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

[Model] Add user-configurable task for models that support both generation and embedding (#9424)

Browse Source
This commit is contained in:
Cyrus Leung
2024-10-19 02:31:58 +08:00
committed by GitHub
parent 7dbe738d65
commit 051eaf6db3
33 changed files with 451 additions and 201 deletions
Download Patch File Download Diff File Expand all files Collapse all files

92
tests/entrypoints/llm/test_chat.py Normal file
View File

@@ -0,0 +1,92 @@
from typing import List
import pytest
from vllm import LLM
from ..openai.test_vision import TEST_IMAGE_URLS
def test_chat():
llm = LLM(model="meta-llama/Meta-Llama-3-8B-Instruct")
prompt1 = "Explain the concept of entropy."
messages = [
{
"role": "system",
"content": "You are a helpful assistant"
},
{
"role": "user",
"content": prompt1
},
]
outputs = llm.chat(messages)
assert len(outputs) == 1
def test_multi_chat():
llm = LLM(model="meta-llama/Meta-Llama-3-8B-Instruct")
prompt1 = "Explain the concept of entropy."
prompt2 = "Explain what among us is."
conversation1 = [
{
"role": "system",
"content": "You are a helpful assistant"
},
{
"role": "user",
"content": prompt1
},
]
conversation2 = [
{
"role": "system",
"content": "You are a helpful assistant"
},
{
"role": "user",
"content": prompt2
},
]
messages = [conversation1, conversation2]
outputs = llm.chat(messages)
assert len(outputs) == 2
@pytest.mark.parametrize("image_urls",
[[TEST_IMAGE_URLS[0], TEST_IMAGE_URLS[1]]])
def test_chat_multi_image(image_urls: List[str]):
llm = LLM(
model="microsoft/Phi-3.5-vision-instruct",
dtype="bfloat16",
max_model_len=4096,
max_num_seqs=5,
enforce_eager=True,
trust_remote_code=True,
limit_mm_per_prompt={"image": 2},
)
messages = [{
"role":
"user",
"content": [
*({
"type": "image_url",
"image_url": {
"url": image_url
}
} for image_url in image_urls),
{
"type": "text",
"text": "What's in this image?"
},
],
}]
outputs = llm.chat(messages)
assert len(outputs) >= 0

88
tests/entrypoints/llm/test_generate.py
View File

@@ -6,7 +6,6 @@ import pytest
from vllm import LLM, RequestOutput, SamplingParams
from ...conftest import cleanup
from ..openai.test_vision import TEST_IMAGE_URLS
MODEL_NAME = "facebook/opt-125m"
@@ -104,90 +103,3 @@ def test_multiple_sampling_params(llm: LLM):
# sampling_params is None, default params should be applied
outputs = llm.generate(PROMPTS, sampling_params=None)
assert len(PROMPTS) == len(outputs)
def test_chat():
llm = LLM(model="meta-llama/Meta-Llama-3-8B-Instruct")
prompt1 = "Explain the concept of entropy."
messages = [
{
"role": "system",
"content": "You are a helpful assistant"
},
{
"role": "user",
"content": prompt1
},
]
outputs = llm.chat(messages)
assert len(outputs) == 1
def test_multi_chat():
llm = LLM(model="meta-llama/Meta-Llama-3-8B-Instruct")
prompt1 = "Explain the concept of entropy."
prompt2 = "Explain what among us is."
conversation1 = [
{
"role": "system",
"content": "You are a helpful assistant"
},
{
"role": "user",
"content": prompt1
},
]
conversation2 = [
{
"role": "system",
"content": "You are a helpful assistant"
},
{
"role": "user",
"content": prompt2
},
]
messages = [conversation1, conversation2]
outputs = llm.chat(messages)
assert len(outputs) == 2
@pytest.mark.parametrize("image_urls",
[[TEST_IMAGE_URLS[0], TEST_IMAGE_URLS[1]]])
def test_chat_multi_image(image_urls: List[str]):
llm = LLM(
model="microsoft/Phi-3.5-vision-instruct",
dtype="bfloat16",
max_model_len=4096,
max_num_seqs=5,
enforce_eager=True,
trust_remote_code=True,
limit_mm_per_prompt={"image": 2},
)
messages = [{
"role":
"user",
"content": [
*({
"type": "image_url",
"image_url": {
"url": image_url
}
} for image_url in image_urls),
{
"type": "text",
"text": "What's in this image?"
},
],
}]
outputs = llm.chat(messages)
assert len(outputs) >= 0

22
tests/entrypoints/llm/test_init.py Normal file
View File

@@ -0,0 +1,22 @@
import pytest
from vllm import LLM
from ...utils import error_on_warning
MODEL_NAME = "facebook/opt-125m"
def test_pos_args_deprecated():
with error_on_warning(DeprecationWarning):
LLM(model=MODEL_NAME, tokenizer=MODEL_NAME)
with error_on_warning(DeprecationWarning):
LLM(MODEL_NAME, tokenizer=MODEL_NAME)
with pytest.warns(DeprecationWarning, match="'tokenizer'"):
LLM(MODEL_NAME, MODEL_NAME)
with pytest.warns(DeprecationWarning,
match="'tokenizer', 'tokenizer_mode'"):
LLM(MODEL_NAME, MODEL_NAME, "auto")

2
tests/entrypoints/openai/test_serving_chat.py
View File

@@ -22,12 +22,12 @@ class MockHFConfig:
@dataclass
class MockModelConfig:
task = "generate"
tokenizer = MODEL_NAME
trust_remote_code = False
tokenizer_mode = "auto"
max_model_len = 100
tokenizer_revision = None
embedding_mode = False
multimodal_config = MultiModalConfig()
hf_config = MockHFConfig()

2
tests/entrypoints/openai/test_vision.py
View File

@@ -23,6 +23,8 @@ TEST_IMAGE_URLS = [
@pytest.fixture(scope="module")
def server():
args = [
"--task",
"generate",
"--dtype",
"bfloat16",
"--max-model-len",

3
tests/entrypoints/test_chat_utils.py
View File

@@ -18,7 +18,8 @@ PHI3V_MODEL_ID = "microsoft/Phi-3.5-vision-instruct"
@pytest.fixture(scope="module")
def phi3v_model_config():
return ModelConfig(PHI3V_MODEL_ID,
PHI3V_MODEL_ID,
task="generate",
tokenizer=PHI3V_MODEL_ID,
tokenizer_mode="auto",
trust_remote_code=True,
dtype="bfloat16",