Skip to content

Token counter

TokenCountingHandler #

Bases: PythonicallyPrintingBaseHandler

LLM和Embedding事件中用于计算标记的回调处理程序。

Parameters:

Name Type Description Default
tokenizer Optional[Callable[[str], List]]

要使用的分词器。默认为全局分词器 (参见llama_index.core.utils.globals_helper)。

 None
event_starts_to_ignore Optional[List[CBEventType]]

要在跟踪开始时忽略的事件类型列表。

 None
event_ends_to_ignore Optional[List[CBEventType]]

要在跟踪结束时忽略的事件类型列表。

 None
Source code in llama_index/core/callbacks/token_counting.py 
 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
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
class TokenCountingHandler(PythonicallyPrintingBaseHandler):
    """LLM和Embedding事件中用于计算标记的回调处理程序。

    Args:
        tokenizer:
            要使用的分词器。默认为全局分词器
            (参见llama_index.core.utils.globals_helper)。
        event_starts_to_ignore: 要在跟踪开始时忽略的事件类型列表。
        event_ends_to_ignore: 要在跟踪结束时忽略的事件类型列表。"""

    def __init__(
        self,
        tokenizer: Optional[Callable[[str], List]] = None,
        event_starts_to_ignore: Optional[List[CBEventType]] = None,
        event_ends_to_ignore: Optional[List[CBEventType]] = None,
        verbose: bool = False,
        logger: Optional[logging.Logger] = None,
    ) -> None:
        self.llm_token_counts: List[TokenCountingEvent] = []
        self.embedding_token_counts: List[TokenCountingEvent] = []
        self.tokenizer = tokenizer or get_tokenizer()

        self._token_counter = TokenCounter(tokenizer=self.tokenizer)
        self._verbose = verbose

        super().__init__(
            event_starts_to_ignore=event_starts_to_ignore or [],
            event_ends_to_ignore=event_ends_to_ignore or [],
            logger=logger,
        )

    def start_trace(self, trace_id: Optional[str] = None) -> None:
        return

    def end_trace(
        self,
        trace_id: Optional[str] = None,
        trace_map: Optional[Dict[str, List[str]]] = None,
    ) -> None:
        return

    def on_event_start(
        self,
        event_type: CBEventType,
        payload: Optional[Dict[str, Any]] = None,
        event_id: str = "",
        parent_id: str = "",
        **kwargs: Any,
    ) -> str:
        return event_id

    def on_event_end(
        self,
        event_type: CBEventType,
        payload: Optional[Dict[str, Any]] = None,
        event_id: str = "",
        **kwargs: Any,
    ) -> None:
        """根据需要计算LLM或嵌入式标记。"""
        if (
            event_type == CBEventType.LLM
            and event_type not in self.event_ends_to_ignore
            and payload is not None
        ):
            self.llm_token_counts.append(
                get_llm_token_counts(
                    token_counter=self._token_counter,
                    payload=payload,
                    event_id=event_id,
                )
            )

            if self._verbose:
                self._print(
                    "LLM Prompt Token Usage: "
                    f"{self.llm_token_counts[-1].prompt_token_count}\n"
                    "LLM Completion Token Usage: "
                    f"{self.llm_token_counts[-1].completion_token_count}",
                )
        elif (
            event_type == CBEventType.EMBEDDING
            and event_type not in self.event_ends_to_ignore
            and payload is not None
        ):
            total_chunk_tokens = 0
            for chunk in payload.get(EventPayload.CHUNKS, []):
                self.embedding_token_counts.append(
                    TokenCountingEvent(
                        event_id=event_id,
                        prompt=chunk,
                        prompt_token_count=self._token_counter.get_string_tokens(chunk),
                        completion="",
                        completion_token_count=0,
                    )
                )
                total_chunk_tokens += self.embedding_token_counts[-1].total_token_count

            if self._verbose:
                self._print(f"Embedding Token Usage: {total_chunk_tokens}")

    @property
    def total_llm_token_count(self) -> int:
        """获取当前的LLM代币总量。"""
        return sum([x.total_token_count for x in self.llm_token_counts])

    @property
    def prompt_llm_token_count(self) -> int:
        """获取当前总的LLM提示令牌计数。"""
        return sum([x.prompt_token_count for x in self.llm_token_counts])

    @property
    def completion_llm_token_count(self) -> int:
        """获取当前的LLM完成令牌总数。"""
        return sum([x.completion_token_count for x in self.llm_token_counts])

    @property
    def total_embedding_token_count(self) -> int:
        """获取当前总嵌入标记计数。"""
        return sum([x.total_token_count for x in self.embedding_token_counts])

    def reset_counts(self) -> None:
        """重置令牌计数。"""
        self.llm_token_counts = []
        self.embedding_token_counts = []

total_llm_token_count property #

total_llm_token_count: int

获取当前的LLM代币总量。

prompt_llm_token_count property #

prompt_llm_token_count: int

获取当前总的LLM提示令牌计数。

completion_llm_token_count property #

completion_llm_token_count: int

获取当前的LLM完成令牌总数。

total_embedding_token_count property #

total_embedding_token_count: int

获取当前总嵌入标记计数。

on_event_end #

on_event_end(
    event_type: CBEventType,
    payload: Optional[Dict[str, Any]] = None,
    event_id: str = "",
    **kwargs: Any
) -> None

根据需要计算LLM或嵌入式标记。

Source code in llama_index/core/callbacks/token_counting.py 
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
189
190
191
192
193
194
def on_event_end(
    self,
    event_type: CBEventType,
    payload: Optional[Dict[str, Any]] = None,
    event_id: str = "",
    **kwargs: Any,
) -> None:
    """根据需要计算LLM或嵌入式标记。"""
    if (
        event_type == CBEventType.LLM
        and event_type not in self.event_ends_to_ignore
        and payload is not None
    ):
        self.llm_token_counts.append(
            get_llm_token_counts(
                token_counter=self._token_counter,
                payload=payload,
                event_id=event_id,
            )
        )

        if self._verbose:
            self._print(
                "LLM Prompt Token Usage: "
                f"{self.llm_token_counts[-1].prompt_token_count}\n"
                "LLM Completion Token Usage: "
                f"{self.llm_token_counts[-1].completion_token_count}",
            )
    elif (
        event_type == CBEventType.EMBEDDING
        and event_type not in self.event_ends_to_ignore
        and payload is not None
    ):
        total_chunk_tokens = 0
        for chunk in payload.get(EventPayload.CHUNKS, []):
            self.embedding_token_counts.append(
                TokenCountingEvent(
                    event_id=event_id,
                    prompt=chunk,
                    prompt_token_count=self._token_counter.get_string_tokens(chunk),
                    completion="",
                    completion_token_count=0,
                )
            )
            total_chunk_tokens += self.embedding_token_counts[-1].total_token_count

        if self._verbose:
            self._print(f"Embedding Token Usage: {total_chunk_tokens}")

reset_counts #

reset_counts() -> None

重置令牌计数。

Source code in llama_index/core/callbacks/token_counting.py 
216
217
218
219
def reset_counts(self) -> None:
    """重置令牌计数。"""
    self.llm_token_counts = []
    self.embedding_token_counts = []