sgl-project--sglang
94057c3d3e
PR Test (NPU) / check-changes (push) Has been cancelled
PR Test (NPU) / pr-gate (push) Has been cancelled
PR Test (NPU) / set-image-config (push) Has been cancelled
PR Test (NPU) / stage-b-test-1-npu-a2 (0) (push) Has been cancelled
PR Test (NPU) / stage-b-test-1-npu-a2 (1) (push) Has been cancelled
PR Test (NPU) / stage-b-test-2-npu-a2 (0) (push) Has been cancelled
PR Test (NPU) / stage-b-test-2-npu-a2 (1) (push) Has been cancelled
PR Test (NPU) / stage-b-test-4-npu-a3 (push) Has been cancelled
PR Test (NPU) / stage-b-test-16-npu-a3 (push) Has been cancelled
PR Test (NPU) / multimodal-gen-test-1-npu-a3 (push) Has been cancelled
PR Test (NPU) / multimodal-gen-test-2-npu-a3 (push) Has been cancelled
PR Test (Arm64) / pr-gate (push) Has been cancelled
PR Test (Arm64) / check-changes (push) Has been cancelled
PR Test (Arm64) / build-test (push) Has been cancelled
PR Test (sgl-router) / gate (push) Has been cancelled
PR Test (sgl-router) / tier-1 — lint (push) Has been cancelled
PR Test (sgl-router) / tier-2 — build + test (push) Has been cancelled
PR Test (sgl-router) / tier-3 — docker (placeholder) (push) Has been cancelled
PR Test (sgl-router) / tier-3 — k8s integration (push) Has been cancelled
PR Test (sgl-router) / tier-3 — e2e (push) Has been cancelled
PR Test (sgl-router) / finish (push) Has been cancelled
PR Test (NPU) / single-node-poc (map[name:qwen3_6_27b_w8a8_1p_in64k_out1k_50ms runner:linux-aarch64-a3-2 test_case:test/registered/ascend/performance/qwen3_6_27b/test_npu_qwen3_6_27b_w8a8_1p_in64k_out1k_50ms.py test_type:perf]) (push) Has been cancelled
PR Test (NPU) / pr-test-npu-finish (push) Has been cancelled
PR Test (Xeon) / pr-gate (push) Has been cancelled
PR Test (Xeon) / check-changes (push) Has been cancelled
PR Test (Xeon) / build-test (, xeon-gnr, base-b-test-cpu) (push) Has been cancelled
PR Test (XPU) / check-changes (push) Has been cancelled
PR Test (XPU) / pr-gate (push) Has been cancelled
PR Test (XPU) / stage-a-test-1-gpu-xpu (push) Has been cancelled
PR Test (XPU) / wait-for-stage-a (push) Has been cancelled
PR Test (XPU) / stage-b-test-1-gpu-xpu (push) Has been cancelled
PR Test (XPU) / finish (push) Has been cancelled
CI Model Inventory / build-inventory (push) Has been cancelled
Lint / lint (push) Has been cancelled
PR Benchmark (SMG Components) / Benchmark Compilation Check (push) Has been cancelled
PR Benchmark (SMG Components) / Benchmark - Manual Policy (push) Has been cancelled
PR Benchmark (SMG Components) / Benchmark - Request Processing (push) Has been cancelled
PR Benchmark (SMG Components) / Benchmark Summary (push) Has been cancelled
PR Test (SMG) / build-wheel (push) Has been cancelled
Release SGLang Model Gateway to PyPI / build on windows (x86_64 - auto) (push) Has been cancelled
Release SGLang Model Gateway to PyPI / build on macos (x86_64 - auto) (push) Has been cancelled
PR Test (SMG) / python-unit-tests (push) Has been cancelled
PR Test (SMG) / unit-tests (push) Has been cancelled
PR Test (SMG) / benchmarks (push) Has been cancelled
PR Test (SMG) / chat-completions (push) Has been cancelled
PR Test (SMG) / chat-completions-4gpu (push) Has been cancelled
PR Test (SMG) / e2e (push) Has been cancelled
PR Test (SMG) / docker-build-test (push) Has been cancelled
PR Test (SMG) / k8s-integration (push) Has been cancelled
PR Test (SMG) / finish (push) Has been cancelled
PR Test (SMG) / summarize-benchmarks (push) Has been cancelled
Release SGLang Model Gateway Docker Image / publish (push) Has been cancelled
Release SGLang Model Gateway to PyPI / build on macos (aarch64 - auto) (push) Has been cancelled
Release SGLang Model Gateway to PyPI / build on linux (aarch64 - auto) (push) Has been cancelled
Release SGLang Model Gateway to PyPI / build on linux (x86_64 - auto) (push) Has been cancelled
Release SGLang Model Gateway to PyPI / build on linux (aarch64 - musllinux_1_1) (push) Has been cancelled
Release SGLang Model Gateway to PyPI / build on linux (x86_64 - musllinux_1_1) (push) Has been cancelled
Release SGLang Model Gateway to PyPI / Build SDist (push) Has been cancelled
Release SGLang Model Gateway to PyPI / Upload to PyPI (push) Has been cancelled
Release SGLang Kernels / build-cu129-matrix (aarch64, 12.9, 3.10, arm-kernel-build-node) (push) Has been cancelled
Release SGLang Kernels / build-cu129-matrix (x86_64, 12.9, 3.10, x64-kernel-build-node) (push) Has been cancelled
Release SGLang Kernels / release-cu129 (push) Has been cancelled
Release SGLang Kernels / build-cu130-matrix (aarch64, 13.0, 3.10, arm-kernel-build-node) (push) Has been cancelled
Release SGLang Kernels / build-cu130-matrix (x86_64, 13.0, 3.10, x64-kernel-build-node) (push) Has been cancelled
Release SGLang Kernels / release-cu130 (push) Has been cancelled
Release SGLang Kernels / build-rocm-matrix (3.10, 700) (push) Has been cancelled
Release SGLang Kernels / build-rocm-matrix (3.10, 720) (push) Has been cancelled
Release SGLang Kernels / release-rocm700 (push) Has been cancelled
Release SGLang Kernels / release-rocm720 (push) Has been cancelled
Release SGLang Kernels / build-musa43 (43, 3.10) (push) Has been cancelled
Release SGLang Kernels / release-musa43 (push) Has been cancelled
273 行
11 KiB
Python
273 行
11 KiB
Python
"""Unit tests for srt/speculative/decoupled_spec_io.
|
|
|
|
decoupled_spec_io is the schema-only IPC layer for decoupled speculative
|
|
decoding: protocol message dataclasses, the cross-process request id codec, and
|
|
the drafter-side reconciliation helpers. These tests drive the real logic (id
|
|
round-trip + parse errors, commit validation, segment coalescing / contiguity /
|
|
prefix extraction, and inbox routing) on CPU; there is no GPU or transport here.
|
|
"""
|
|
|
|
import unittest
|
|
|
|
from sglang.srt.speculative.decoupled_spec_io import (
|
|
DraftClose,
|
|
DraftControlBatch,
|
|
DraftMeshMessage,
|
|
DraftMeshMessageType,
|
|
DraftReqKey,
|
|
DraftSync,
|
|
VerifierCommitSegment,
|
|
VerifyCommit,
|
|
build_draft_scheduler_rid,
|
|
parse_draft_scheduler_rid,
|
|
)
|
|
from sglang.test.ci.ci_register import register_cpu_ci
|
|
from sglang.test.test_utils import CustomTestCase
|
|
|
|
register_cpu_ci(est_time=6, suite="base-a-test-cpu")
|
|
|
|
|
|
def _commit(rid, *, pre, tokens, src_verifier_rank=0, drafter_rank=0) -> VerifyCommit:
|
|
return VerifyCommit(
|
|
request_id=rid,
|
|
src_verifier_rank=src_verifier_rank,
|
|
dst_drafter_rank=drafter_rank,
|
|
pre_verify_committed_len=pre,
|
|
committed_tokens=list(tokens),
|
|
)
|
|
|
|
|
|
def _segment(
|
|
rid, *, pre=0, drafter_rank=0, src_verifier_rank=0
|
|
) -> VerifierCommitSegment:
|
|
return VerifierCommitSegment(
|
|
draft_key=DraftReqKey(src_verifier_rank=src_verifier_rank, request_id=rid),
|
|
dst_drafter_rank=drafter_rank,
|
|
pre_verify_committed_len=pre,
|
|
)
|
|
|
|
|
|
class TestDraftSchedulerRid(CustomTestCase):
|
|
def test_round_trip(self):
|
|
key = DraftReqKey(src_verifier_rank=3, request_id="req-1")
|
|
rid = build_draft_scheduler_rid(key)
|
|
self.assertEqual(rid, "draft:3:req-1")
|
|
self.assertEqual(parse_draft_scheduler_rid(rid), key)
|
|
|
|
def test_request_id_containing_colon_round_trips(self):
|
|
# request_id may contain ':' — parse splits on the first ':' only.
|
|
key = DraftReqKey(src_verifier_rank=1, request_id="a:b:c")
|
|
self.assertEqual(parse_draft_scheduler_rid(build_draft_scheduler_rid(key)), key)
|
|
|
|
def test_parse_invalid_rid_raises(self):
|
|
for bad in ["no-prefix", "draft:", "draft:0:", "draft:notint:r"]:
|
|
with self.subTest(rid=bad):
|
|
with self.assertRaises(ValueError):
|
|
parse_draft_scheduler_rid(bad)
|
|
|
|
|
|
class TestVerifyCommitValidation(CustomTestCase):
|
|
def test_empty_tokens_raises(self):
|
|
with self.assertRaises(ValueError):
|
|
_commit("r", pre=0, tokens=[]).validate_committed_tokens()
|
|
|
|
def test_negative_pre_len_raises(self):
|
|
with self.assertRaises(ValueError):
|
|
_commit("r", pre=-1, tokens=[1]).validate_committed_tokens()
|
|
|
|
def test_valid_commit_passes(self):
|
|
# Should not raise.
|
|
_commit("r", pre=0, tokens=[1, 2]).validate_committed_tokens()
|
|
|
|
|
|
class TestVerifierCommitSegment(CustomTestCase):
|
|
def test_append_coalesces_contiguous_commits(self):
|
|
seg = _segment("r", pre=0)
|
|
seg.append_message(_commit("r", pre=0, tokens=[10, 11]))
|
|
self.assertEqual(seg.committed_tokens, [10, 11])
|
|
self.assertEqual(seg.end_committed_len, 2)
|
|
seg.append_message(_commit("r", pre=2, tokens=[12]))
|
|
self.assertEqual(seg.committed_tokens, [10, 11, 12])
|
|
self.assertEqual(seg.end_committed_len, 3)
|
|
|
|
def test_append_wrong_request_raises(self):
|
|
seg = _segment("r", pre=0)
|
|
with self.assertRaises(RuntimeError):
|
|
seg.append_message(_commit("other", pre=0, tokens=[1]))
|
|
|
|
def test_append_wrong_drafter_rank_raises(self):
|
|
seg = _segment("r", pre=0, drafter_rank=0)
|
|
with self.assertRaises(RuntimeError):
|
|
seg.append_message(_commit("r", pre=0, tokens=[1], drafter_rank=7))
|
|
|
|
def test_append_non_contiguous_raises(self):
|
|
seg = _segment("r", pre=0)
|
|
seg.append_message(_commit("r", pre=0, tokens=[10])) # end -> 1
|
|
with self.assertRaises(RuntimeError):
|
|
seg.append_message(_commit("r", pre=5, tokens=[11])) # gap
|
|
|
|
def test_append_runs_message_validation(self):
|
|
seg = _segment("r", pre=0)
|
|
with self.assertRaises(ValueError):
|
|
seg.append_message(_commit("r", pre=0, tokens=[])) # empty -> validate
|
|
|
|
def test_extract_prefix_splits_segment(self):
|
|
seg = _segment("r", pre=0)
|
|
seg.append_message(_commit("r", pre=0, tokens=[10, 11, 12, 13]))
|
|
prefix = seg.extract_prefix(2)
|
|
self.assertEqual(prefix.committed_tokens, [10, 11])
|
|
self.assertEqual(prefix.pre_verify_committed_len, 0)
|
|
# Remainder stays in the original segment, with pre advanced by 2.
|
|
self.assertEqual(seg.committed_tokens, [12, 13])
|
|
self.assertEqual(seg.pre_verify_committed_len, 2)
|
|
self.assertEqual(seg.end_committed_len, 4)
|
|
|
|
def test_extract_prefix_bounds(self):
|
|
seg = _segment("r", pre=0)
|
|
seg.append_message(_commit("r", pre=0, tokens=[10, 11]))
|
|
with self.assertRaises(ValueError):
|
|
seg.extract_prefix(0)
|
|
with self.assertRaises(ValueError):
|
|
seg.extract_prefix(3) # exceeds segment length
|
|
|
|
|
|
class TestDraftControlInbox(CustomTestCase):
|
|
def _inbox(self):
|
|
from sglang.srt.speculative.decoupled_spec_io import DraftControlInbox
|
|
|
|
return DraftControlInbox()
|
|
|
|
def _sync(self, rid, drafter_rank=0):
|
|
return DraftSync(
|
|
request_id=rid, src_verifier_rank=0, dst_drafter_rank=drafter_rank
|
|
)
|
|
|
|
def _close(self, rid, drafter_rank=0):
|
|
return DraftClose(
|
|
request_id=rid,
|
|
src_verifier_rank=0,
|
|
dst_drafter_rank=drafter_rank,
|
|
reason="x",
|
|
)
|
|
|
|
def test_add_control_batch_routes_each_message_type(self):
|
|
inbox = self._inbox()
|
|
inbox.add_control_batch_locked(
|
|
DraftControlBatch(
|
|
dst_drafter_rank=0,
|
|
sync_messages=[self._sync("s")],
|
|
verify_commit_messages=[_commit("c", pre=0, tokens=[1])],
|
|
close_messages=[self._close("x")],
|
|
)
|
|
)
|
|
self.assertEqual([m.request_id for m in inbox.sync_messages], ["s"])
|
|
self.assertIn(DraftReqKey(0, "c"), inbox.verifier_commit_segments)
|
|
self.assertIn(DraftReqKey(0, "x"), inbox.close_keys)
|
|
|
|
def test_close_drops_pending_segment_and_sync(self):
|
|
inbox = self._inbox()
|
|
inbox.add_control_batch_locked(
|
|
DraftControlBatch(
|
|
dst_drafter_rank=0,
|
|
sync_messages=[self._sync("r")],
|
|
verify_commit_messages=[_commit("r", pre=0, tokens=[1])],
|
|
)
|
|
)
|
|
inbox.add_close_key_locked(DraftReqKey(0, "r"))
|
|
self.assertEqual(inbox.sync_messages, [])
|
|
self.assertNotIn(DraftReqKey(0, "r"), inbox.verifier_commit_segments)
|
|
self.assertIn(DraftReqKey(0, "r"), inbox.close_keys)
|
|
|
|
def test_verify_commit_for_closed_key_is_ignored(self):
|
|
inbox = self._inbox()
|
|
inbox.add_close_key_locked(DraftReqKey(0, "r"))
|
|
inbox.add_verify_commit_locked(_commit("r", pre=0, tokens=[1]))
|
|
self.assertNotIn(DraftReqKey(0, "r"), inbox.verifier_commit_segments)
|
|
|
|
def test_extract_ready_controls_full_consume(self):
|
|
inbox = self._inbox()
|
|
inbox.add_control_batch_locked(
|
|
DraftControlBatch(
|
|
dst_drafter_rank=0,
|
|
sync_messages=[self._sync("s")],
|
|
verify_commit_messages=[_commit("c", pre=0, tokens=[1, 2])],
|
|
close_messages=[self._close("x")],
|
|
)
|
|
)
|
|
ready = inbox.extract_ready_controls_locked(
|
|
lambda seg: len(seg.committed_tokens)
|
|
)
|
|
self.assertEqual([m.request_id for m in ready.sync_messages], ["s"])
|
|
self.assertEqual({k.request_id for k in ready.close_keys}, {"x"})
|
|
self.assertEqual(len(ready.ready_commit_segments), 1)
|
|
self.assertEqual(ready.ready_commit_segments[0].committed_tokens, [1, 2])
|
|
# Fully consumed -> the segment is gone; inbox drained.
|
|
self.assertTrue(inbox.is_empty())
|
|
|
|
def test_extract_ready_controls_zero_consumable_keeps_segment(self):
|
|
inbox = self._inbox()
|
|
inbox.add_verify_commit_locked(_commit("c", pre=0, tokens=[1, 2]))
|
|
ready = inbox.extract_ready_controls_locked(lambda seg: 0)
|
|
self.assertEqual(ready.ready_commit_segments, [])
|
|
# Segment is left buffered for a later step.
|
|
self.assertIn(DraftReqKey(0, "c"), inbox.verifier_commit_segments)
|
|
|
|
def test_extract_ready_controls_partial_consume_buffers_remainder(self):
|
|
inbox = self._inbox()
|
|
inbox.add_verify_commit_locked(_commit("c", pre=0, tokens=[1, 2, 3]))
|
|
ready = inbox.extract_ready_controls_locked(lambda seg: 1)
|
|
self.assertEqual(ready.ready_commit_segments[0].committed_tokens, [1])
|
|
# Remainder [2, 3] stays buffered with pre advanced to 1.
|
|
seg = inbox.verifier_commit_segments[DraftReqKey(0, "c")]
|
|
self.assertEqual(seg.committed_tokens, [2, 3])
|
|
self.assertEqual(seg.pre_verify_committed_len, 1)
|
|
|
|
def test_close_in_same_batch_drops_sync_and_commit(self):
|
|
# add_control_batch applies close first, so a same-key sync/commit in the
|
|
# same batch is dropped/ignored: close wins within one batch.
|
|
inbox = self._inbox()
|
|
inbox.add_control_batch_locked(
|
|
DraftControlBatch(
|
|
dst_drafter_rank=0,
|
|
sync_messages=[self._sync("r")],
|
|
verify_commit_messages=[_commit("r", pre=0, tokens=[1])],
|
|
close_messages=[self._close("r")],
|
|
)
|
|
)
|
|
self.assertEqual(inbox.sync_messages, [])
|
|
self.assertNotIn(DraftReqKey(0, "r"), inbox.verifier_commit_segments)
|
|
self.assertEqual({k.request_id for k in inbox.close_keys}, {"r"})
|
|
|
|
def test_two_commits_same_key_coalesce_in_inbox(self):
|
|
# A second commit for an existing key appends to the buffered segment.
|
|
inbox = self._inbox()
|
|
inbox.add_verify_commit_locked(_commit("r", pre=0, tokens=[1, 2]))
|
|
inbox.add_verify_commit_locked(_commit("r", pre=2, tokens=[3]))
|
|
seg = inbox.verifier_commit_segments[DraftReqKey(0, "r")]
|
|
self.assertEqual(seg.committed_tokens, [1, 2, 3])
|
|
self.assertEqual(seg.end_committed_len, 3)
|
|
|
|
|
|
class TestDraftMeshMessageEnvelope(CustomTestCase):
|
|
def test_from_control_batch_sets_discriminant_and_slot(self):
|
|
batch = DraftControlBatch(dst_drafter_rank=0)
|
|
msg = DraftMeshMessage.from_control_batch(batch)
|
|
self.assertEqual(msg.message_type, DraftMeshMessageType.CONTROL_BATCH)
|
|
self.assertIs(msg.control_batch, batch)
|
|
self.assertIsNone(msg.tail_stream_output_batch)
|
|
|
|
def test_from_tail_stream_output_batch_sets_discriminant_and_slot(self):
|
|
from sglang.srt.speculative.decoupled_spec_io import DraftTailStreamOutputBatch
|
|
|
|
batch = DraftTailStreamOutputBatch()
|
|
msg = DraftMeshMessage.from_tail_stream_output_batch(batch)
|
|
self.assertEqual(
|
|
msg.message_type, DraftMeshMessageType.TAIL_STREAM_OUTPUT_BATCH
|
|
)
|
|
self.assertIs(msg.tail_stream_output_batch, batch)
|
|
self.assertIsNone(msg.control_batch)
|
|
|
|
|
|
if __name__ == "__main__":
|
|
unittest.main(verbosity=3)
|