Repository navigation
Expand file tree
/
Copy pathoutputs.py
More file actions
398 lines (327 loc) · 14.3 KB
/
Copy pathoutputs.py
File metadata and controls
398 lines (327 loc) · 14.3 KB
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
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
# SPDX-License-Identifier: Apache-2.0
# SPDX-FileCopyrightText: Copyright contributors to the vLLM project
from collections.abc import MutableSequence
from collections.abc import Sequence as GenericSequence
from dataclasses import dataclass
from typing import Any, Generic
import numpy as np
import torch
from typing_extensions import TypeVar
from vllm.logger import init_logger
from vllm.logprobs import PromptLogprobs, SampleLogprobs
from vllm.lora.request import LoRARequest
from vllm.v1.metrics.stats import RequestSpecDecodeMetrics, RequestStateStats
logger = init_logger(__name__)
@dataclass
class SamplingMask:
"""Per-token sampling support sets aligned with completion token IDs.
Each inner list contains the vocabulary token IDs that survived
top-k / top-p / min-p filtering for the corresponding generated token.
"""
token_ids: list[list[int]]
@dataclass
class CompletionOutput:
"""The output data of one completion output of a request.
Args:
index: The index of the output in the request.
text: The generated output text.
token_ids: The token IDs of the generated output text.
cumulative_logprob: The cumulative log probability of the generated
output text.
logprobs: The log probabilities of the top probability words at each
position if the logprobs are requested.
sampling_mask: The post-processing token support set for each generated
token, if requested.
finish_reason: The reason why the sequence is finished.
stop_reason: The stop string or token id that caused the completion
to stop, None if the completion finished for some other reason
including encountering the EOS token.
lora_request: The LoRA request that was used to generate the output.
spec_decode_metrics: Per-sequence speculative-decoding acceptance metrics,
populated on finish when speculative decoding ran and
``--per-request-spec-decode-metrics`` is enabled; None otherwise.
Surfaced in the response as ``metrics.speculative_decoding`` for
single-sequence (``n == 1``) requests.
"""
index: int
text: str
token_ids: GenericSequence[int]
cumulative_logprob: float | None
logprobs: SampleLogprobs | None
routed_experts: np.ndarray | None = None # [seq_len,layer_num,topk]
finish_reason: str | None = None
stop_reason: int | str | None = None
lora_request: LoRARequest | None = None
sampling_mask: SamplingMask | None = None
spec_decode_metrics: RequestSpecDecodeMetrics | None = None
def finished(self) -> bool:
return self.finish_reason is not None
def __repr__(self) -> str:
return (
f"CompletionOutput(index={self.index}, "
f"text={self.text!r}, "
f"token_ids={self.token_ids}, "
f"routed_experts={self.routed_experts}, "
f"sampling_mask={self.sampling_mask}, "
f"cumulative_logprob={self.cumulative_logprob}, "
f"logprobs={self.logprobs}, "
f"finish_reason={self.finish_reason}, "
f"stop_reason={self.stop_reason})"
)
@dataclass
class PoolingOutput:
"""The output data of one pooling output of a request.
Args:
data: The extracted hidden states.
"""
data: torch.Tensor
def __repr__(self) -> str:
return f"PoolingOutput(data={self.data})"
def __eq__(self, other: object) -> bool:
return isinstance(other, self.__class__) and bool(
(self.data == other.data).all()
)
class RequestOutput:
"""The output data of a completion request to the LLM.
Args:
request_id: The unique ID of the request.
prompt: The prompt string of the request.
For encoder/decoder models, this is the
decoder input prompt.
prompt_token_ids: The token IDs of the prompt.
For encoder/decoder models, this is the
decoder input prompt token ids.
prompt_logprobs: The log probabilities to return per prompt token.
outputs: The output sequences of the request.
finished: Whether the whole request is finished.
metrics: Metrics associated with the request.
lora_request: The LoRA request that was used to generate the output.
encoder_prompt: The encoder prompt string of the request.
None if decoder-only.
encoder_prompt_token_ids: The token IDs of the encoder prompt.
None if decoder-only.
num_cached_tokens: The number of tokens with prefix cache hit.
num_cache_creation_tokens: Prompt tokens currently counted as local
prefix-cache writes for this request.
kv_transfer_params: The params for remote K/V transfer.
ec_transfer_params: The params for remote encoder-cache transfer.
"""
def __init__(
self,
request_id: str,
prompt: str | None,
prompt_token_ids: list[int] | None,
prompt_logprobs: PromptLogprobs | None,
outputs: list[CompletionOutput],
finished: bool,
metrics: RequestStateStats | None = None,
lora_request: LoRARequest | None = None,
encoder_prompt: str | None = None,
encoder_prompt_token_ids: list[int] | None = None,
num_cached_tokens: int | None = None,
num_cache_creation_tokens: int | None = None,
*,
kv_transfer_params: dict[str, Any] | None = None,
ec_transfer_params: dict[str, Any] | None = None,
# Forward compatibility, code that uses args added in new release can
# still run with older versions of vLLM without breaking.
**kwargs: Any,
) -> None:
if kwargs:
logger.warning_once(
"RequestOutput: Ignoring extra arguments: %s", str(kwargs)
)
self.request_id = request_id
self.prompt = prompt
self.prompt_token_ids = prompt_token_ids
self.prompt_logprobs = prompt_logprobs
self.outputs = outputs
self.finished = finished
self.metrics = metrics
self.lora_request = lora_request
self.encoder_prompt = encoder_prompt
self.encoder_prompt_token_ids = encoder_prompt_token_ids
self.num_cached_tokens = num_cached_tokens
self.num_cache_creation_tokens = num_cache_creation_tokens
self.kv_transfer_params = kv_transfer_params
self.ec_transfer_params = ec_transfer_params
def add(self, next_output: "RequestOutput", aggregate: bool) -> None:
"""Merge subsequent RequestOutput into this one"""
self.finished |= next_output.finished
self.kv_transfer_params = next_output.kv_transfer_params
self.ec_transfer_params = next_output.ec_transfer_params
for next_completion in next_output.outputs:
for i, completion in enumerate(self.outputs):
if completion.index == next_completion.index:
if aggregate:
# Merge outputs with same index
completion.text += next_completion.text
if not isinstance(completion.token_ids, MutableSequence):
completion.token_ids = list(completion.token_ids)
completion.token_ids.extend(next_completion.token_ids)
if next_completion.logprobs:
assert completion.logprobs is not None
completion.logprobs.extend(next_completion.logprobs) # type: ignore[arg-type]
completion.cumulative_logprob = (
next_completion.cumulative_logprob
)
# R3 is returned on the terminal output and must survive
# aggregation with earlier chunks that have no R3.
if next_completion.routed_experts is not None:
completion.routed_experts = next_completion.routed_experts
completion.finish_reason = next_completion.finish_reason
completion.stop_reason = next_completion.stop_reason
else:
# Replace the output with the new one
self.outputs[i] = next_completion
break
else:
self.outputs.append(next_completion)
def __repr__(self) -> str:
return (
f"RequestOutput(request_id={self.request_id}, "
f"prompt={self.prompt!r}, "
f"prompt_token_ids={self.prompt_token_ids}, "
f"encoder_prompt={self.encoder_prompt!r}, "
f"encoder_prompt_token_ids={self.encoder_prompt_token_ids}, "
f"prompt_logprobs={self.prompt_logprobs}, "
f"outputs={self.outputs}, "
f"finished={self.finished}, "
f"metrics={self.metrics}, "
f"lora_request={self.lora_request}, "
f"num_cached_tokens={self.num_cached_tokens}, "
f"num_cache_creation_tokens={self.num_cache_creation_tokens})"
)
# Sentinel to indicate request is finished, used with streaming inputs.
STREAM_FINISHED = RequestOutput(
request_id="",
prompt=None,
prompt_token_ids=None,
prompt_logprobs=None,
outputs=[],
finished=True,
)
_O = TypeVar("_O", default=PoolingOutput)
class PoolingRequestOutput(Generic[_O]):
"""The output data of a pooling request to the LLM.
Args:
request_id (str): A unique identifier for the pooling request.
outputs (PoolingOutput): The pooling results for the given input.
prompt_token_ids (list[int]): A list of token IDs used in the prompt.
num_cached_tokens: The number of tokens with prefix cache hit.
finished (bool): A flag indicating whether the pooling is completed.
"""
def __init__(
self,
request_id: str,
outputs: _O,
prompt_token_ids: list[int],
num_cached_tokens: int,
finished: bool,
):
self.request_id = request_id
self.prompt_token_ids = prompt_token_ids
self.num_cached_tokens = num_cached_tokens
self.finished = finished
self.outputs = outputs
def __repr__(self) -> str:
return (
f"{type(self).__name__}(request_id={self.request_id!r}, "
f"outputs={self.outputs!r}, "
f"prompt_token_ids={self.prompt_token_ids}, "
f"num_cached_tokens={self.num_cached_tokens}, "
f"finished={self.finished})"
)
@dataclass
class EmbeddingOutput:
"""The output data of one embedding output of a request.
Args:
embedding: The embedding vector, which is a list of floats.
Its length depends on the hidden dimension of the model.
"""
embedding: list[float]
@staticmethod
def from_base(pooling_output: PoolingOutput) -> "EmbeddingOutput":
pooled_data = pooling_output.data
if pooled_data.ndim != 1:
raise ValueError("pooled_data should be a 1-D embedding vector")
return EmbeddingOutput(pooled_data.tolist())
@property
def hidden_size(self) -> int:
return len(self.embedding)
def __repr__(self) -> str:
return f"EmbeddingOutput(hidden_size={self.hidden_size})"
class EmbeddingRequestOutput(PoolingRequestOutput[EmbeddingOutput]):
@staticmethod
def from_base(
request_output: PoolingRequestOutput,
) -> "EmbeddingRequestOutput":
return EmbeddingRequestOutput(
request_id=request_output.request_id,
outputs=EmbeddingOutput.from_base(request_output.outputs),
prompt_token_ids=request_output.prompt_token_ids,
num_cached_tokens=request_output.num_cached_tokens,
finished=request_output.finished,
)
@dataclass
class ClassificationOutput:
"""The output data of one classification output of a request.
Args:
probs: The probability vector, which is a list of floats.
Its length depends on the number of classes.
"""
probs: list[float]
@staticmethod
def from_base(pooling_output: PoolingOutput) -> "ClassificationOutput":
# pooling_output shape: (num_classes)
pooled_data = pooling_output.data
if pooled_data.ndim != 1:
raise ValueError("pooled_data should be a 1-D probability vector")
return ClassificationOutput(pooled_data.tolist())
@property
def num_classes(self) -> int:
return len(self.probs)
def __repr__(self) -> str:
return f"ClassificationOutput(num_classes={self.num_classes})"
class ClassificationRequestOutput(PoolingRequestOutput[ClassificationOutput]):
@staticmethod
def from_base(
request_output: PoolingRequestOutput,
) -> "ClassificationRequestOutput":
return ClassificationRequestOutput(
request_id=request_output.request_id,
outputs=ClassificationOutput.from_base(request_output.outputs),
prompt_token_ids=request_output.prompt_token_ids,
num_cached_tokens=request_output.num_cached_tokens,
finished=request_output.finished,
)
@dataclass
class ScoringOutput:
"""The output data of one scoring output of a request.
Args:
score: The similarity score, which is a scalar value.
"""
score: float
@staticmethod
def from_base(pooling_output: PoolingOutput) -> "ScoringOutput":
# pooling_output shape:
# classify task: (num_classes) num_classes == 1
# embed task: a scalar value
pooled_data = pooling_output.data.squeeze()
if pooled_data.ndim != 0:
raise ValueError("pooled_data should be a scalar score")
return ScoringOutput(pooled_data.item())
def __repr__(self) -> str:
return f"ScoringOutput(score={self.score})"
class ScoringRequestOutput(PoolingRequestOutput[ScoringOutput]):
@staticmethod
def from_base(
request_output: PoolingRequestOutput,
) -> "ScoringRequestOutput":
return ScoringRequestOutput(
request_id=request_output.request_id,
outputs=ScoringOutput.from_base(request_output.outputs),
prompt_token_ids=request_output.prompt_token_ids,
num_cached_tokens=request_output.num_cached_tokens,
finished=request_output.finished,
)