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

[Chore] Cleanup guided namespace, move to structured outputs config (#22772)

Browse Source 
Signed-off-by: Aaron Pham <contact@aarnphm.xyz>
Signed-off-by: Harry Mellor <19981378+hmellor@users.noreply.github.com>
Co-authored-by: Harry Mellor <19981378+hmellor@users.noreply.github.com>
This commit is contained in:
Aaron Pham
2025-09-18 05:20:27 -04:00
committed by GitHub
parent 05b044e698
commit 29283e8976
51 changed files with 579 additions and 806 deletions
Download Patch File Download Diff File Expand all files Collapse all files

79
tests/entrypoints/openai/test_completion.py
View File

@@ -1,6 +1,6 @@
# SPDX-License-Identifier: Apache-2.0
# SPDX-FileCopyrightText: Copyright contributors to the vLLM project
# imports for guided decoding tests
# imports for structured outputs tests
import json
import os
from typing import Optional
@@ -23,8 +23,6 @@ MODEL_NAME = "HuggingFaceH4/zephyr-7b-beta"
# technically these adapters use a different base model,
# but we're not testing generation quality here
GUIDED_DECODING_BACKENDS = ["outlines", "xgrammar", "guidance"]
@pytest.fixture(scope="module")
def default_server_args(zephyr_lora_files):
@@ -595,12 +593,13 @@ async def test_allowed_token_ids(client: openai.AsyncOpenAI):
@pytest.mark.asyncio
@pytest.mark.parametrize("guided_decoding_backend", GUIDED_DECODING_BACKENDS)
async def test_guided_json_completion(client: openai.AsyncOpenAI,
guided_decoding_backend: str,
sample_json_schema, is_v1_server: bool):
async def test_structured_outputs_json_completion(
client: openai.AsyncOpenAI,
sample_json_schema,
is_v1_server: bool,
):
if not is_v1_server:
pytest.skip("Guided decoding is only supported in v1 engine")
pytest.skip("structured outputs is only supported in v1 engine")
completion = await client.completions.create(
model=MODEL_NAME,
@@ -609,8 +608,7 @@ async def test_guided_json_completion(client: openai.AsyncOpenAI,
n=3,
temperature=1.0,
max_tokens=500,
extra_body=dict(guided_json=sample_json_schema,
guided_decoding_backend=guided_decoding_backend))
extra_body=dict(structured_outputs=dict(json=sample_json_schema)))
assert completion.id is not None
assert len(completion.choices) == 3
@@ -620,12 +618,13 @@ async def test_guided_json_completion(client: openai.AsyncOpenAI,
@pytest.mark.asyncio
@pytest.mark.parametrize("guided_decoding_backend", GUIDED_DECODING_BACKENDS)
async def test_guided_regex_completion(client: openai.AsyncOpenAI,
guided_decoding_backend: str,
sample_regex, is_v1_server: bool):
async def test_structured_outputs_regex_completion(
client: openai.AsyncOpenAI,
sample_regex,
is_v1_server: bool,
):
if not is_v1_server:
pytest.skip("Guided decoding is only supported in v1 engine")
pytest.skip("structured outputs is only supported in v1 engine")
completion = await client.completions.create(
model=MODEL_NAME,
@@ -633,8 +632,7 @@ async def test_guided_regex_completion(client: openai.AsyncOpenAI,
n=3,
temperature=1.0,
max_tokens=20,
extra_body=dict(guided_regex=sample_regex,
guided_decoding_backend=guided_decoding_backend))
extra_body=dict(structured_outputs=dict(regex=sample_regex)))
assert completion.id is not None
assert len(completion.choices) == 3
@@ -644,13 +642,13 @@ async def test_guided_regex_completion(client: openai.AsyncOpenAI,
@pytest.mark.asyncio
@pytest.mark.parametrize("guided_decoding_backend", GUIDED_DECODING_BACKENDS)
async def test_guided_choice_completion(client: openai.AsyncOpenAI,
guided_decoding_backend: str,
sample_guided_choice,
is_v1_server: bool):
async def test_structured_outputs_choice_completion(
client: openai.AsyncOpenAI,
sample_structured_outputs_choices,
is_v1_server: bool,
):
if not is_v1_server:
pytest.skip("Guided decoding is only supported in v1 engine")
pytest.skip("structured outputs is only supported in v1 engine")
completion = await client.completions.create(
model=MODEL_NAME,
@@ -658,20 +656,21 @@ async def test_guided_choice_completion(client: openai.AsyncOpenAI,
n=2,
temperature=1.0,
max_tokens=10,
extra_body=dict(guided_choice=sample_guided_choice,
guided_decoding_backend=guided_decoding_backend))
extra_body=dict(structured_outputs=dict(
choice=sample_structured_outputs_choices)))
assert completion.id is not None
assert len(completion.choices) == 2
for i in range(2):
assert completion.choices[i].text in sample_guided_choice
assert completion.choices[i].text in sample_structured_outputs_choices
@pytest.mark.asyncio
async def test_guided_grammar(client: openai.AsyncOpenAI,
sample_sql_statements, is_v1_server: bool):
async def test_structured_outputs_grammar(client: openai.AsyncOpenAI,
sample_sql_statements,
is_v1_server: bool):
if not is_v1_server:
pytest.skip("Guided grammar is only supported in v1 engine")
pytest.skip("grammar is only supported in v1 engine")
completion = await client.completions.create(
model=MODEL_NAME,
@@ -679,7 +678,8 @@ async def test_guided_grammar(client: openai.AsyncOpenAI,
"table_1 where it is equals to 1"),
temperature=1.0,
max_tokens=500,
extra_body=dict(guided_grammar=sample_sql_statements))
extra_body=dict(
structured_outputs=dict(grammar=sample_sql_statements), ))
content = completion.choices[0].text
@@ -730,27 +730,26 @@ async def test_echo_logprob_completion(client: openai.AsyncOpenAI,
@pytest.mark.asyncio
@pytest.mark.parametrize("guided_decoding_backend", GUIDED_DECODING_BACKENDS)
async def test_guided_decoding_type_error(client: openai.AsyncOpenAI,
guided_decoding_backend: str,
sample_json_schema, sample_regex,
is_v1_server: bool):
async def test_structured_outputs_type_error(client: openai.AsyncOpenAI,
sample_json_schema, sample_regex,
is_v1_server: bool):
if not is_v1_server:
pytest.skip("Guided decoding is only supported in v1 engine")
pytest.skip("structured outputs is only supported in v1 engine")
with pytest.raises(openai.BadRequestError):
_ = await client.completions.create(
model=MODEL_NAME,
prompt="Give an example JSON that fits this schema: 42",
extra_body=dict(guided_json=42,
guided_decoding_backend=guided_decoding_backend))
extra_body=dict(structured_outputs=dict(json=42)))
with pytest.raises(openai.BadRequestError):
_ = await client.completions.create(
model=MODEL_NAME,
prompt="Give an example string that fits this regex",
extra_body=dict(guided_regex=sample_regex,
guided_json=sample_json_schema))
extra_body=dict(structured_outputs=dict(
regex=sample_regex,
json=sample_json_schema,
)))
@pytest.mark.asyncio