File size: 8,557 Bytes
b4b665d
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
import abc
import os
from dataclasses import dataclass, field
from typing import Hashable, List, Optional, TypeVar
from .consts import (COMPLETE_ALPHABET, WHITESPACE_CHARACTERS, DEFAULT_MAX_CONSECUTIVE_WHITESPACES, 
                     DEFAULT_FORCE_JSON_FIELD_ORDER, CONFIG_ENV_VAR_MAX_CONSECUTIVE_WHITESPACES, 
                     CONFIG_ENV_VAR_STRICT_JSON_FIELD_ORDER, CONFIG_ENV_VAR_MAX_JSON_ARRAY_LENGTH,
                     DEFAULT_MAX_JSON_ARRAY_LENGTH)


def _parse_bool(s: str) -> bool:
    return s and (s.strip().lower() in ['true', '1'])


def _env_or_default_field(env_var: str, default_val):
    default_val_type = type(default_val)
    parser_func = _parse_bool if default_val_type == bool else default_val_type
    def factory_func():
        return parser_func(os.environ.get(env_var, str(default_val)))
    return field(default_factory=factory_func)


@dataclass
class CharacterLevelParserConfig:
    alphabet: str = COMPLETE_ALPHABET
    max_consecutive_whitespaces: int = _env_or_default_field(CONFIG_ENV_VAR_MAX_CONSECUTIVE_WHITESPACES, 
                                                             DEFAULT_MAX_CONSECUTIVE_WHITESPACES)
    """How many consective whitespaces the JsonSchemaParser will allow"""
    force_json_field_order: bool = _env_or_default_field(CONFIG_ENV_VAR_STRICT_JSON_FIELD_ORDER, 
                                                         DEFAULT_FORCE_JSON_FIELD_ORDER)
    """Whether the JsonSchemaParser will force fields to appear in the 
    order of the 'required' field in the schema"""
    max_json_array_length: int = _env_or_default_field(CONFIG_ENV_VAR_MAX_JSON_ARRAY_LENGTH,
                                                       DEFAULT_MAX_JSON_ARRAY_LENGTH)
    """What is the maximum json array length if not specified by the schema. Helps the LLM
    avoid infinite loops."""


class CharacterLevelParser(abc.ABC):
    """CharacterLevelParser is an interface for classes that can parse strings one character at a time, and determine which characters are allowed at any specific time"""

    def __init__(self, config: Optional[CharacterLevelParserConfig] = None):
        self._config = config or CharacterLevelParserConfig()
    
    @abc.abstractmethod
    def add_character(self, new_character: str) -> 'CharacterLevelParser':
        """Add a character to the parser, and return a new parser that represents the state of the parser after the character has been added. This has to be
        an immutable operation - the original CharacterLevelParser (self) must not be modified."""
        raise NotImplementedError()

    @abc.abstractmethod
    def get_allowed_characters(self) -> str:
        """Return a string containing all characters that are allowed at the current point in the parsing process."""
        raise NotImplementedError()
    
    @abc.abstractmethod
    def can_end(self) -> bool:
        """Return True if the parser is in a state where it can end (potentially finished parsing the desired structure), and False otherwise."""
        raise NotImplementedError()
    
    def shortcut_key(self) -> Optional[Hashable]:
        """Optional. Return a key that denotes that this state is a repeating state, full tree traversal should be avoided."""
        return None
    
    def cache_key(self) -> Optional[Hashable]:
        """Optional. Return a key that denotes that this state is a repeating state, and if it is visited again, results can be cached."""
        return None
    
    @property
    def config(self) -> CharacterLevelParserConfig:
        return self._config
    
    @config.setter
    def config(self, new_config: CharacterLevelParserConfig):
        self._config = new_config
        return self


class StringParser(CharacterLevelParser):
    """RegexParser is an example CharacterLevelParser that only allows an exact string. It is a debugging / learning tool
    to show how CharacterLevelParser works together with TokenizerPrefixTree to filter the allowed tokens (some of whom may contain multiple characters)"""
    def __init__(self, string: str):
        self.target_str = string

    def add_character(self, new_character: str) -> CharacterLevelParser:
        if self.target_str.startswith(new_character):
            return StringParser(self.target_str[len(new_character):])
        else:
            raise ValueError(f"Expected '{self.target_str[0]}' but got '{new_character}'")

    def get_allowed_characters(self) -> str:
        return self.target_str[0] if self.target_str else ""

    def can_end(self) -> bool:
        return not self.target_str
    

class ForceStopParser(CharacterLevelParser):
    """A simple parser that forbids any characters except the stop token. Used to force stop LM operation"""
    def __init__(self, allow_whitespace: bool = False):
        self.allow_whitespace = allow_whitespace
    def add_character(self, new_character: str) -> CharacterLevelParser:
        return self
    def get_allowed_characters(self) -> str:
        return WHITESPACE_CHARACTERS if self.allow_whitespace else ""
    def can_end(self) -> bool:
        return True
    

class UnionParser(CharacterLevelParser):
    """A parser that allows a string that would be allowed by any of several different parsers"""
    def __init__(self, parsers: List[CharacterLevelParser]):
        self.parsers = parsers

    def add_character(self, new_character: str) -> CharacterLevelParser:
        # This is a bit of a performance hit, as it means get_allowed_characters() is called twice.
        relevant_parsers = [parser for parser in self.parsers if new_character in parser.get_allowed_characters()]
        next_parsers = [parser.add_character(new_character) for parser in relevant_parsers]
        if len(next_parsers) == 1:
            return next_parsers[0]
        return UnionParser(next_parsers)
    
    def get_allowed_characters(self) -> str:
        allowed = "".join([parser.get_allowed_characters() for parser in self.parsers])
        return "".join(set(allowed))
    
    def can_end(self) -> bool:
        return any([parser.can_end() for parser in self.parsers])
    
    def shortcut_key(self) -> Optional[Hashable]:
        unique_shortcut_keys = set(parser.shortcut_key() for parser in self.parsers)
        if len(unique_shortcut_keys) == 1:
            return next(iter(unique_shortcut_keys))
        return None
    
    def cache_key(self) -> Optional[Hashable]:
        all_cache_keys = tuple(parser.cache_key() for parser in self.parsers)
        if all(key is not None for key in all_cache_keys):
            return ('union', all_cache_keys)
        return None


class SequenceParser(CharacterLevelParser):
    """A parser that is a sequence of multiple parsers."""
    def __init__(self, parsers: List[CharacterLevelParser]):
        self.parsers = parsers

    def add_character(self, new_character: str) -> CharacterLevelParser:
        legal_parsers = []
        # Tricky edge case: if the first parser can both end and accept the character,
        # and the second parser can also accept, we don't know which scenario we are dealing
        # with, so we need to return a UnionParser.
        for idx, parser in enumerate(self.parsers):
            if new_character in parser.get_allowed_characters():
                updated_parser = parser.add_character(new_character)
                next_parsers = [updated_parser] + self.parsers[idx+1:]
                if len(next_parsers) == 1:
                    legal_parsers.append(next_parsers[0])
                else:
                    legal_parsers.append(SequenceParser(next_parsers))
            if not parser.can_end():
                break
        if len(legal_parsers) == 1:
            return legal_parsers[0]
        return UnionParser(legal_parsers)
    
    def get_allowed_characters(self) -> str:
        allowed_characters = set()
        for parser in self.parsers:
            allowed_characters.update(parser.get_allowed_characters())
            if not parser.can_end():
                break
        return "".join(allowed_characters)
    
    def can_end(self) -> bool:
        return all([parser.can_end() for parser in self.parsers])
    
    def shortcut_key(self) -> Optional[str]:
        return self.parsers[0].shortcut_key() if len(self.parsers) == 1 else None
    
    def cache_key(self) -> Optional[Hashable]:
        all_cache_keys = tuple(parser.cache_key() for parser in self.parsers)
        if all(key is not None for key in all_cache_keys):
            return ('sequence', all_cache_keys)
        return None