chore: update CI test est_time values (#38238)

Co-authored-by: sglang-bot <sglang-bot@users.noreply.github.com>

test/registered/unit/spec/test_adaptive_runtime_state.py

@@ -6,7 +6,7 @@ from sglang.srt.speculative.adaptive_runtime_state import (
 )
 from sglang.test.ci.ci_register import register_cpu_ci
 
-register_cpu_ci(est_time=1, suite="base-a-test-cpu")
+register_cpu_ci(est_time=4, suite="base-a-test-cpu")
 
 
 def _make_state(steps: int) -> SpecRuntimeState:

test/registered/unit/spec/test_decode_bookkeeping_ownership.py

@@ -20,7 +20,7 @@ from pathlib import Path
 from sglang.test.ci.ci_register import register_cpu_ci
 from sglang.test.test_utils import CustomTestCase
 
-register_cpu_ci(est_time=8, suite="base-a-test-cpu")
+register_cpu_ci(est_time=22, suite="base-a-test-cpu")
 
 _REPO_ROOT = Path(__file__).resolve().parents[4]
 _SRT_DIR = _REPO_ROOT / "python" / "sglang" / "srt"

test/registered/unit/spec/test_decoupled_spec_io.py

@@ -24,7 +24,7 @@ from sglang.srt.speculative.decoupled_spec_io import (
 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")
+register_cpu_ci(est_time=10, suite="base-a-test-cpu")
 
 
 def _commit(rid, *, pre, tokens, src_verifier_rank=0, drafter_rank=0) -> VerifyCommit:

test/registered/unit/spec/test_dflash_extra_buffer_lazy.py

@@ -13,7 +13,7 @@ from sglang.srt.arg_groups.mamba_hook import validate_mamba_extra_buffer
 from sglang.test.ci.ci_register import register_cuda_ci
 from sglang.test.test_utils import CustomTestCase
 
-register_cuda_ci(est_time=10, stage="base-b", runner_config="1-gpu-small")
+register_cuda_ci(est_time=9, stage="base-b", runner_config="1-gpu-small")
 
 from sglang.srt.model_executor.forward_batch_info import ForwardMode
 from sglang.srt.runtime_context import override_platform

test/registered/unit/spec/test_dflash_logits.py

@@ -12,7 +12,7 @@ from sglang.srt.models.dflash import (
 from sglang.srt.speculative.dflash_utils import parse_dflash_draft_config
 from sglang.test.ci.ci_register import register_cpu_ci
 
-register_cpu_ci(est_time=35, suite="base-a-test-cpu")
+register_cpu_ci(est_time=38, suite="base-a-test-cpu")
 
 
 def test_dflash_unary_logit_transform():

test/registered/unit/spec/test_dflash_overlap_hostsync.py

@@ -12,7 +12,7 @@ from sglang.srt.runtime_context import get_context
 from sglang.test.ci.ci_register import register_cuda_ci
 from sglang.test.test_utils import CustomTestCase
 
-register_cuda_ci(est_time=30, stage="base-b", runner_config="1-gpu-small")
+register_cuda_ci(est_time=13, stage="base-b", runner_config="1-gpu-small")
 
 _HAS_CUDA = torch.cuda.is_available()
 

test/registered/unit/spec/test_draft_construction_isolation.py

@@ -23,7 +23,7 @@ from sglang.srt.runtime_context import get_context, get_flags, get_model, publis
 from sglang.test.ci.ci_register import register_cpu_ci
 from sglang.test.test_utils import CustomTestCase
 
-register_cpu_ci(est_time=5, suite="base-a-test-cpu")
+register_cpu_ci(est_time=12, suite="base-a-test-cpu")
 
 
 class _AlwaysDisables:

test/registered/unit/spec/test_draft_per_runner_config.py

@@ -26,7 +26,7 @@ from sglang.srt.runtime_context import get_context, get_model
 from sglang.test.ci.ci_register import register_cpu_ci
 from sglang.test.test_utils import CustomTestCase
 
-register_cpu_ci(est_time=5, suite="base-a-test-cpu")
+register_cpu_ci(est_time=12, suite="base-a-test-cpu")
 
 
 class _StopConstruction(Exception):

test/registered/unit/spec/test_dspark_target_hidden_projection.py

@@ -12,7 +12,7 @@ from sglang.srt.speculative.dspark_components.dspark_kv_inject import (
 from sglang.test.ci.ci_register import register_cpu_ci
 from sglang.test.test_utils import CustomTestCase
 
-register_cpu_ci(est_time=1, suite="base-a-test-cpu")
+register_cpu_ci(est_time=8, suite="base-a-test-cpu")
 
 
 class _Attention:

test/registered/unit/spec/test_eagle_draft_cuda_graph_runner.py

@@ -25,7 +25,7 @@ from sglang.srt.speculative.eagle_draft_cuda_graph_runner import (
 from sglang.test.ci.ci_register import register_cpu_ci
 from sglang.test.test_utils import CustomTestCase
 
-register_cpu_ci(est_time=5, suite="base-a-test-cpu")
+register_cpu_ci(est_time=10, suite="base-a-test-cpu")
 
 CAPTURE_BS = 4
 SEQ_LEN_FILL_VALUE = 1

test/registered/unit/spec/test_eagle_draft_extend_logits.py

@@ -10,7 +10,7 @@ from sglang.srt.model_executor.forward_batch_info import (
 )
 from sglang.test.ci.ci_register import register_cpu_ci
 
-register_cpu_ci(est_time=1, suite="base-a-test-cpu")
+register_cpu_ci(est_time=7, suite="base-a-test-cpu")
 
 
 class TestEagleDraftExtendLogitsPruning(unittest.TestCase):

test/registered/unit/spec/test_eagle_seeded_coins.py

@@ -21,7 +21,7 @@ from sglang.srt.speculative.eagle_utils import _seeded_verify_coins, _verify_coi
 from sglang.test.ci.ci_register import register_cuda_ci
 from sglang.test.test_utils import CustomTestCase
 
-register_cuda_ci(est_time=20, stage="base-b", runner_config="1-gpu-small")
+register_cuda_ci(est_time=10, stage="base-b", runner_config="1-gpu-small")
 
 DRAFT_TOKEN_NUM = 4
 

test/registered/unit/spec/test_eagle_worker_v2_topk1_fastpath.py

@@ -24,7 +24,7 @@ from sglang.test.test_utils import CustomTestCase
 register_amd_ci(est_time=20, stage="stage-b", runner_config="1-gpu-small-amd")
 
 
-register_cpu_ci(est_time=20, suite="base-a-test-cpu")
+register_cpu_ci(est_time=12, suite="base-a-test-cpu")
 
 DEVICE = "cuda" if torch.cuda.is_available() else "cpu"
 

test/registered/unit/spec/test_fast_prefill_plan.py

@@ -27,7 +27,7 @@ try:
 except ImportError:
     _HAS_FLASHINFER = False
 
-register_cuda_ci(est_time=30, stage="base-b", runner_config="1-gpu-small")
+register_cuda_ci(est_time=10, stage="base-b", runner_config="1-gpu-small")
 
 # Draft-extend layout: constant qo (num_tokens_per_req per req), page_size 1.
 # Non-uniform seq_lens make cumsum non-trivial, so a wrong per-row kv split is

test/registered/unit/spec/test_multi_layer_eagle_shared_read_event.py

@@ -14,7 +14,7 @@ from sglang.srt.speculative.multi_layer_eagle_worker_v2 import (
 from sglang.test.ci.ci_register import register_cpu_ci
 from sglang.test.test_utils import CustomTestCase
 
-register_cpu_ci(est_time=5, suite="base-a-test-cpu")
+register_cpu_ci(est_time=8, suite="base-a-test-cpu")
 
 RAW_BS = 2
 RAW_NUM_TOKENS = 3

test/registered/unit/spec/test_ngram_corpus.py

@@ -13,7 +13,7 @@ from sglang.srt.speculative.cpp_ngram.ngram_corpus import NgramCorpus
 from sglang.test.ci.ci_register import register_cpu_ci
 from sglang.test.test_utils import CustomTestCase
 
-register_cpu_ci(est_time=26, suite="base-a-test-cpu")
+register_cpu_ci(est_time=20, suite="base-a-test-cpu")
 
 
 def _make_corpus(match_type="BFS", **kwargs):

test/registered/unit/spec/test_ngram_mamba_verify_update.py

@@ -7,7 +7,7 @@ import sglang.srt
 from sglang.test.ci.ci_register import register_cpu_ci
 from sglang.test.test_utils import CustomTestCase
 
-register_cpu_ci(est_time=5, suite="base-a-test-cpu")
+register_cpu_ci(est_time=23, suite="base-a-test-cpu")
 
 
 class TestNgramLastCorrectStepIndices(CustomTestCase):

test/registered/unit/spec/test_plugin_hook_signatures.py

@@ -25,7 +25,7 @@ from sglang.srt.speculative.spec_registry import CustomSpecAlgo
 from sglang.test.ci.ci_register import register_cpu_ci
 from sglang.test.test_utils import CustomTestCase
 
-register_cpu_ci(est_time=5, suite="base-a-test-cpu")
+register_cpu_ci(est_time=10, suite="base-a-test-cpu")
 
 
 def _dispatched_methods():

test/registered/unit/spec/test_resolve_swa_kv_pool.py

@@ -11,7 +11,7 @@ from sglang.srt.speculative.spec_info import SpeculativeAlgorithm
 from sglang.test.ci.ci_register import register_amd_ci, register_cuda_ci
 from sglang.test.test_utils import CustomTestCase
 
-register_cuda_ci(est_time=5, stage="base-b", runner_config="1-gpu-large")
+register_cuda_ci(est_time=9, stage="base-b", runner_config="1-gpu-large")
 register_amd_ci(est_time=8, suite="stage-b-test-1-gpu-large-amd")
 
 _RESOLVERS = (

test/registered/unit/spec/test_spec_cpu_overlap_constraint.py

@@ -7,7 +7,7 @@ from sglang.srt.server_args import ServerArgs
 from sglang.test.ci.ci_register import register_cpu_ci
 from sglang.test.test_utils import CustomTestCase
 
-register_cpu_ci(est_time=20, suite="base-a-test-cpu")
+register_cpu_ci(est_time=10, suite="base-a-test-cpu")
 
 
 def _make_spec_args(device: str, algorithm: str = "EAGLE", **overrides) -> ServerArgs:

test/registered/unit/spec/test_spec_registry.py

@@ -16,7 +16,7 @@ from sglang.srt.speculative.spec_registry import (
 from sglang.test.ci.ci_register import register_cpu_ci
 from sglang.test.test_utils import CustomTestCase
 
-register_cpu_ci(est_time=5, suite="base-a-test-cpu")
+register_cpu_ci(est_time=11, suite="base-a-test-cpu")
 
 
 class _RegistryIsolated(CustomTestCase):

test/registered/unit/spec/test_spec_utils_traverse_tree.py

@@ -14,7 +14,7 @@ import torch
 from sglang.srt.speculative.spec_utils import GrammarTree, traverse_tree
 from sglang.test.ci.ci_register import register_cpu_ci
 
-register_cpu_ci(est_time=4, suite="base-a-test-cpu")
+register_cpu_ci(est_time=10, suite="base-a-test-cpu")
 
 
 class TestTraverseTreePassesIntsToGrammar(unittest.TestCase):

test/registered/unit/spec/test_suffix_attention_merge_dispatch.py

@@ -11,7 +11,7 @@ from sglang.kernels.ops.attention.suffix_attention_merge import (
 from sglang.test.ci.ci_register import register_cpu_ci
 from sglang.test.test_utils import CustomTestCase
 
-register_cpu_ci(est_time=1, suite="base-a-test-cpu")
+register_cpu_ci(est_time=8, suite="base-a-test-cpu")
 
 
 class TestSuffixAttentionMergeDispatch(CustomTestCase):

test/registered/unit/spec/test_uno_request_validation.py

@@ -7,7 +7,7 @@ from sglang.srt.speculative.uno_validation import validate_uno_request
 from sglang.test.ci.ci_register import register_cpu_ci
 from sglang.test.test_utils import CustomTestCase
 
-register_cpu_ci(est_time=1, suite="base-a-test-cpu")
+register_cpu_ci(est_time=7, suite="base-a-test-cpu")
 
 
 def _make_request(**overrides):

test/registered/unit/spec/test_uno_tree_config.py

@@ -8,7 +8,7 @@ from sglang.srt.arg_groups.speculative_hook import _handle_uno
 from sglang.test.ci.ci_register import register_cpu_ci
 from sglang.test.test_utils import CustomTestCase
 
-register_cpu_ci(est_time=1, suite="base-a-test-cpu")
+register_cpu_ci(est_time=8, suite="base-a-test-cpu")
 
 
 class TestUnoTreeConfig(CustomTestCase):

test/registered/unit/spec/test_uno_tree_sparse_sampling.py

@@ -13,7 +13,7 @@ from sglang.srt.speculative.uno_utils import sample_uno_tree_target_tokens
 from sglang.test.ci.ci_register import register_cpu_ci
 from sglang.test.test_utils import CustomTestCase
 
-register_cpu_ci(est_time=1, suite="base-a-test-cpu")
+register_cpu_ci(est_time=8, suite="base-a-test-cpu")
 
 
 class TestUnoTreeSparseSampling(CustomTestCase):