Update deprecated Python 3.8 typing (#13971)
This commit is contained in:
@@ -1,7 +1,5 @@
|
||||
# SPDX-License-Identifier: Apache-2.0
|
||||
|
||||
from typing import List, Tuple
|
||||
|
||||
import pytest
|
||||
import torch
|
||||
from transformers import AutoTokenizer
|
||||
@@ -17,8 +15,8 @@ from vllm.transformers_utils.tokenizer_group import init_tokenizer_from_configs
|
||||
|
||||
from tests.v1.engine.utils import FULL_STRINGS # isort: skip
|
||||
|
||||
EngineCoreSampleLogprobsType = List[Tuple[torch.Tensor, torch.Tensor]]
|
||||
EngineCorePromptLogprobsType = Tuple[torch.Tensor, torch.Tensor]
|
||||
EngineCoreSampleLogprobsType = list[tuple[torch.Tensor, torch.Tensor]]
|
||||
EngineCorePromptLogprobsType = tuple[torch.Tensor, torch.Tensor]
|
||||
|
||||
|
||||
def _build_test_vectors_no_logprobs() -> DummyOutputProcessorTestVectors:
|
||||
|
||||
@@ -2,7 +2,7 @@
|
||||
|
||||
import asyncio
|
||||
from contextlib import ExitStack
|
||||
from typing import List, Optional, Tuple
|
||||
from typing import Optional
|
||||
|
||||
import pytest
|
||||
|
||||
@@ -47,7 +47,7 @@ async def generate(engine: AsyncLLM,
|
||||
prompt: PromptType,
|
||||
output_kind: RequestOutputKind,
|
||||
max_tokens: int,
|
||||
prompt_logprobs: Optional[int] = None) -> Tuple[int, str]:
|
||||
prompt_logprobs: Optional[int] = None) -> tuple[int, str]:
|
||||
# Ensure generate doesn't complete too fast for cancellation test.
|
||||
await asyncio.sleep(0.2)
|
||||
|
||||
@@ -114,7 +114,7 @@ async def test_async_llm_refuses_prompt_logprobs_with_apc(
|
||||
(VISION_ENGINE_ARGS, VISION_PROMPT)])
|
||||
@pytest.mark.asyncio
|
||||
async def test_load(monkeypatch, output_kind: RequestOutputKind,
|
||||
engine_args_and_prompt: Tuple[AsyncEngineArgs,
|
||||
engine_args_and_prompt: tuple[AsyncEngineArgs,
|
||||
PromptType]):
|
||||
# TODO(rickyx): Remove monkeypatch once we have a better way to test V1
|
||||
# so that in the future when we switch, we don't have to change all the
|
||||
@@ -160,7 +160,7 @@ async def test_load(monkeypatch, output_kind: RequestOutputKind,
|
||||
(VISION_ENGINE_ARGS, VISION_PROMPT)])
|
||||
@pytest.mark.asyncio
|
||||
async def test_abort(monkeypatch, output_kind: RequestOutputKind,
|
||||
engine_args_and_prompt: Tuple[AsyncEngineArgs,
|
||||
engine_args_and_prompt: tuple[AsyncEngineArgs,
|
||||
PromptType]):
|
||||
|
||||
with monkeypatch.context() as m, ExitStack() as after:
|
||||
@@ -177,7 +177,7 @@ async def test_abort(monkeypatch, output_kind: RequestOutputKind,
|
||||
request_ids = [f"request-{i}" for i in range(NUM_REQUESTS)]
|
||||
|
||||
# Create concurrent requests.
|
||||
tasks: List[asyncio.Task] = []
|
||||
tasks: list[asyncio.Task] = []
|
||||
for request_id in request_ids:
|
||||
tasks.append(
|
||||
asyncio.create_task(
|
||||
|
||||
@@ -5,7 +5,6 @@ import threading
|
||||
import time
|
||||
import uuid
|
||||
from concurrent.futures import Future
|
||||
from typing import List
|
||||
|
||||
import pytest
|
||||
from transformers import AutoTokenizer
|
||||
@@ -213,7 +212,7 @@ def test_engine_core_concurrent_batches(monkeypatch):
|
||||
class DummyExecutor(UniProcExecutor):
|
||||
|
||||
def initialize_from_config(
|
||||
self, kv_cache_configs: List[KVCacheConfig]) -> None:
|
||||
self, kv_cache_configs: list[KVCacheConfig]) -> None:
|
||||
super().initialize_from_config(kv_cache_configs)
|
||||
|
||||
# This executor actually can only run 1 batch at a time
|
||||
|
||||
@@ -3,7 +3,7 @@
|
||||
import asyncio
|
||||
import time
|
||||
import uuid
|
||||
from typing import Dict, List, Optional
|
||||
from typing import Optional
|
||||
|
||||
import pytest
|
||||
from transformers import AutoTokenizer
|
||||
@@ -44,7 +44,7 @@ def make_request(params: SamplingParams) -> EngineCoreRequest:
|
||||
)
|
||||
|
||||
|
||||
def loop_until_done(client: EngineCoreClient, outputs: Dict):
|
||||
def loop_until_done(client: EngineCoreClient, outputs: dict):
|
||||
|
||||
while True:
|
||||
engine_core_outputs = client.get_output().outputs
|
||||
@@ -62,7 +62,7 @@ def loop_until_done(client: EngineCoreClient, outputs: Dict):
|
||||
break
|
||||
|
||||
|
||||
async def loop_until_done_async(client: EngineCoreClient, outputs: Dict):
|
||||
async def loop_until_done_async(client: EngineCoreClient, outputs: dict):
|
||||
|
||||
while True:
|
||||
engine_core_outputs = (await client.get_output_async()).outputs
|
||||
@@ -121,7 +121,7 @@ def test_engine_core_client(monkeypatch, multiprocessing_mode: bool):
|
||||
client.add_request(request)
|
||||
time.sleep(0.01)
|
||||
|
||||
outputs: Dict[str, List] = {req_id: [] for req_id in request_ids}
|
||||
outputs: dict[str, list] = {req_id: [] for req_id in request_ids}
|
||||
loop_until_done(client, outputs)
|
||||
|
||||
for req_id in request_ids:
|
||||
@@ -207,7 +207,7 @@ async def test_engine_core_client_asyncio(monkeypatch):
|
||||
await client.add_request_async(request)
|
||||
await asyncio.sleep(0.01)
|
||||
|
||||
outputs: Dict[str, List] = {req_id: [] for req_id in request_ids}
|
||||
outputs: dict[str, list] = {req_id: [] for req_id in request_ids}
|
||||
await loop_until_done_async(client, outputs)
|
||||
|
||||
for req_id in request_ids:
|
||||
|
||||
@@ -1,7 +1,7 @@
|
||||
# SPDX-License-Identifier: Apache-2.0
|
||||
|
||||
import random
|
||||
from typing import Dict, List, Optional, Tuple
|
||||
from typing import Optional
|
||||
|
||||
import pytest
|
||||
|
||||
@@ -47,9 +47,9 @@ def vllm_model_apc(vllm_runner, monkeypatch):
|
||||
|
||||
|
||||
def _get_test_sampling_params(
|
||||
prompt_list: List[str],
|
||||
prompt_list: list[str],
|
||||
seed: Optional[int] = 42,
|
||||
) -> Tuple[List[SamplingParams], List[int]]:
|
||||
) -> tuple[list[SamplingParams], list[int]]:
|
||||
"""Generate random sampling params for a batch."""
|
||||
|
||||
def get_mostly_n_gt1() -> int:
|
||||
@@ -81,7 +81,7 @@ def test_parallel_sampling(vllm_model, example_prompts) -> None:
|
||||
|
||||
# Validate each request response
|
||||
for out, n in zip(outputs, n_list):
|
||||
completion_counts: Dict[str, int] = {}
|
||||
completion_counts: dict[str, int] = {}
|
||||
# Assert correct number of completions
|
||||
assert len(out.outputs) == n, (
|
||||
f"{len(out.outputs)} completions; {n} expected.")
|
||||
|
||||
@@ -2,7 +2,7 @@
|
||||
|
||||
import math
|
||||
import time
|
||||
from typing import Dict, List, Optional
|
||||
from typing import Optional
|
||||
|
||||
import pytest
|
||||
|
||||
@@ -112,12 +112,12 @@ def test_incremental_detokenization(request_output_kind: RequestOutputKind,
|
||||
|
||||
|
||||
def _validate_logprobs(
|
||||
gen_tokens: Dict[str, List[int]],
|
||||
gen_logprobs: Dict[str, Optional[SampleLogprobs]],
|
||||
gen_prompt_logprobs: Dict[str, Optional[PromptLogprobs]],
|
||||
gen_cumulative_logprob: Dict[str, float],
|
||||
gen_tokens: dict[str, list[int]],
|
||||
gen_logprobs: dict[str, Optional[SampleLogprobs]],
|
||||
gen_prompt_logprobs: dict[str, Optional[PromptLogprobs]],
|
||||
gen_cumulative_logprob: dict[str, float],
|
||||
dtv: DummyOutputProcessorTestVectors,
|
||||
request_id_list: List[str],
|
||||
request_id_list: list[str],
|
||||
num_sample_logprobs: Optional[int],
|
||||
num_prompt_logprobs: Optional[int],
|
||||
) -> None:
|
||||
|
||||
+25
-25
@@ -2,7 +2,7 @@
|
||||
|
||||
import random
|
||||
from dataclasses import dataclass
|
||||
from typing import List, Optional, Tuple, Union
|
||||
from typing import Optional, Union
|
||||
|
||||
import torch
|
||||
from transformers import PreTrainedTokenizer, PreTrainedTokenizerFast
|
||||
@@ -61,7 +61,7 @@ def _create_random_top_logprob_test_vector(
|
||||
|
||||
|
||||
def _create_random_top_logprob_test_matrix(
|
||||
shape: Tuple,
|
||||
shape: tuple,
|
||||
lower: float,
|
||||
upper: float,
|
||||
) -> torch.Tensor:
|
||||
@@ -90,7 +90,7 @@ def _create_random_top_token_test_vector(
|
||||
lower: int,
|
||||
upper: int,
|
||||
sampled_token_id: int,
|
||||
adjust_num_logprobs: bool = True) -> Tuple[torch.Tensor, int]:
|
||||
adjust_num_logprobs: bool = True) -> tuple[torch.Tensor, int]:
|
||||
"""Create a random vector of top logprob token indices
|
||||
|
||||
Use to create fake sample logprobs for testing. The sampled token
|
||||
@@ -141,11 +141,11 @@ def _create_random_top_token_test_vector(
|
||||
|
||||
|
||||
def _create_random_top_token_test_matrix(
|
||||
shape: Tuple[int, int],
|
||||
shape: tuple[int, int],
|
||||
lower: int,
|
||||
upper: int,
|
||||
tokens_list: List[int],
|
||||
) -> Tuple[torch.Tensor, torch.Tensor]:
|
||||
tokens_list: list[int],
|
||||
) -> tuple[torch.Tensor, torch.Tensor]:
|
||||
"""Create a random matrix of top logprob token indices
|
||||
|
||||
Use to create fake prompt logprobs for testing.
|
||||
@@ -160,7 +160,7 @@ def _create_random_top_token_test_matrix(
|
||||
upper: upper range of token ids
|
||||
|
||||
Returns:
|
||||
Tuple containing:
|
||||
tuple containing:
|
||||
- 2D num_tokens x num_logprobs+1 torch Tensor of token ids
|
||||
- 1D tensor of ranks of prompt tokens in their respective
|
||||
rows, or random values
|
||||
@@ -206,10 +206,10 @@ def decode_token(
|
||||
|
||||
|
||||
def generate_dummy_sample_logprobs(
|
||||
sampled_tokens_list: List,
|
||||
sampled_tokens_list: list,
|
||||
num_logprobs: int,
|
||||
tokenizer: PreTrainedTokenizer,
|
||||
) -> List[Tuple[List[int], List[float], int]]:
|
||||
) -> list[tuple[list[int], list[float], int]]:
|
||||
"""Generate dummy sample logprobs
|
||||
|
||||
Generate a test data structure which imitates the list of sample logprobs
|
||||
@@ -221,7 +221,7 @@ def generate_dummy_sample_logprobs(
|
||||
tokenizer: model tokenizer to use for detokenization
|
||||
|
||||
Returns
|
||||
List of (top token ids vector, logprobs vector, sampled token rank)
|
||||
list of (top token ids vector, logprobs vector, sampled token rank)
|
||||
Python lists tuples; in each tuple the logprobs and top token ids
|
||||
vectors have the same length which is either `num_logprobs` or
|
||||
`num_logprobs+1`. Sampled token rank is the rank (index+1) of the
|
||||
@@ -253,7 +253,7 @@ def generate_dummy_sample_logprobs(
|
||||
|
||||
|
||||
def generate_dummy_prompt_logprobs_tensors(
|
||||
prompt_tokens_list: List,
|
||||
prompt_tokens_list: list,
|
||||
num_logprobs: int,
|
||||
tokenizer: PreTrainedTokenizer,
|
||||
) -> LogprobsTensors:
|
||||
@@ -269,7 +269,7 @@ def generate_dummy_prompt_logprobs_tensors(
|
||||
tokenizer: model tokenizer to use for detokenization
|
||||
|
||||
Returns
|
||||
Single Tuple of (logprobs matrix, top token ids matrix) torch Tensor,
|
||||
Single tuple of (logprobs matrix, top token ids matrix) torch Tensor,
|
||||
where both matrices have dimensions
|
||||
num_prompt_tokens x num_logprobs
|
||||
"""
|
||||
@@ -301,19 +301,19 @@ class DummyOutputProcessorTestVectors:
|
||||
tokenizer: GeneralTokenizerType
|
||||
tokenizer_group: BaseTokenizerGroup
|
||||
vllm_config: EngineArgs
|
||||
full_tokens: List[List[int]] # Prompt + generated tokens
|
||||
prompt_tokens: List[List[int]]
|
||||
generation_tokens: List[List[int]]
|
||||
full_tokens: list[list[int]] # Prompt + generated tokens
|
||||
prompt_tokens: list[list[int]]
|
||||
generation_tokens: list[list[int]]
|
||||
# Each request is associated with a tuple of
|
||||
# (top tokens, top logprobs, ranks) prompt logprobs tensors
|
||||
prompt_logprobs: List[LogprobsTensors]
|
||||
prompt_logprobs: list[LogprobsTensors]
|
||||
# Each request is associated with a sample logprobs; a request's
|
||||
# sample logprobs are a list of (top tokens, top logprobs, ranks)
|
||||
# sample logprobs tensors at each sequence position
|
||||
generation_logprobs: List[List[Tuple[List[int], List[float], int]]]
|
||||
prompt_strings: List[str]
|
||||
prompt_strings_len: List[int]
|
||||
generation_strings: List[str]
|
||||
generation_logprobs: list[list[tuple[list[int], list[float], int]]]
|
||||
prompt_strings: list[str]
|
||||
prompt_strings_len: list[int]
|
||||
generation_strings: list[str]
|
||||
|
||||
|
||||
class MockEngineCore:
|
||||
@@ -321,18 +321,18 @@ class MockEngineCore:
|
||||
|
||||
def __init__(
|
||||
self,
|
||||
tokens_list: List[List[int]],
|
||||
tokens_list: list[list[int]],
|
||||
# For each request, for each sampled token offset,
|
||||
# a tuple of
|
||||
# (list of topk token ids, list of sample logprob vals, rank)
|
||||
generated_logprobs_raw: Optional[List[List[Tuple[List[int],
|
||||
List[float],
|
||||
generated_logprobs_raw: Optional[list[list[tuple[list[int],
|
||||
list[float],
|
||||
int]]]] = None,
|
||||
# For each request, a tuple of
|
||||
# (prompt logprob val matrix, prompt logprob tok id matrix);
|
||||
# each matrix has dimensions
|
||||
# (num prompt toks) x (num prompt logprobs+1)
|
||||
prompt_logprobs_raw: Optional[List[LogprobsTensors]] = None,
|
||||
prompt_logprobs_raw: Optional[list[LogprobsTensors]] = None,
|
||||
) -> None:
|
||||
self.tokens_list = tokens_list
|
||||
self.current_idx = 0
|
||||
@@ -341,7 +341,7 @@ class MockEngineCore:
|
||||
self.prompt_logprobs_raw = prompt_logprobs_raw
|
||||
self.do_prompt_logprobs = prompt_logprobs_raw is not None
|
||||
|
||||
def get_outputs(self) -> List[EngineCoreOutput]:
|
||||
def get_outputs(self) -> list[EngineCoreOutput]:
|
||||
do_logprobs = self.do_logprobs
|
||||
do_prompt_logprobs = self.do_prompt_logprobs
|
||||
token_idx = self.current_idx
|
||||
|
||||
Reference in New Issue
Block a user