[Mini-VLLM] Repair test suite after SM86 cut

- Restore example connector, test helpers, and dropped patch tests
- Guard MLA metadata import in the example connector
- Point the PLE test at common.ops.ple after upstream #60021 move
- Cut tests for removed features: tools, DRY, epd proxy, STT batch,
  diffusion scheduler, INC/Quark PLE cases, mamba-2 mixer

Co-authored-by: Hinata <hinata@idihore.id>
Signed-off-by: MOVZX <movzx@yahoo.com>
This commit is contained in:
2026-10-19 01:06:30 +07:00
co-authored by Hinata
parent 0152af1336
commit 54a396586e
112 changed files with 55726 additions and 10371 deletions
@@ -16,12 +16,9 @@ from vllm.assets.audio import AudioAsset
from vllm.connections import HTTPConnection
from vllm.entrypoints.launchers.run_batch import (
BatchRequestOutput,
BatchTranscriptionRequest,
download_bytes_from_url,
make_transcription_wrapper,
upload_data,
)
from vllm.entrypoints.serve.engine.protocol import ErrorResponse
from vllm.exceptions import VLLMValidationError
from vllm.utils.mem_constants import MiB_bytes
@@ -644,122 +641,10 @@ def test_reasoning_parser():
assert len(reasoning) > 0
def test_transcription():
with (
tempfile.NamedTemporaryFile("w") as input_file,
tempfile.NamedTemporaryFile("r") as output_file,
):
input_file.write(INPUT_TRANSCRIPTION_BATCH)
input_file.flush()
proc = subprocess.Popen(
[
"vllm",
"run-batch",
"-i",
input_file.name,
"-o",
output_file.name,
"--model",
SPEECH_LARGE_MODEL_NAME,
],
)
proc.communicate()
proc.wait()
assert proc.returncode == 0, f"{proc=}"
contents = output_file.read()
print(f"\n\ncontents: {contents}\n\n")
for line in contents.strip().split("\n"):
BatchRequestOutput.model_validate_json(line)
line_dict = json.loads(line)
assert isinstance(line_dict, dict)
assert line_dict["error"] is None
response_body = line_dict["response"]["body"]
assert response_body is not None
assert "text" in response_body
assert "usage" in response_body
def test_transcription_http_url():
with (
tempfile.NamedTemporaryFile("w") as input_file,
tempfile.NamedTemporaryFile("r") as output_file,
):
input_file.write(INPUT_TRANSCRIPTION_HTTP_BATCH)
input_file.flush()
proc = subprocess.Popen(
[
"vllm",
"run-batch",
"-i",
input_file.name,
"-o",
output_file.name,
"--model",
SPEECH_LARGE_MODEL_NAME,
],
)
proc.communicate()
proc.wait()
assert proc.returncode == 0, f"{proc=}"
contents = output_file.read()
for line in contents.strip().split("\n"):
BatchRequestOutput.model_validate_json(line)
line_dict = json.loads(line)
assert isinstance(line_dict, dict)
assert line_dict["error"] is None
response_body = line_dict["response"]["body"]
assert response_body is not None
assert "text" in response_body
assert "usage" in response_body
transcription_text = response_body["text"]
assert "Mary had a little lamb" in transcription_text
def test_translation():
with (
tempfile.NamedTemporaryFile("w") as input_file,
tempfile.NamedTemporaryFile("r") as output_file,
):
input_file.write(INPUT_TRANSLATION_BATCH)
input_file.flush()
proc = subprocess.Popen(
[
"vllm",
"run-batch",
"-i",
input_file.name,
"-o",
output_file.name,
"--model",
SPEECH_SMALL_MODEL_NAME,
],
)
proc.communicate()
proc.wait()
assert proc.returncode == 0, f"{proc=}"
contents = output_file.read()
for line in contents.strip().split("\n"):
BatchRequestOutput.model_validate_json(line)
line_dict = json.loads(line)
assert isinstance(line_dict, dict)
assert line_dict["error"] is None
response_body = line_dict["response"]["body"]
assert response_body is not None
assert "text" in response_body
translation_text = response_body["text"]
translation_text_lower = str(translation_text).strip().lower()
assert "mary" in translation_text_lower or "lamb" in translation_text_lower
def test_tool_calling():
@@ -974,28 +859,6 @@ async def test_download_bytes_backslash_bypass():
)
@pytest.mark.asyncio
async def test_transcription_wrapper_rejects_oversized_data_url_before_decode(
monkeypatch: pytest.MonkeyPatch,
):
monkeypatch.setattr(envs, "VLLM_MAX_AUDIO_CLIP_FILESIZE_MB", 1)
handler = AsyncMock()
wrapped_handler = make_transcription_wrapper(is_translation=False)(handler)
request = BatchTranscriptionRequest.model_validate(
{
"model": SPEECH_LARGE_MODEL_NAME,
"file_url": f"data:audio/wav;base64,{_OVERSIZED_BASE64_AUDIO}",
"response_format": "json",
}
)
with patch("vllm.entrypoints.launchers.run_batch.base64.b64decode") as decode:
response = await wrapped_handler(request)
assert isinstance(response, ErrorResponse)
assert "Maximum file size exceeded" in response.error.message
decode.assert_not_called()
handler.assert_not_awaited()
@pytest.mark.asyncio
@@ -1019,35 +882,6 @@ async def test_download_bytes_allows_http_body_at_audio_limit(
await _close_async_connection(connection)
@pytest.mark.asyncio
async def test_transcription_wrapper_rejects_oversized_http_before_handler(
monkeypatch: pytest.MonkeyPatch,
batch_http_server: _BatchHTTPServer,
):
monkeypatch.setattr(envs, "VLLM_MAX_AUDIO_CLIP_FILESIZE_MB", 1)
connection = HTTPConnection()
monkeypatch.setattr(
"vllm.entrypoints.launchers.run_batch.global_http_connection",
connection,
)
handler = AsyncMock()
wrapped_handler = make_transcription_wrapper(is_translation=False)(handler)
request = BatchTranscriptionRequest.model_validate(
{
"model": SPEECH_LARGE_MODEL_NAME,
"file_url": _batch_http_url(batch_http_server, "/oversized-chunked"),
"response_format": "json",
}
)
try:
response = await wrapped_handler(request)
finally:
await _close_async_connection(connection)
assert isinstance(response, ErrorResponse)
assert "Maximum file size exceeded" in response.error.message
handler.assert_not_awaited()
# ---------------------------------------------------------------------------
@@ -11,7 +11,6 @@ import pytest
import pytest_asyncio
from transformers import AutoTokenizer
import vllm.envs as envs
from tests.utils import RemoteOpenAIServer
from vllm.config import ModelConfig
from vllm.v1.engine.detokenizer import check_stop_strings
@@ -140,7 +139,7 @@ async def test_generate_rejects_min_tokens_above_filled_max_tokens(client):
@pytest.mark.asyncio
@pytest.mark.skipif(
envs.VLLM_USE_RUST_FRONTEND,
False,
reason="sampling mask output is not supported by the Rust frontend",
)
@pytest.mark.parametrize(
@@ -218,7 +217,7 @@ async def test_generate_defaults_max_tokens_when_omitted(client):
@pytest.mark.asyncio
@pytest.mark.skipif(
envs.VLLM_USE_RUST_FRONTEND,
False,
reason="parallel sampling (n > 1) is not supported by the Rust frontend",
)
async def test_generate_returns_all_choices_when_n_greater_than_one(client):
@@ -485,7 +484,7 @@ async def test_stop_string_workflow(client, tokenizer, messages):
@pytest.mark.asyncio
@pytest.mark.skipif(
envs.VLLM_USE_RUST_FRONTEND,
False,
reason="--tokens-only is not supported by the Rust frontend",
)
@pytest.mark.parametrize("server", [["--tokens-only"]], indirect=True)
@@ -617,7 +616,7 @@ async def _render_then_generate(client, body) -> tuple[dict, list[int]]:
# either reasoning field is dropped between render and generate.
@pytest.mark.asyncio
@pytest.mark.skipif(
envs.VLLM_USE_RUST_FRONTEND,
False,
reason="the Rust frontend does not forward reasoning fields to the engine",
)
@pytest.mark.parametrize(
@@ -638,7 +637,7 @@ async def test_render_then_generate_forwards_reasoning_parser_kwargs(client, mes
@pytest.mark.asyncio
@pytest.mark.skipif(
envs.VLLM_USE_RUST_FRONTEND,
False,
reason="the Rust frontend does not forward reasoning fields to the engine",
)
@pytest.mark.parametrize(
@@ -679,7 +678,7 @@ def _text_mode_payload(token_ids, **sampling_params) -> dict:
@pytest.mark.asyncio
@pytest.mark.skipif(
envs.VLLM_USE_RUST_FRONTEND,
False,
reason="output_mode is not supported by the Rust frontend",
)
async def test_text_mode_matches_chat_completions(client, tokenizer, messages):
@@ -707,7 +706,7 @@ async def test_text_mode_matches_chat_completions(client, tokenizer, messages):
@pytest.mark.asyncio
@pytest.mark.skipif(
envs.VLLM_USE_RUST_FRONTEND,
False,
reason="output_mode is not supported by the Rust frontend",
)
async def test_text_mode_stream_matches_non_stream(client, tokenizer, messages):
@@ -734,7 +733,7 @@ async def test_text_mode_stream_matches_non_stream(client, tokenizer, messages):
@pytest.mark.asyncio
@pytest.mark.skipif(
envs.VLLM_USE_RUST_FRONTEND,
False,
reason="output_mode is not supported by the Rust frontend",
)
@pytest.mark.parametrize(
@@ -759,7 +758,7 @@ async def test_text_mode_invalid_requests_return_400(client, overrides, expected
@pytest.mark.asyncio
@pytest.mark.skipif(
envs.VLLM_USE_RUST_FRONTEND,
False,
reason="--tokens-only is not supported by the Rust frontend",
)
@pytest.mark.parametrize("server", [["--tokens-only"]], indirect=True)
@@ -778,7 +777,7 @@ async def test_text_mode_rejected_when_tokens_only(client):
@pytest.mark.asyncio
@pytest.mark.skipif(
envs.VLLM_USE_RUST_FRONTEND,
False,
reason="the Rust frontend does not serve /inference/v1/abort_requests",
)
@pytest.mark.parametrize("output_mode", ["tokens", "text"])
@@ -825,7 +824,7 @@ async def test_stream_delivers_abort_finish_chunk(
@pytest.mark.asyncio
@pytest.mark.skipif(
envs.VLLM_USE_RUST_FRONTEND,
False,
reason="the Rust frontend does not serve /inference/v1/abort_requests",
)
async def test_abort_requests_is_served_without_tokens_only(client):