[Mini-vLLM] Cut dead subsystems: models, entrypoints, backends, utils, tests

- Drop dead model modules, keeping files live imports require
- Remove anthropic, cohere, codegen entrypoints and batch STT modes
- Delete unused attention backends, MoE experts, and NPU/XPU util leaves
- Strip dead third_party trees and ROCm leaves with no CUDA path
- Remove mooncake EC connectors: dependency missing since branch base
- Re-audit verified by full import sweep and Qwen3.5 smoke test

Co-authored-by: Hinata <hinata@idihore.id>
Signed-off-by: MOVZX <movzx@yahoo.com>
This commit is contained in:
2026-10-19 00:20:48 +07:00
co-authored by Hinata
parent 6a95f62e2f
commit 81074e94a3
940 changed files with 11 additions and 380533 deletions
-51
View File
@@ -1,51 +0,0 @@
# SPDX-License-Identifier: Apache-2.0
# SPDX-FileCopyrightText: Copyright contributors to the vLLM project
import pytest
from transformers import AutoTokenizer
from vllm import SamplingParams
from vllm.v1.engine import EngineCoreRequest
from vllm.v1.engine.detokenizer import FastIncrementalDetokenizer
PROMPT = "Hello, my name is Lee, and I'm a student in the " + "college of engineering"
@pytest.mark.parametrize(
"min_tokens,stop,truth",
[
(0, None, " is Lee, and I'm a student in the college of engineering"),
(0, "e", " is L"),
(5, "e", " is Lee, and I'm a stud"),
],
)
def test_min_tokens_with_stop(min_tokens: int, stop: str, truth: str):
"""Test for a specific min_tokens and stop.
See https://github.com/vllm-project/vllm/pull/22014
"""
tokenizer = AutoTokenizer.from_pretrained("facebook/opt-125m")
all_prompt_ids = tokenizer(PROMPT, add_special_tokens=False).input_ids
# The prompt is "Hello, my name is"
prompt_token_ids = all_prompt_ids[:4]
params = SamplingParams(
stop=stop,
min_tokens=min_tokens,
)
request = EngineCoreRequest(
request_id="",
prompt_token_ids=prompt_token_ids,
mm_features=None,
sampling_params=params,
pooling_params=None,
arrival_time=0.0,
lora_request=None,
cache_salt=None,
data_parallel_rank=None,
)
detokenizer = FastIncrementalDetokenizer(tokenizer, request)
detokenizer.update(all_prompt_ids[4:], False)
assert detokenizer.output_text == truth
-69
View File
@@ -1,69 +0,0 @@
# SPDX-License-Identifier: Apache-2.0
# SPDX-FileCopyrightText: Copyright contributors to the vLLM project
"""Test the different finish_reason="stop" situations during generation:
1. One of the provided stop strings
2. One of the provided stop tokens
3. The EOS token
Run `pytest tests/engine/test_stop_reason.py`.
"""
import pytest
import transformers
from vllm import SamplingParams
MODEL = "distilbert/distilgpt2"
STOP_STR = "."
SEED = 42
MAX_TOKENS = 1024
@pytest.fixture
def vllm_model(vllm_runner):
with vllm_runner(MODEL) as vllm_model:
yield vllm_model
def test_stop_reason(vllm_model, example_prompts):
tokenizer = transformers.AutoTokenizer.from_pretrained(MODEL)
stop_token_id = tokenizer.convert_tokens_to_ids(STOP_STR)
llm = vllm_model.llm
# test stop token
outputs = llm.generate(
example_prompts,
sampling_params=SamplingParams(
ignore_eos=True,
seed=SEED,
max_tokens=MAX_TOKENS,
stop_token_ids=[stop_token_id],
),
)
for output in outputs:
output = output.outputs[0]
assert output.finish_reason == "stop"
assert output.stop_reason == stop_token_id
# test stop string
outputs = llm.generate(
example_prompts,
sampling_params=SamplingParams(
ignore_eos=True, seed=SEED, max_tokens=MAX_TOKENS, stop="."
),
)
for output in outputs:
output = output.outputs[0]
assert output.finish_reason == "stop"
assert output.stop_reason == STOP_STR
# test EOS token
outputs = llm.generate(
example_prompts,
sampling_params=SamplingParams(seed=SEED, max_tokens=MAX_TOKENS),
)
for output in outputs:
output = output.outputs[0]
assert output.finish_reason == "length" or (
output.finish_reason == "stop" and output.stop_reason is None
)
-120
View File
@@ -1,120 +0,0 @@
# SPDX-License-Identifier: Apache-2.0
# SPDX-FileCopyrightText: Copyright contributors to the vLLM project
from typing import Any
import pytest
from vllm import LLM, SamplingParams
MODEL = "meta-llama/llama-2-7b-hf"
MAX_TOKENS = 200
def _test_stopping(
llm: LLM,
expected_output: str,
expected_reason: Any,
stop: list[str] | None = None,
stop_token_ids: list[int] | None = None,
include_in_output: bool = False,
) -> None:
output = llm.generate(
"A story about vLLM:\n",
SamplingParams(
temperature=0.0,
max_tokens=MAX_TOKENS,
stop=stop,
stop_token_ids=stop_token_ids,
include_stop_str_in_output=include_in_output,
),
)[0].outputs[0]
assert output is not None
assert output.text == expected_output
assert output.stop_reason == expected_reason
def _stop_basic(llm):
_test_stopping(
llm,
stop=["."],
include_in_output=False,
expected_output="VLLM is a 100% volunteer organization",
expected_reason=".",
)
_test_stopping(
llm,
stop=["."],
include_in_output=True,
expected_output="VLLM is a 100% volunteer organization.",
expected_reason=".",
)
def _stop_multi_tokens(llm):
_test_stopping(
llm,
stop=["group of peo", "short"],
include_in_output=False,
expected_output="VLLM is a 100% volunteer organization. We are a ",
expected_reason="group of peo",
)
_test_stopping(
llm,
stop=["group of peo", "short"],
include_in_output=True,
expected_output="VLLM is a 100% volunteer organization. We are a group of peo",
expected_reason="group of peo",
)
def _stop_partial_token(llm):
_test_stopping(
llm,
stop=["gani"],
include_in_output=False,
expected_output="VLLM is a 100% volunteer or",
expected_reason="gani",
)
_test_stopping(
llm,
stop=["gani"],
include_in_output=True,
expected_output="VLLM is a 100% volunteer organi",
expected_reason="gani",
)
def _stop_token_id(llm):
# token id 13013 => " organization"
_test_stopping(
llm,
stop_token_ids=[13013],
include_in_output=False,
expected_output="VLLM is a 100% volunteer",
expected_reason=13013,
)
_test_stopping(
llm,
stop_token_ids=[13013],
include_in_output=True,
expected_output="VLLM is a 100% volunteer organization",
expected_reason=13013,
)
@pytest.mark.skip_global_cleanup
def test_stop_strings():
llm = LLM(MODEL, enforce_eager=True)
_stop_basic(llm)
_stop_multi_tokens(llm)
_stop_partial_token(llm)
# FIXME: this does not respect include_in_output=False
# _stop_token_id(llm)