biondizzle/vllm
1
0
Fork 0
Code Issues Pull Requests Actions 2 Packages Projects Releases Wiki Activity

[Deprecation] Remove prompt_token_ids arg fallback in LLM.generate and LLM.embed (#18800)

Browse Source 
Signed-off-by: DarkLight1337 <tlleungac@connect.ust.hk>
This commit is contained in:
Cyrus Leung
2025-08-22 10:56:57 +08:00
committed by GitHub
parent 19fe1a0510
commit 8896eb72eb
24 changed files with 116 additions and 467 deletions
Download Patch File Download Diff File Expand all files Collapse all files

15
tests/entrypoints/llm/test_chat.py
View File

@@ -18,10 +18,9 @@ def text_llm():
enforce_eager=True,
seed=0)
with llm.deprecate_legacy_api():
yield weakref.proxy(llm)
yield weakref.proxy(llm)
del llm
del llm
cleanup_dist_env_and_memory()
@@ -88,10 +87,9 @@ def vision_llm():
seed=0,
)
with llm.deprecate_legacy_api():
yield weakref.proxy(llm)
yield weakref.proxy(llm)
del llm
del llm
cleanup_dist_env_and_memory()
@@ -158,10 +156,9 @@ def thinking_llm():
seed=0,
)
with llm.deprecate_legacy_api():
yield weakref.proxy(llm)
yield weakref.proxy(llm)
del llm
del llm
cleanup_dist_env_and_memory()

5
tests/entrypoints/llm/test_classify.py
View File

@@ -35,10 +35,9 @@ def llm():
enforce_eager=True,
seed=0)
with llm.deprecate_legacy_api():
yield weakref.proxy(llm)
yield weakref.proxy(llm)
del llm
del llm
cleanup_dist_env_and_memory()

5
tests/entrypoints/llm/test_embedding.py
View File

@@ -26,10 +26,9 @@ def llm():
enforce_eager=True,
seed=0)
with llm.deprecate_legacy_api():
yield weakref.proxy(llm)
yield weakref.proxy(llm)
del llm
del llm
cleanup_dist_env_and_memory()

52
tests/entrypoints/llm/test_encode.py
View File

@@ -5,11 +5,9 @@ import weakref
import pytest
from vllm import LLM, PoolingParams, PoolingRequestOutput
from vllm import LLM, PoolingParams
from vllm.distributed import cleanup_dist_env_and_memory
from ...models.utils import check_embeddings_close
MODEL_NAME = "intfloat/multilingual-e5-small"
PROMPTS = [
@@ -48,57 +46,13 @@ def llm():
enforce_eager=True,
seed=0)
with llm.deprecate_legacy_api():
yield weakref.proxy(llm)
yield weakref.proxy(llm)
del llm
del llm
cleanup_dist_env_and_memory()
def assert_outputs_match(o1: list[PoolingRequestOutput],
o2: list[PoolingRequestOutput]):
check_embeddings_close(
embeddings_0_lst=[o.outputs.data for o in o1],
embeddings_1_lst=[o.outputs.data for o in o2],
name_0="hf",
name_1="vllm",
tol=1e-2,
)
@pytest.mark.skip_global_cleanup
@pytest.mark.parametrize('prompt_token_ids', TOKEN_IDS)
def test_v1_v2_api_consistency_single_prompt_tokens(llm: LLM,
prompt_token_ids):
pooling_params = PoolingParams()
with pytest.warns(DeprecationWarning, match="'prompt_token_ids'"):
v1_output = llm.encode(prompt_token_ids=prompt_token_ids,
pooling_params=pooling_params)
v2_output = llm.encode({"prompt_token_ids": prompt_token_ids},
pooling_params=pooling_params)
assert_outputs_match(v1_output, v2_output)
@pytest.mark.skip_global_cleanup
def test_v1_v2_api_consistency_multi_prompt_tokens(llm: LLM):
pooling_params = PoolingParams()
with pytest.warns(DeprecationWarning, match="'prompt_token_ids'"):
v1_output = llm.encode(prompt_token_ids=TOKEN_IDS,
pooling_params=pooling_params)
v2_output = llm.encode(
[{
"prompt_token_ids": p
} for p in TOKEN_IDS],
pooling_params=pooling_params,
)
assert_outputs_match(v1_output, v2_output)
@pytest.mark.skip_global_cleanup
def test_multiple_pooling_params(llm: LLM):
pooling_params = [

43
tests/entrypoints/llm/test_generate.py
View File

@@ -5,7 +5,7 @@ import weakref
import pytest
from vllm import LLM, RequestOutput, SamplingParams
from vllm import LLM, SamplingParams
from vllm.distributed import cleanup_dist_env_and_memory
MODEL_NAME = "distilbert/distilgpt2"
@@ -41,50 +41,13 @@ def llm():
gpu_memory_utilization=0.10,
enforce_eager=True)
with llm.deprecate_legacy_api():
yield weakref.proxy(llm)
yield weakref.proxy(llm)
del llm
del llm
cleanup_dist_env_and_memory()
def assert_outputs_equal(o1: list[RequestOutput], o2: list[RequestOutput]):
assert [o.outputs for o in o1] == [o.outputs for o in o2]
@pytest.mark.skip_global_cleanup
@pytest.mark.parametrize('prompt_token_ids', TOKEN_IDS)
def test_v1_v2_api_consistency_single_prompt_tokens(llm: LLM,
prompt_token_ids):
sampling_params = SamplingParams(temperature=0.0, top_p=1.0)
with pytest.warns(DeprecationWarning, match="'prompt_token_ids'"):
v1_output = llm.generate(prompt_token_ids=prompt_token_ids,
sampling_params=sampling_params)
v2_output = llm.generate({"prompt_token_ids": prompt_token_ids},
sampling_params=sampling_params)
assert_outputs_equal(v1_output, v2_output)
@pytest.mark.skip_global_cleanup
def test_v1_v2_api_consistency_multi_prompt_tokens(llm: LLM):
sampling_params = SamplingParams(temperature=0.0, top_p=1.0)
with pytest.warns(DeprecationWarning, match="'prompt_token_ids'"):
v1_output = llm.generate(prompt_token_ids=TOKEN_IDS,
sampling_params=sampling_params)
v2_output = llm.generate(
[{
"prompt_token_ids": p
} for p in TOKEN_IDS],
sampling_params=sampling_params,
)
assert_outputs_equal(v1_output, v2_output)
@pytest.mark.skip_global_cleanup
def test_multiple_sampling_params(llm: LLM):
sampling_params = [

5
tests/entrypoints/llm/test_generate_multiple_loras.py
View File

@@ -48,10 +48,9 @@ def llm(request, monkeypatch_module):
max_num_seqs=128,
enforce_eager=True)
with llm.deprecate_legacy_api():
yield weakref.proxy(llm)
yield weakref.proxy(llm)
del llm
del llm
cleanup_dist_env_and_memory()

5
tests/entrypoints/llm/test_reward.py
View File

@@ -36,10 +36,9 @@ def llm():
trust_remote_code=True,
seed=0)
with llm.deprecate_legacy_api():
yield weakref.proxy(llm)
yield weakref.proxy(llm)
del llm
del llm
cleanup_dist_env_and_memory()

5
tests/entrypoints/llm/test_score.py
View File

@@ -33,10 +33,9 @@ def llm():
enforce_eager=True,
seed=0)
with llm.deprecate_legacy_api():
yield weakref.proxy(llm)
yield weakref.proxy(llm)
del llm
del llm
cleanup_dist_env_and_memory()