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

[V0 deprecation] Guided decoding (#21347)

Browse Source 
Signed-off-by: Reza Barazesh <rezabarazesh@meta.com>
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:
Reza Barazesh
2025-07-29 03:15:30 -07:00
committed by GitHub
parent a4528f0cac
commit 37efc63b64
29 changed files with 103 additions and 2809 deletions
Download Patch File Download Diff File Expand all files Collapse all files

151
tests/entrypoints/openai/test_chat.py
View File

@@ -488,7 +488,9 @@ async def test_chat_completion_stream_options(client: openai.AsyncOpenAI,
@pytest.mark.asyncio
async def test_guided_choice_chat(client: openai.AsyncOpenAI,
sample_guided_choice):
sample_guided_choice, is_v1_server: bool):
if not is_v1_server:
pytest.skip("Guided decoding is only supported in v1 engine")
messages = [{
"role": "system",
"content": "you are a helpful assistant"
@@ -524,8 +526,10 @@ async def test_guided_choice_chat(client: openai.AsyncOpenAI,
@pytest.mark.asyncio
async def test_guided_json_chat(client: openai.AsyncOpenAI,
sample_json_schema):
async def test_guided_json_chat(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")
messages = [{
"role": "system",
@@ -568,7 +572,10 @@ async def test_guided_json_chat(client: openai.AsyncOpenAI,
@pytest.mark.asyncio
async def test_guided_regex_chat(client: openai.AsyncOpenAI, sample_regex):
async def test_guided_regex_chat(client: openai.AsyncOpenAI, sample_regex,
is_v1_server: bool):
if not is_v1_server:
pytest.skip("Guided decoding is only supported in v1 engine")
messages = [{
"role": "system",
@@ -653,7 +660,10 @@ async def test_guided_choice_chat_logprobs(client: openai.AsyncOpenAI,
@pytest.mark.asyncio
async def test_named_tool_use(client: openai.AsyncOpenAI, sample_json_schema):
async def test_named_tool_use(client: openai.AsyncOpenAI, sample_json_schema,
is_v1_server: bool):
if not is_v1_server:
pytest.skip("Tool use is only supported in v1 engine")
messages = [{
"role": "system",
"content": "you are a helpful assistant"
@@ -741,131 +751,6 @@ async def test_named_tool_use(client: openai.AsyncOpenAI, sample_json_schema):
assert json1["age"] != json2["age"]
@pytest.mark.asyncio
@pytest.mark.parametrize("model_name", [MODEL_NAME])
async def test_required_tool_use(client: openai.AsyncOpenAI,
is_v1_server: bool, model_name: str):
if is_v1_server:
pytest.skip(
"tool_choice='required' requires features unsupported on V1")
tools = [
{
"type": "function",
"function": {
"name": "get_current_weather",
"description": "Get the current weather in a given location",
"parameters": {
"type": "object",
"properties": {
"city": {
"type": "string",
"description":
"The city to find the weather for, e.g. 'Vienna'",
"default": "Vienna",
},
"country": {
"type":
"string",
"description":
"The country that the city is in, e.g. 'Austria'",
},
"unit": {
"type": "string",
"description":
"The unit to fetch the temperature in",
"enum": ["celsius", "fahrenheit"],
},
},
"required": ["country", "unit"],
},
},
},
{
"type": "function",
"function": {
"name": "get_forecast",
"description": "Get the weather forecast for a given location",
"parameters": {
"type": "object",
"properties": {
"city": {
"type": "string",
"description":
"The city to get the forecast for, e.g. 'Vienna'",
"default": "Vienna",
},
"country": {
"type":
"string",
"description":
"The country that the city is in, e.g. 'Austria'",
},
"days": {
"type":
"integer",
"description":
"Number of days to get the forecast for (1-7)",
},
"unit": {
"type": "string",
"description":
"The unit to fetch the temperature in",
"enum": ["celsius", "fahrenheit"],
},
},
"required": ["country", "days", "unit"],
},
},
},
]
messages = [
{
"role": "user",
"content": "Hi! How are you doing today?"
},
{
"role": "assistant",
"content": "I'm doing well! How can I help you?"
},
{
"role":
"user",
"content":
"Can you tell me what the current weather is in Berlin and the "\
"forecast for the next 5 days, in fahrenheit?",
},
]
# Non-streaming test
chat_completion = await client.chat.completions.create(
messages=messages,
model=model_name,
tools=tools,
tool_choice="required",
)
assert chat_completion.choices[0].message.tool_calls is not None
assert len(chat_completion.choices[0].message.tool_calls) > 0
# Streaming test
stream = await client.chat.completions.create(
messages=messages,
model=model_name,
tools=tools,
tool_choice="required",
stream=True,
)
output = []
async for chunk in stream:
if chunk.choices and chunk.choices[0].delta.tool_calls:
output.extend(chunk.choices[0].delta.tool_calls)
assert len(output) > 0
@pytest.mark.asyncio
async def test_inconsistent_tool_choice_and_tools(client: openai.AsyncOpenAI,
sample_json_schema):
@@ -948,7 +833,11 @@ async def test_response_format_json_object(client: openai.AsyncOpenAI):
@pytest.mark.asyncio
async def test_response_format_json_schema(client: openai.AsyncOpenAI):
async def test_response_format_json_schema(client: openai.AsyncOpenAI,
is_v1_server: bool):
if not is_v1_server:
pytest.skip(
"JSON schema response format is only supported in v1 engine")
prompt = 'what is 1+1? The format is "result": 2'
# Check that this prompt cannot lead to a valid JSON without json_schema
for _ in range(2):

36
tests/entrypoints/openai/test_completion.py
View File

@@ -28,7 +28,7 @@ MODEL_NAME = "HuggingFaceH4/zephyr-7b-beta"
# but we're not testing generation quality here
LORA_NAME = "typeof/zephyr-7b-beta-lora"
GUIDED_DECODING_BACKENDS = ["outlines", "lm-format-enforcer", "xgrammar"]
GUIDED_DECODING_BACKENDS = ["outlines", "xgrammar", "guidance"]
@pytest.fixture(scope="module")
@@ -95,6 +95,14 @@ def server(default_server_args, request):
os.environ['VLLM_USE_V1'] = original_value
@pytest.fixture
def is_v1_server(server):
import os
# For completion tests, we assume v0 since there's no explicit v1 setup
return os.environ.get('VLLM_USE_V1', '0') == '1'
@pytest_asyncio.fixture
async def client(server):
async with server.get_async_client() as async_client:
@@ -631,7 +639,10 @@ async def test_allowed_token_ids(client: openai.AsyncOpenAI):
@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):
sample_json_schema, is_v1_server: bool):
if not is_v1_server:
pytest.skip("Guided decoding is only supported in v1 engine")
completion = await client.completions.create(
model=MODEL_NAME,
prompt=f"Give an example JSON for an employee profile "
@@ -653,7 +664,10 @@ async def test_guided_json_completion(client: openai.AsyncOpenAI,
@pytest.mark.parametrize("guided_decoding_backend", GUIDED_DECODING_BACKENDS)
async def test_guided_regex_completion(client: openai.AsyncOpenAI,
guided_decoding_backend: str,
sample_regex):
sample_regex, is_v1_server: bool):
if not is_v1_server:
pytest.skip("Guided decoding is only supported in v1 engine")
completion = await client.completions.create(
model=MODEL_NAME,
prompt=f"Give an example IPv4 address with this regex: {sample_regex}",
@@ -674,7 +688,11 @@ async def test_guided_regex_completion(client: openai.AsyncOpenAI,
@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):
sample_guided_choice,
is_v1_server: bool):
if not is_v1_server:
pytest.skip("Guided decoding is only supported in v1 engine")
completion = await client.completions.create(
model=MODEL_NAME,
prompt="The best language for type-safe systems programming is ",
@@ -692,7 +710,9 @@ async def test_guided_choice_completion(client: openai.AsyncOpenAI,
@pytest.mark.asyncio
async def test_guided_grammar(client: openai.AsyncOpenAI,
sample_sql_statements):
sample_sql_statements, is_v1_server: bool):
if not is_v1_server:
pytest.skip("Guided grammar is only supported in v1 engine")
completion = await client.completions.create(
model=MODEL_NAME,
@@ -754,7 +774,11 @@ async def test_echo_logprob_completion(client: openai.AsyncOpenAI,
@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):
sample_json_schema, sample_regex,
is_v1_server: bool):
if not is_v1_server:
pytest.skip("Guided decoding is only supported in v1 engine")
with pytest.raises(openai.BadRequestError):
_ = await client.completions.create(
model=MODEL_NAME,

26
tests/entrypoints/openai/test_prompt_validation.py
View File

@@ -9,6 +9,11 @@ import regex as re
from ...utils import RemoteOpenAIServer
@pytest.fixture(scope="function", autouse=True)
def use_v1_only(monkeypatch):
monkeypatch.setenv('VLLM_USE_V1', '1')
@pytest.mark.asyncio
async def test_empty_prompt():
model_name = "gpt2"
@@ -37,24 +42,3 @@ async def test_out_of_vocab_token_ids():
prompt=[999999],
max_tokens=5,
temperature=0.0)
@pytest.mark.asyncio
async def test_reject_multistep_with_guided_decoding():
model_name = "gpt2"
server_args = ["--enforce-eager", "--num-scheduler-steps", "8"]
with RemoteOpenAIServer(model_name, server_args) as remote_server:
client = remote_server.get_async_client()
with pytest.raises(
openai.BadRequestError,
match=re.compile(
'.*Guided decoding .* multi-step decoding.*').pattern):
await client.completions.create(
model=model_name,
prompt="Hello",
max_tokens=5,
temperature=0.0,
extra_body={"response_format": {
"type": "json_object"
}})