MOVZXandHinata
5aa2ac64f8
Production runs one static model behind the OpenAI-compatible API. Every adapter path and non-OpenAI HTTP surface is unreachable there. - Delete the LoRA stack: vllm/lora, LoRARequest on the engine wire, the scheduler, worker, cudagraph and MoE plumbing, LoRA metrics, the lora CLI group, and the LoRA tests. - Drop the plugin loader in vllm/plugins. Platform resolution is now CUDA-only and the general/stat-logger plugin hooks are gone. - Delete the speech-to-text and realtime entrypoints, the scale-out render routes, and vllm/config/speech_to_text.py. GenerationTask is generate-only. - Cut DecodeBenchConnector and the disagg-prefill doc and image. The Multi, Offloading and Example connectors stay; production uses the CPU KV offload tier. - Delete the ROCm-only fused KDA chunk kernel (2346 lines) with its CMake arch gate, declarations and op registrations. The CUDA build never defined the guard and no Python caller existed. - Delete the orphan Mistral and Pixtral chat fixtures and examples/. - Fix _check_model: removing the LoRA early return made every request fail with "The model ... does not exist. Valid aliases: ...". Verified on an RTX 3090: import sweep reports 1242 modules with 7 known optional-dependency failures; a live server answers chat, completions, responses, tokenize and detokenize with zero log errors and 404 for the removed surfaces; collection yields 8558 tests with the 3 pre-existing errors; ruff adds no new finding; the incremental _C build succeeds. Co-authored-by: Hinata <hinata@idihore.id> Signed-off-by: MOVZX <movzx@yahoo.com>
241 lines
7.6 KiB
Python
241 lines
7.6 KiB
Python
# SPDX-License-Identifier: Apache-2.0
|
|
# SPDX-FileCopyrightText: Copyright contributors to the vLLM project
|
|
|
|
from typing import Any
|
|
|
|
import msgspec
|
|
import pytest
|
|
|
|
from vllm.distributed.kv_events import BlockRemoved, BlockStored
|
|
|
|
# Minimal ExternalBlockHash for testing (bytes are a valid ExternalBlockHash).
|
|
_FAKE_HASH: bytes = b"\xab" * 32
|
|
|
|
|
|
class _LegacyBlockStored(
|
|
msgspec.Struct,
|
|
omit_defaults=True, # type: ignore[call-arg]
|
|
gc=False, # type: ignore[call-arg]
|
|
tag="BlockStored", # type: ignore[call-arg]
|
|
):
|
|
"""BlockStored wire schema before locality was added."""
|
|
|
|
block_hashes: list[bytes]
|
|
parent_block_hash: bytes | None
|
|
token_ids: list[int]
|
|
block_size: int
|
|
medium: str | None
|
|
extra_keys: list[tuple[Any, ...] | None] | None = None
|
|
group_idx: int | None = None
|
|
kv_cache_spec_kind: str | None = None
|
|
kv_cache_spec_sliding_window: int | None = None
|
|
|
|
|
|
class _LegacyBlockRemoved(
|
|
msgspec.Struct,
|
|
omit_defaults=True, # type: ignore[call-arg]
|
|
gc=False, # type: ignore[call-arg]
|
|
tag="BlockRemoved", # type: ignore[call-arg]
|
|
):
|
|
"""BlockRemoved wire schema before locality was added."""
|
|
|
|
block_hashes: list[bytes]
|
|
medium: str | None
|
|
group_idx: int | None = None
|
|
|
|
|
|
class _PreSessionBlockStored(
|
|
msgspec.Struct,
|
|
omit_defaults=True, # type: ignore[call-arg]
|
|
gc=False, # type: ignore[call-arg]
|
|
tag="BlockStored", # type: ignore[call-arg]
|
|
):
|
|
"""BlockStored wire schema before session_id was added."""
|
|
|
|
block_hashes: list[bytes]
|
|
parent_block_hash: bytes | None
|
|
token_ids: list[int]
|
|
block_size: int
|
|
medium: str | None
|
|
extra_keys: list[tuple[Any, ...] | None] | None = None
|
|
group_idx: int | None = None
|
|
kv_cache_spec_kind: str | None = None
|
|
kv_cache_spec_sliding_window: int | None = None
|
|
locality: str | None = None
|
|
|
|
|
|
def _make_block_stored(
|
|
group_idx: int | None = None,
|
|
kv_cache_spec_sliding_window: int | None = None,
|
|
locality: str | None = None,
|
|
session_id: str | None = None,
|
|
) -> BlockStored:
|
|
return BlockStored(
|
|
block_hashes=[_FAKE_HASH],
|
|
parent_block_hash=None,
|
|
token_ids=[1, 2, 3, 4],
|
|
block_size=4,
|
|
medium="GPU",
|
|
group_idx=group_idx,
|
|
kv_cache_spec_sliding_window=kv_cache_spec_sliding_window,
|
|
locality=locality,
|
|
session_id=session_id,
|
|
)
|
|
|
|
|
|
def _make_block_removed(
|
|
group_idx: int | None = None,
|
|
locality: str | None = None,
|
|
) -> BlockRemoved:
|
|
return BlockRemoved(
|
|
block_hashes=[_FAKE_HASH],
|
|
medium="GPU",
|
|
group_idx=group_idx,
|
|
locality=locality,
|
|
)
|
|
|
|
|
|
def test_block_stored_default_group_idx_is_none():
|
|
"""group_idx defaults to None when not provided."""
|
|
event = _make_block_stored()
|
|
assert event.group_idx is None
|
|
|
|
|
|
def test_block_removed_default_group_idx_is_none():
|
|
"""group_idx defaults to None when not provided."""
|
|
event = _make_block_removed()
|
|
assert event.group_idx is None
|
|
|
|
|
|
@pytest.mark.parametrize("group_idx", [1, 2, 3])
|
|
def test_block_stored_hash_differs_by_group_idx(group_idx: int):
|
|
"""BlockStored events that differ only in group_idx must hash differently."""
|
|
other_group_idx = group_idx + 1
|
|
event_a = _make_block_stored(group_idx=group_idx)
|
|
event_b = _make_block_stored(group_idx=other_group_idx)
|
|
assert hash(event_a) != hash(event_b)
|
|
|
|
|
|
def test_block_stored_hash_same_for_equal_group_idx():
|
|
"""Two BlockStored events with identical fields produce the same hash."""
|
|
event_a = _make_block_stored(group_idx=1)
|
|
event_b = _make_block_stored(group_idx=1)
|
|
assert hash(event_a) == hash(event_b)
|
|
|
|
|
|
def test_block_stored_hash_differs_by_session_id():
|
|
event_a = _make_block_stored(session_id="session-a")
|
|
event_b = _make_block_stored(session_id="session-b")
|
|
assert hash(event_a) != hash(event_b)
|
|
|
|
|
|
@pytest.mark.parametrize("group_idx", [1, 2, 3])
|
|
def test_block_removed_hash_differs_by_group_idx(group_idx: int):
|
|
"""BlockRemoved events that differ only in group_idx must hash differently."""
|
|
other_group_idx = group_idx + 1
|
|
event_a = _make_block_removed(group_idx=group_idx)
|
|
event_b = _make_block_removed(group_idx=other_group_idx)
|
|
assert hash(event_a) != hash(event_b)
|
|
|
|
|
|
def test_block_removed_hash_same_for_equal_group_idx():
|
|
"""Two BlockRemoved events with identical fields produce the same hash."""
|
|
event_a = _make_block_removed(group_idx=1)
|
|
event_b = _make_block_removed(group_idx=1)
|
|
assert hash(event_a) == hash(event_b)
|
|
|
|
|
|
def test_block_stored_hash_differs_by_sliding_window():
|
|
event_a = _make_block_stored(group_idx=1, kv_cache_spec_sliding_window=128)
|
|
event_b = _make_block_stored(group_idx=1, kv_cache_spec_sliding_window=256)
|
|
assert hash(event_a) != hash(event_b)
|
|
|
|
|
|
@pytest.mark.parametrize(
|
|
("event_a", "event_b"),
|
|
[
|
|
(
|
|
_make_block_stored(locality="LOCAL"),
|
|
_make_block_stored(locality="REMOTE"),
|
|
),
|
|
(
|
|
_make_block_removed(locality="LOCAL"),
|
|
_make_block_removed(locality="REMOTE"),
|
|
),
|
|
],
|
|
)
|
|
def test_event_hash_differs_by_locality(
|
|
event_a: BlockStored | BlockRemoved,
|
|
event_b: BlockStored | BlockRemoved,
|
|
):
|
|
assert hash(event_a) != hash(event_b)
|
|
|
|
|
|
def test_block_stored_locality_is_wire_compatible():
|
|
legacy = _LegacyBlockStored(
|
|
block_hashes=[_FAKE_HASH],
|
|
parent_block_hash=None,
|
|
token_ids=[1, 2, 3, 4],
|
|
block_size=4,
|
|
medium="GPU",
|
|
group_idx=2,
|
|
kv_cache_spec_sliding_window=128,
|
|
)
|
|
legacy_payload = msgspec.msgpack.encode(legacy)
|
|
assert (
|
|
msgspec.msgpack.encode(
|
|
_make_block_stored(
|
|
group_idx=2,
|
|
kv_cache_spec_sliding_window=128,
|
|
)
|
|
)
|
|
== legacy_payload
|
|
)
|
|
assert msgspec.msgpack.decode(legacy_payload, type=BlockStored).locality is None
|
|
new_payload = msgspec.msgpack.encode(_make_block_stored(locality="LOCAL"))
|
|
assert msgspec.msgpack.decode(new_payload)["locality"] == "LOCAL"
|
|
assert msgspec.msgpack.decode(new_payload, type=_LegacyBlockStored).medium == "GPU"
|
|
|
|
|
|
def test_block_removed_locality_is_wire_compatible():
|
|
legacy = _LegacyBlockRemoved(block_hashes=[_FAKE_HASH], medium="GPU")
|
|
legacy_payload = msgspec.msgpack.encode(legacy)
|
|
assert msgspec.msgpack.encode(_make_block_removed()) == legacy_payload
|
|
assert msgspec.msgpack.decode(legacy_payload, type=BlockRemoved).locality is None
|
|
new_payload = msgspec.msgpack.encode(_make_block_removed(locality="REMOTE"))
|
|
assert msgspec.msgpack.decode(new_payload)["locality"] == "REMOTE"
|
|
assert msgspec.msgpack.decode(new_payload, type=_LegacyBlockRemoved).medium == "GPU"
|
|
|
|
|
|
def test_block_stored_session_id_is_wire_compatible():
|
|
pre_session = _PreSessionBlockStored(
|
|
block_hashes=[_FAKE_HASH],
|
|
parent_block_hash=None,
|
|
token_ids=[1, 2, 3, 4],
|
|
block_size=4,
|
|
medium="GPU",
|
|
group_idx=2,
|
|
kv_cache_spec_sliding_window=128,
|
|
locality="LOCAL",
|
|
)
|
|
pre_session_payload = msgspec.msgpack.encode(pre_session)
|
|
assert (
|
|
msgspec.msgpack.encode(
|
|
_make_block_stored(
|
|
group_idx=2,
|
|
kv_cache_spec_sliding_window=128,
|
|
locality="LOCAL",
|
|
)
|
|
)
|
|
== pre_session_payload
|
|
)
|
|
assert (
|
|
msgspec.msgpack.decode(pre_session_payload, type=BlockStored).session_id is None
|
|
)
|
|
|
|
new_payload = msgspec.msgpack.encode(_make_block_stored(session_id="session-1"))
|
|
assert msgspec.msgpack.decode(new_payload)["session_id"] == "session-1"
|
|
assert (
|
|
msgspec.msgpack.decode(new_payload, type=_PreSessionBlockStored).medium == "GPU"
|
|
)
|