/token

First 10 tokens:
  Token 0: ID=  127 hex=7f                   (1 bytes)
  Token 1: ID=15580 hex=454c                 (2 bytes)
  Token 2: ID=   70 hex=46                   (1 bytes)
  Token 3: ID=    2 hex=02                   (1 bytes)
  Token 4: ID= 1516 hex=0101                 (2 bytes)
  Token 5: ID= 2624 hex=000000000000000000   (9 bytes)
  Token 6: ID= 1046 hex=0300                 (2 bytes)
  Token 7: ID= 5675 hex=3e00                 (2 bytes)
  Token 8: ID= 1099 hex=01000000             (4 bytes)
  Token 9: ID=   48 hex=30                   (1 bytes)

Decoded: 7f454c4602010100000000000000000003003e000100000030...
(ELF header: 7f 45 4c 46 = ELF magic bytes)

---

Citation

If you use this tokenizer in your research, please cite:

@article{bommarito2025binarybpe,
  title={Binary BPE: Cross-Platform Tokenization for Binary Analysis},
  author={Bommarito II, Michael J.},
  journal={arXiv preprint},
  year={2025},
  note={Preprint coming soon}
}

Author: Michael J. Bommarito II ([email protected])

---

Generated: November 12, 2025 Training Script: train_tokenizers.sh Analysis Script: analyze_tokenizer.py