Skip to content

Wandb

WandbCallbackHandler #

Bases: BaseCallbackHandler

回调处理程序,用于将事件记录到wandb。

注意:这是一个测试版功能。我们代码库中的使用方式和接口可能会发生变化。

使用WandbCallbackHandler来将跟踪事件记录到wandb。这个处理程序对于调试和可视化跟踪事件非常有用。它捕获事件的有效负载并将其记录到wandb。该处理程序还跟踪事件的开始和结束。这对于调试LLM调用特别有用。

WandbCallbackHandler还可以使用persist_index方法将索引和图记录到wandb。这将把索引保存为wandb中的artifact。load_storage_context方法可用于从wandb中加载索引artifact。该方法将返回一个StorageContext对象,可用于构建索引,使用load_index_from_storageload_indices_from_storageload_graph_from_storage函数。

Source code in llama_index/callbacks/wandb/base.py
 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
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
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
class WandbCallbackHandler(BaseCallbackHandler):
    """回调处理程序,用于将事件记录到wandb。

    注意:这是一个测试版功能。我们代码库中的使用方式和接口可能会发生变化。

    使用`WandbCallbackHandler`来将跟踪事件记录到wandb。这个处理程序对于调试和可视化跟踪事件非常有用。它捕获事件的有效负载并将其记录到wandb。该处理程序还跟踪事件的开始和结束。这对于调试LLM调用特别有用。

    `WandbCallbackHandler`还可以使用`persist_index`方法将索引和图记录到wandb。这将把索引保存为wandb中的artifact。`load_storage_context`方法可用于从wandb中加载索引artifact。该方法将返回一个`StorageContext`对象,可用于构建索引,使用`load_index_from_storage`、`load_indices_from_storage`或`load_graph_from_storage`函数。

    Args:
        event_starts_to_ignore(Optional[List[CBEventType]]):要忽略跟踪事件开始时的事件类型列表。
        event_ends_to_ignore(Optional[List[CBEventType]]):要忽略跟踪事件结束时的事件类型列表。"""

    def __init__(
        self,
        run_args: Optional[WandbRunArgs] = None,
        tokenizer: Optional[Callable[[str], List]] = None,
        event_starts_to_ignore: Optional[List[CBEventType]] = None,
        event_ends_to_ignore: Optional[List[CBEventType]] = None,
    ) -> None:
        try:
            import wandb
            from wandb.sdk.data_types import trace_tree

            self._wandb = wandb
            self._trace_tree = trace_tree
        except ImportError:
            raise ImportError(
                "WandbCallbackHandler requires wandb. "
                "Please install it with `pip install wandb`."
            )

        from llama_index.core.indices import (
            ComposableGraph,
            GPTEmptyIndex,
            GPTKeywordTableIndex,
            GPTRAKEKeywordTableIndex,
            GPTSimpleKeywordTableIndex,
            GPTSQLStructStoreIndex,
            GPTTreeIndex,
            GPTVectorStoreIndex,
            SummaryIndex,
        )

        self._IndexType = (
            ComposableGraph,
            GPTKeywordTableIndex,
            GPTSimpleKeywordTableIndex,
            GPTRAKEKeywordTableIndex,
            SummaryIndex,
            GPTEmptyIndex,
            GPTTreeIndex,
            GPTVectorStoreIndex,
            GPTSQLStructStoreIndex,
        )

        self._run_args = run_args
        # Check if a W&B run is already initialized; if not, initialize one
        self._ensure_run(should_print_url=(self._wandb.run is None))  # type: ignore[attr-defined]

        self._event_pairs_by_id: Dict[str, List[CBEvent]] = defaultdict(list)
        self._cur_trace_id: Optional[str] = None
        self._trace_map: Dict[str, List[str]] = defaultdict(list)

        self.tokenizer = tokenizer or get_tokenizer()
        self._token_counter = TokenCounter(tokenizer=self.tokenizer)

        event_starts_to_ignore = (
            event_starts_to_ignore if event_starts_to_ignore else []
        )
        event_ends_to_ignore = event_ends_to_ignore if event_ends_to_ignore else []
        super().__init__(
            event_starts_to_ignore=event_starts_to_ignore,
            event_ends_to_ignore=event_ends_to_ignore,
        )

    def on_event_start(
        self,
        event_type: CBEventType,
        payload: Optional[Dict[str, Any]] = None,
        event_id: str = "",
        parent_id: str = "",
        **kwargs: Any,
    ) -> str:
        """按事件类型存储事件开始数据。

Args:
    event_type (CBEventType): 要存储的事件类型。
    payload (Optional[Dict[str, Any]]): 要存储的有效负载。
    event_id (str): 要存储的事件ID。
    parent_id (str): 父事件ID。
"""
        event = CBEvent(event_type, payload=payload, id_=event_id)
        self._event_pairs_by_id[event.id_].append(event)
        return event.id_

    def on_event_end(
        self,
        event_type: CBEventType,
        payload: Optional[Dict[str, Any]] = None,
        event_id: str = "",
        **kwargs: Any,
    ) -> None:
        """按事件类型存储事件结束数据。

Args:
    event_type (CBEventType): 要存储的事件类型。
    payload (Optional[Dict[str, Any]]): 要存储的有效负载。
    event_id (str): 要存储的事件ID。
"""
        event = CBEvent(event_type, payload=payload, id_=event_id)
        self._event_pairs_by_id[event.id_].append(event)
        self._trace_map = defaultdict(list)

    def start_trace(self, trace_id: Optional[str] = None) -> None:
        """启动一个跟踪。"""
        self._trace_map = defaultdict(list)
        self._cur_trace_id = trace_id
        self._start_time = datetime.now()

    def end_trace(
        self,
        trace_id: Optional[str] = None,
        trace_map: Optional[Dict[str, List[str]]] = None,
    ) -> None:
        # Ensure W&B run is initialized
        self._ensure_run()

        self._trace_map = trace_map or defaultdict(list)
        self._end_time = datetime.now()

        # Log the trace map to wandb
        # We can control what trace ids we want to log here.
        self.log_trace_tree()

        # TODO (ayulockin): Log the LLM token counts to wandb when weave is ready

    def log_trace_tree(self) -> None:
        """将跟踪树记录到wandb。"""
        try:
            child_nodes = self._trace_map["root"]
            root_span = self._convert_event_pair_to_wb_span(
                self._event_pairs_by_id[child_nodes[0]],
                trace_id=self._cur_trace_id if len(child_nodes) > 1 else None,
            )

            if len(child_nodes) == 1:
                child_nodes = self._trace_map[child_nodes[0]]
                root_span = self._build_trace_tree(child_nodes, root_span)
            else:
                root_span = self._build_trace_tree(child_nodes, root_span)
            if root_span:
                root_trace = self._trace_tree.WBTraceTree(root_span)
                if self._wandb.run:  # type: ignore[attr-defined]
                    self._wandb.run.log({"trace": root_trace})  # type: ignore[attr-defined]
                self._wandb.termlog("Logged trace tree to W&B.")  # type: ignore[attr-defined]
        except Exception as e:
            print(f"Failed to log trace tree to W&B: {e}")
            # ignore errors to not break user code

    def persist_index(
        self, index: "IndexType", index_name: str, persist_dir: Union[str, None] = None
    ) -> None:
        """将索引上传到wandb作为一个artifact。您可以在这里了解更多关于W&B artifacts的信息:https://docs.wandb.ai/guides/artifacts。

对于`ComposableGraph`索引,根id被存储为artifact元数据。

Args:
    index (IndexType): 要上传的索引。
    index_name (str): 索引的名称。这将被用作artifact的名称。
    persist_dir (Union[str, None]): 持久化索引的目录。如果为None,将创建并使用临时目录。
"""
        if persist_dir is None:
            persist_dir = f"{self._wandb.run.dir}/storage"  # type: ignore
            _default_persist_dir = True
        if not os.path.exists(persist_dir):
            os.makedirs(persist_dir)

        if isinstance(index, self._IndexType):
            try:
                index.storage_context.persist(persist_dir)  # type: ignore

                metadata = None
                # For the `ComposableGraph` index, store the root id as metadata
                if isinstance(index, self._IndexType[0]):
                    root_id = index.root_id
                    metadata = {"root_id": root_id}

                self._upload_index_as_wb_artifact(persist_dir, index_name, metadata)
            except Exception as e:
                # Silently ignore errors to not break user code
                self._print_upload_index_fail_message(e)

        # clear the default storage dir
        if _default_persist_dir:
            shutil.rmtree(persist_dir, ignore_errors=True)

    def load_storage_context(
        self, artifact_url: str, index_download_dir: Union[str, None] = None
    ) -> "StorageContext":
        """从wandb下载索引并返回存储上下文。

使用此存储上下文,可以使用`load_index_from_storage`、`load_indices_from_storage`或`load_graph_from_storage`函数将索引加载到内存中。

Args:
    artifact_url(str):要下载的artifact的url。artifact的url的格式为:`entity/project/index_name:version`,可以在W&B UI中找到。
    index_download_dir(Union[str, None]):要下载索引的目录。
"""
        from llama_index.core.storage.storage_context import StorageContext

        artifact = self._wandb.use_artifact(artifact_url, type="storage_context")  # type: ignore[attr-defined]
        artifact_dir = artifact.download(root=index_download_dir)

        return StorageContext.from_defaults(persist_dir=artifact_dir)

    def _upload_index_as_wb_artifact(
        self, dir_path: str, artifact_name: str, metadata: Optional[Dict]
    ) -> None:
        """实用函数,将一个目录上传到W&B作为一个artifact。"""
        artifact = self._wandb.Artifact(artifact_name, type="storage_context")  # type: ignore[attr-defined]

        if metadata:
            artifact.metadata = metadata

        artifact.add_dir(dir_path)
        self._wandb.run.log_artifact(artifact)  # type: ignore

    def _build_trace_tree(
        self, events: List[str], span: "trace_tree.Span"
    ) -> "trace_tree.Span":
        """从跟踪地图构建跟踪树。"""
        for child_event in events:
            child_span = self._convert_event_pair_to_wb_span(
                self._event_pairs_by_id[child_event]
            )
            child_span = self._build_trace_tree(
                self._trace_map[child_event], child_span
            )
            span.add_child_span(child_span)

        return span

    def _convert_event_pair_to_wb_span(
        self,
        event_pair: List[CBEvent],
        trace_id: Optional[str] = None,
    ) -> "trace_tree.Span":
        """将一对事件转换为一个wandb跟踪树跨度。"""
        start_time_ms, end_time_ms = self._get_time_in_ms(event_pair)

        if trace_id is None:
            event_type = event_pair[0].event_type
            span_kind = self._map_event_type_to_span_kind(event_type)
        else:
            event_type = trace_id  # type: ignore
            span_kind = None

        wb_span = self._trace_tree.Span(
            name=f"{event_type}",
            span_kind=span_kind,
            start_time_ms=start_time_ms,
            end_time_ms=end_time_ms,
        )

        inputs, outputs, wb_span = self._add_payload_to_span(wb_span, event_pair)
        wb_span.add_named_result(inputs=inputs, outputs=outputs)  # type: ignore

        return wb_span

    def _map_event_type_to_span_kind(
        self, event_type: CBEventType
    ) -> Union[None, "trace_tree.SpanKind"]:
        """将CBEventType映射到wandb跟踪树SpanKind。"""
        if event_type == CBEventType.CHUNKING:
            span_kind = None
        elif event_type == CBEventType.NODE_PARSING:
            span_kind = None
        elif event_type == CBEventType.EMBEDDING:
            # TODO: add span kind for EMBEDDING when it's available
            span_kind = None
        elif event_type == CBEventType.LLM:
            span_kind = self._trace_tree.SpanKind.LLM
        elif event_type == CBEventType.QUERY:
            span_kind = self._trace_tree.SpanKind.AGENT
        elif event_type == CBEventType.AGENT_STEP:
            span_kind = self._trace_tree.SpanKind.AGENT
        elif event_type == CBEventType.RETRIEVE:
            span_kind = self._trace_tree.SpanKind.TOOL
        elif event_type == CBEventType.SYNTHESIZE:
            span_kind = self._trace_tree.SpanKind.CHAIN
        elif event_type == CBEventType.TREE:
            span_kind = self._trace_tree.SpanKind.CHAIN
        elif event_type == CBEventType.SUB_QUESTION:
            span_kind = self._trace_tree.SpanKind.CHAIN
        elif event_type == CBEventType.RERANKING:
            span_kind = self._trace_tree.SpanKind.CHAIN
        elif event_type == CBEventType.FUNCTION_CALL:
            span_kind = self._trace_tree.SpanKind.TOOL
        else:
            span_kind = None

        return span_kind

    def _add_payload_to_span(
        self, span: "trace_tree.Span", event_pair: List[CBEvent]
    ) -> Tuple[Optional[Dict[str, Any]], Optional[Dict[str, Any]], "trace_tree.Span"]:
        """将事件的有效负载添加到跨度中。"""
        assert len(event_pair) == 2
        event_type = event_pair[0].event_type
        inputs = None
        outputs = None

        if event_type == CBEventType.NODE_PARSING:
            # TODO: disabled full detailed inputs/outputs due to UI lag
            inputs, outputs = self._handle_node_parsing_payload(event_pair)
        elif event_type == CBEventType.LLM:
            inputs, outputs, span = self._handle_llm_payload(event_pair, span)
        elif event_type == CBEventType.QUERY:
            inputs, outputs = self._handle_query_payload(event_pair)
        elif event_type == CBEventType.EMBEDDING:
            inputs, outputs = self._handle_embedding_payload(event_pair)

        return inputs, outputs, span

    def _handle_node_parsing_payload(
        self, event_pair: List[CBEvent]
    ) -> Tuple[Dict[str, Any], Dict[str, Any]]:
        """处理NODE_PARSING事件的有效负载。"""
        inputs = event_pair[0].payload
        outputs = event_pair[-1].payload

        if inputs and EventPayload.DOCUMENTS in inputs:
            documents = inputs.pop(EventPayload.DOCUMENTS)
            inputs["num_documents"] = len(documents)

        if outputs and EventPayload.NODES in outputs:
            nodes = outputs.pop(EventPayload.NODES)
            outputs["num_nodes"] = len(nodes)

        return inputs or {}, outputs or {}

    def _handle_llm_payload(
        self, event_pair: List[CBEvent], span: "trace_tree.Span"
    ) -> Tuple[Dict[str, Any], Dict[str, Any], "trace_tree.Span"]:
        """处理LLM事件的有效负载。"""
        inputs = event_pair[0].payload
        outputs = event_pair[-1].payload

        assert isinstance(inputs, dict) and isinstance(outputs, dict)

        # Get `original_template` from Prompt
        if EventPayload.PROMPT in inputs:
            inputs[EventPayload.PROMPT] = inputs[EventPayload.PROMPT]

        # Format messages
        if EventPayload.MESSAGES in inputs:
            inputs[EventPayload.MESSAGES] = "\n".join(
                [str(x) for x in inputs[EventPayload.MESSAGES]]
            )

        token_counts = get_llm_token_counts(self._token_counter, outputs)
        metadata = {
            "formatted_prompt_tokens_count": token_counts.prompt_token_count,
            "prediction_tokens_count": token_counts.completion_token_count,
            "total_tokens_used": token_counts.total_token_count,
        }
        span.attributes = metadata

        # Make `response` part of `outputs`
        outputs = {EventPayload.RESPONSE: str(outputs[EventPayload.RESPONSE])}

        return inputs, outputs, span

    def _handle_query_payload(
        self, event_pair: List[CBEvent]
    ) -> Tuple[Optional[Dict[str, Any]], Dict[str, Any]]:
        """处理QUERY事件的有效负载。"""
        inputs = event_pair[0].payload
        outputs = event_pair[-1].payload

        if outputs:
            response_obj = outputs[EventPayload.RESPONSE]
            response = str(outputs[EventPayload.RESPONSE])

            if type(response).__name__ == "Response":
                response = response_obj.response
            elif type(response).__name__ == "StreamingResponse":
                response = response_obj.get_response().response
        else:
            response = " "

        outputs = {"response": response}

        return inputs, outputs

    def _handle_embedding_payload(
        self,
        event_pair: List[CBEvent],
    ) -> Tuple[Optional[Dict[str, Any]], Dict[str, Any]]:
        event_pair[0].payload
        outputs = event_pair[-1].payload

        chunks = []
        if outputs:
            chunks = outputs.get(EventPayload.CHUNKS, [])

        return {}, {"num_chunks": len(chunks)}

    def _get_time_in_ms(self, event_pair: List[CBEvent]) -> Tuple[int, int]:
        """获取事件对的开始和结束时间,单位为毫秒。"""
        start_time = datetime.strptime(event_pair[0].time, TIMESTAMP_FORMAT)
        end_time = datetime.strptime(event_pair[1].time, TIMESTAMP_FORMAT)

        start_time_in_ms = int(
            (start_time - datetime(1970, 1, 1)).total_seconds() * 1000
        )
        end_time_in_ms = int((end_time - datetime(1970, 1, 1)).total_seconds() * 1000)

        return start_time_in_ms, end_time_in_ms

    def _ensure_run(self, should_print_url: bool = False) -> None:
        """确保存在一个活跃的W&B运行。

如果不存在,则将使用提供的run_args启动一个新的运行。
"""
        if self._wandb.run is None:  # type: ignore[attr-defined]
            # Make a shallow copy of the run args, so we don't modify the original
            run_args = self._run_args or {}  # type: ignore
            run_args: dict = {**run_args}  # type: ignore

            # Prefer to run in silent mode since W&B has a lot of output
            # which can be undesirable when dealing with text-based models.
            if "settings" not in run_args:  # type: ignore
                run_args["settings"] = {"silent": True}  # type: ignore

            # Start the run and add the stream table
            self._wandb.init(**run_args)  # type: ignore[attr-defined]
            self._wandb.run._label(repo="llama_index")  # type: ignore

            if should_print_url:
                self._print_wandb_init_message(
                    self._wandb.run.settings.run_url  # type: ignore
                )

    def _print_wandb_init_message(self, run_url: str) -> None:
        """当W&B初始化时,在终端打印一条消息。"""
        self._wandb.termlog(  # type: ignore[attr-defined]
            f"Streaming LlamaIndex events to W&B at {run_url}\n"
            "`WandbCallbackHandler` is currently in beta.\n"
            "Please report any issues to https://github.com/wandb/wandb/issues "
            "with the tag `llamaindex`."
        )

    def _print_upload_index_fail_message(self, e: Exception) -> None:
        """当上传索引失败时,在终端打印一条消息。"""
        self._wandb.termlog(  # type: ignore[attr-defined]
            f"Failed to upload index to W&B with the following error: {e}\n"
        )

    def finish(self) -> None:
        """完成回调处理程序。"""
        self._wandb.finish()  # type: ignore[attr-defined]

on_event_start #

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

按事件类型存储事件开始数据。

Parameters:

Name Type Description Default
event_type CBEventType

要存储的事件类型。

required
payload Optional[Dict[str, Any]]

要存储的有效负载。

None
event_id str

要存储的事件ID。

''
parent_id str

父事件ID。

''
Source code in llama_index/callbacks/wandb/base.py
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
    def on_event_start(
        self,
        event_type: CBEventType,
        payload: Optional[Dict[str, Any]] = None,
        event_id: str = "",
        parent_id: str = "",
        **kwargs: Any,
    ) -> str:
        """按事件类型存储事件开始数据。

Args:
    event_type (CBEventType): 要存储的事件类型。
    payload (Optional[Dict[str, Any]]): 要存储的有效负载。
    event_id (str): 要存储的事件ID。
    parent_id (str): 父事件ID。
"""
        event = CBEvent(event_type, payload=payload, id_=event_id)
        self._event_pairs_by_id[event.id_].append(event)
        return event.id_

on_event_end #

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

按事件类型存储事件结束数据。

Parameters:

Name Type Description Default
event_type CBEventType

要存储的事件类型。

required
payload Optional[Dict[str, Any]]

要存储的有效负载。

None
event_id str

要存储的事件ID。

''
Source code in llama_index/callbacks/wandb/base.py
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
    def on_event_end(
        self,
        event_type: CBEventType,
        payload: Optional[Dict[str, Any]] = None,
        event_id: str = "",
        **kwargs: Any,
    ) -> None:
        """按事件类型存储事件结束数据。

Args:
    event_type (CBEventType): 要存储的事件类型。
    payload (Optional[Dict[str, Any]]): 要存储的有效负载。
    event_id (str): 要存储的事件ID。
"""
        event = CBEvent(event_type, payload=payload, id_=event_id)
        self._event_pairs_by_id[event.id_].append(event)
        self._trace_map = defaultdict(list)

start_trace #

start_trace(trace_id: Optional[str] = None) -> None

启动一个跟踪。

Source code in llama_index/callbacks/wandb/base.py
201
202
203
204
205
def start_trace(self, trace_id: Optional[str] = None) -> None:
    """启动一个跟踪。"""
    self._trace_map = defaultdict(list)
    self._cur_trace_id = trace_id
    self._start_time = datetime.now()

log_trace_tree #

log_trace_tree() -> None

将跟踪树记录到wandb。

Source code in llama_index/callbacks/wandb/base.py
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
def log_trace_tree(self) -> None:
    """将跟踪树记录到wandb。"""
    try:
        child_nodes = self._trace_map["root"]
        root_span = self._convert_event_pair_to_wb_span(
            self._event_pairs_by_id[child_nodes[0]],
            trace_id=self._cur_trace_id if len(child_nodes) > 1 else None,
        )

        if len(child_nodes) == 1:
            child_nodes = self._trace_map[child_nodes[0]]
            root_span = self._build_trace_tree(child_nodes, root_span)
        else:
            root_span = self._build_trace_tree(child_nodes, root_span)
        if root_span:
            root_trace = self._trace_tree.WBTraceTree(root_span)
            if self._wandb.run:  # type: ignore[attr-defined]
                self._wandb.run.log({"trace": root_trace})  # type: ignore[attr-defined]
            self._wandb.termlog("Logged trace tree to W&B.")  # type: ignore[attr-defined]
    except Exception as e:
        print(f"Failed to log trace tree to W&B: {e}")

persist_index #

persist_index(
    index: IndexType,
    index_name: str,
    persist_dir: Union[str, None] = None,
) -> None

将索引上传到wandb作为一个artifact。您可以在这里了解更多关于W&B artifacts的信息:https://docs.wandb.ai/guides/artifacts。

对于ComposableGraph索引,根id被存储为artifact元数据。

Parameters:

Name Type Description Default
index IndexType

要上传的索引。

required
index_name str

索引的名称。这将被用作artifact的名称。

required
persist_dir Union[str, None]

持久化索引的目录。如果为None,将创建并使用临时目录。

None
Source code in llama_index/callbacks/wandb/base.py
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
    def persist_index(
        self, index: "IndexType", index_name: str, persist_dir: Union[str, None] = None
    ) -> None:
        """将索引上传到wandb作为一个artifact。您可以在这里了解更多关于W&B artifacts的信息:https://docs.wandb.ai/guides/artifacts。

对于`ComposableGraph`索引,根id被存储为artifact元数据。

Args:
    index (IndexType): 要上传的索引。
    index_name (str): 索引的名称。这将被用作artifact的名称。
    persist_dir (Union[str, None]): 持久化索引的目录。如果为None,将创建并使用临时目录。
"""
        if persist_dir is None:
            persist_dir = f"{self._wandb.run.dir}/storage"  # type: ignore
            _default_persist_dir = True
        if not os.path.exists(persist_dir):
            os.makedirs(persist_dir)

        if isinstance(index, self._IndexType):
            try:
                index.storage_context.persist(persist_dir)  # type: ignore

                metadata = None
                # For the `ComposableGraph` index, store the root id as metadata
                if isinstance(index, self._IndexType[0]):
                    root_id = index.root_id
                    metadata = {"root_id": root_id}

                self._upload_index_as_wb_artifact(persist_dir, index_name, metadata)
            except Exception as e:
                # Silently ignore errors to not break user code
                self._print_upload_index_fail_message(e)

        # clear the default storage dir
        if _default_persist_dir:
            shutil.rmtree(persist_dir, ignore_errors=True)

load_storage_context #

load_storage_context(
    artifact_url: str,
    index_download_dir: Union[str, None] = None,
) -> StorageContext

从wandb下载索引并返回存储上下文。

使用此存储上下文,可以使用load_index_from_storageload_indices_from_storageload_graph_from_storage函数将索引加载到内存中。

Parameters:

Name Type Description Default
artifact_url(str):要下载的artifact的url。artifact的url的格式为:`entity/project/index_name

version`,可以在W&B UI中找到。

required
Source code in llama_index/callbacks/wandb/base.py
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
    def load_storage_context(
        self, artifact_url: str, index_download_dir: Union[str, None] = None
    ) -> "StorageContext":
        """从wandb下载索引并返回存储上下文。

使用此存储上下文,可以使用`load_index_from_storage`、`load_indices_from_storage`或`load_graph_from_storage`函数将索引加载到内存中。

Args:
    artifact_url(str):要下载的artifact的url。artifact的url的格式为:`entity/project/index_name:version`,可以在W&B UI中找到。
    index_download_dir(Union[str, None]):要下载索引的目录。
"""
        from llama_index.core.storage.storage_context import StorageContext

        artifact = self._wandb.use_artifact(artifact_url, type="storage_context")  # type: ignore[attr-defined]
        artifact_dir = artifact.download(root=index_download_dir)

        return StorageContext.from_defaults(persist_dir=artifact_dir)

finish #

finish() -> None

完成回调处理程序。

Source code in llama_index/callbacks/wandb/base.py
546
547
548
def finish(self) -> None:
    """完成回调处理程序。"""
    self._wandb.finish()  # type: ignore[attr-defined]