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

[Core] Deprecating block manager v1 and make block manager v2 default (#8704)

Browse Source 
Removing the block manager v1. This is the initial piece of prefix-caching-centric design. In order to achieve prefix-caching-centric design, we need to simplify the code path so that we only use v2 block manager (which has much higher performance on prefix caching).
This commit is contained in:
Kuntai Du
2024-10-17 11:38:15 -05:00
committed by GitHub
parent 5eda21e773
commit 81ede99ca4
45 changed files with 206 additions and 2109 deletions
Download Patch File Download Diff File Expand all files Collapse all files

68
tests/spec_decode/e2e/test_compatibility.py
View File

@@ -1,27 +1,15 @@
import pytest
from tests.utils import check_deprecated_block_manager_usage
from vllm import SamplingParams
from .conftest import get_output_from_llm_generator
@pytest.fixture(scope="module", autouse=True)
def check_deprecated_block_manager():
check_deprecated_block_manager_usage(
'tests/spec_decode/e2e/test_compatibility.py')
@pytest.mark.parametrize(
"common_llm_kwargs",
[{
"model": "JackFram/llama-68m",
"speculative_model": "JackFram/llama-68m",
"num_speculative_tokens": 5,
# Required for spec decode.
"use_v2_block_manager": True
}])
@pytest.mark.parametrize("common_llm_kwargs", [{
"model": "JackFram/llama-68m",
"speculative_model": "JackFram/llama-68m",
"num_speculative_tokens": 5,
}])
@pytest.mark.parametrize("per_test_common_llm_kwargs", [
{
"enable_chunked_prefill": True,
@@ -51,16 +39,11 @@ def test_spec_decode_xfail_chunked_prefill(test_llm_generator):
sampling_params)
@pytest.mark.parametrize(
"common_llm_kwargs",
[{
"model": "meta-llama/Llama-2-7b-chat-hf",
"speculative_model": "JackFram/llama-68m",
"num_speculative_tokens": 5,
# Required for spec decode.
"use_v2_block_manager": True
}])
@pytest.mark.parametrize("common_llm_kwargs", [{
"model": "meta-llama/Llama-2-7b-chat-hf",
"speculative_model": "JackFram/llama-68m",
"num_speculative_tokens": 5,
}])
@pytest.mark.parametrize(
"per_test_common_llm_kwargs",
[
@@ -101,34 +84,3 @@ def test_spec_decode_xfail_spec_max_model_len(test_llm_generator):
with pytest.raises(ValueError, match="cannot be larger than"):
get_output_from_llm_generator(test_llm_generator, prompts,
sampling_params)
@pytest.mark.parametrize("common_llm_kwargs", [{
"model": "JackFram/llama-68m",
"speculative_model": "JackFram/llama-68m",
"num_speculative_tokens": 5,
"use_v2_block_manager": False,
}])
@pytest.mark.parametrize("per_test_common_llm_kwargs", [{}])
@pytest.mark.parametrize("test_llm_kwargs", [{}])
@pytest.mark.parametrize("seed", [1])
def test_spec_decode_xfail_block_manager_v1(test_llm_generator):
"""Verify that speculative decoding with block manager v1 fails.
"""
output_len = 128
temperature = 0.0
prompts = [
"Hello, my name is",
]
sampling_params = SamplingParams(
max_tokens=output_len,
ignore_eos=True,
temperature=temperature,
)
with pytest.raises(ValueError,
match="Speculative decoding requires usage of the V2"):
get_output_from_llm_generator(test_llm_generator, prompts,
sampling_params)