Register CPU CI for 17 e2e tests and partition xeon base-c suite (#35227)

Co-authored-by: Zhang, Mingxu <mingxu.zhang@intel.com> Co-authored-by: MingxuZh <109504044+MingxuZh@users.noreply.github.com> Co-authored-by: Copilot Autofix powered by AI <175728472+Copilot@users.noreply.github.com> Co-authored-by: Claude <noreply@anthropic.com>
2026-08-24 10:49:38 +08:00
parent 11b1b4c374
commit 514b997e6c
128 changed files with 312 additions and 162 deletions
@@ -83,11 +83,35 @@ jobs:
    runs-on: ${{ matrix.runner }}

    strategy:
+      fail-fast: false
      matrix:
        include:
+          # GNR runs only the TP suite (needs >=6 NUMA nodes for --tp 6),
+          # full machine, no partitioning. 4 files: 3 TP + autoround.
          - runner: xeon-gnr
-            suite: base-b-test-cpu
-            partition_args: ""
+            role: gnr
+            suite: "base-b-tp-test-cpu"
+            name: gnr
+
+          # Each SPR box (2 sockets / 2 NUMA, 32 cores each) is ONE job that
+          # runs two socket-pinned containers in parallel, each taking one
+          # partition of a 4-way split over the merged non-TP pool
+          # (base-b non-TP + base-c). Building once per job keeps both
+          # containers on the same locally-built image.
+          - runner: xeon-spr
+            role: spr
+            suite: "base-b-test-cpu,base-c-test-cpu"
+            name: spr1
+            part_size: 4
+            part_lo: 0        # socket0
+            part_hi: 1        # socket1
+          - runner: xeon-spr
+            role: spr
+            suite: "base-b-test-cpu,base-c-test-cpu"
+            name: spr2
+            part_size: 4
+            part_lo: 2        # socket0
+            part_hi: 3        # socket1
    steps:
      - name: Cleanup workspace
        run: |
@@ -100,10 +124,13 @@ jobs:
          clean: false
          ref: ${{ inputs.ref || github.ref }}

-      - name: Build and Push
+      - name: Build image
        run: |
-          version=$(cat python/sglang/version.py | cut -d'"' -f2)
-          tag=v${version}-xeon
+          # Fixed tag (no SHA): a new build overwrites the previous image so a
+          # self-hosted box does not accumulate multi-GB images across PRs.
+          # --no-cache keeps the CPU image clean (past sgl-eval/accelerate gaps
+          # came from stale layers). One build per job; on SPR both socket
+          # containers reuse this locally-built image.
          PR_REPO=${{ github.event.pull_request.head.repo.clone_url }}
          PR_HEAD_REF=${{ github.head_ref }}

@@ -112,37 +139,89 @@ jobs:
            ${PR_HEAD_REF:+--build-arg VER_SGLANG=$PR_HEAD_REF} \
            . -f docker/xeon.Dockerfile  -t sglang_xeon --no-cache

-      - name: Run container
+      # ==================== GNR: full machine, TP suite ==================== #
+      - name: Run container (GNR)
+        if: matrix.role == 'gnr'
        run: |
+          docker rm -f ci_sglang_${{ matrix.name }} || true
          docker run -dt \
            -v ${{ github.workspace }}:/sglang-checkout/ --ipc=host \
            -v $HOME/.cache/huggingface:/root/.cache/huggingface \
            -e HF_TOKEN="$(cat ~/huggingface_token.txt)" \
-            --name ci_sglang_${{ matrix.runner }} \
+            --name ci_sglang_${{ matrix.name }} \
            sglang_xeon

-      - name: Check AMX support
-        id: check_amx
+      - name: Check AMX support (GNR)
+        if: matrix.role == 'gnr'
        timeout-minutes: 5
        run: |
-          docker exec -w /sglang-checkout/ ci_sglang_${{ matrix.runner }} \
+          docker exec -w /sglang-checkout/ ci_sglang_${{ matrix.name }} \
            bash -c "source /opt/.venv/bin/activate && python3 -c 'import torch; import sgl_kernel; assert torch.cpu._is_amx_tile_supported(); assert hasattr(torch.ops.sgl_kernel, \"convert_weight_packed\"); '"

-      - name: Run unit tests
+      - name: Run unit tests (GNR)
+        if: matrix.role == 'gnr'
        timeout-minutes: 120
        run: |
-          docker exec -w /sglang-checkout/ ci_sglang_${{ matrix.runner }} \
-            bash -c "source /opt/.venv/bin/activate && cd ./test && python3 run_suite.py --hw cpu --suite ${{ matrix.suite }} ${{ matrix.partition_args }}"
+          docker exec -w /sglang-checkout/ ci_sglang_${{ matrix.name }} \
+            bash -c "source /opt/.venv/bin/activate && cd ./test && python3 run_suite.py --hw cpu --suite ${{ matrix.suite }} --timeout-from-est-time"

+      # ============ SPR: two socket-pinned containers in parallel ========== #
+      - name: Run unit tests (SPR, two sockets in parallel)
+        if: matrix.role == 'spr'
+        timeout-minutes: 120
+        run: |
+          set -uo pipefail
+          HF_TOKEN_VAL="$(cat ~/huggingface_token.txt)"
+
+          # $1 socket id, $2 cpuset-cpus, $3 cpuset-mems, $4 partition id
+          run_socket () {
+            local sid=$1 cpus=$2 mems=$3 pid=$4
+            local cname=ci_sglang_${{ matrix.name }}_s${sid}
+            docker rm -f "$cname" >/dev/null 2>&1 || true
+            docker run -dt \
+              --cpuset-cpus="$cpus" --cpuset-mems="$mems" \
+              -v ${{ github.workspace }}:/sglang-checkout/ --ipc=host \
+              -v $HOME/.cache/huggingface:/root/.cache/huggingface \
+              -e HF_TOKEN="$HF_TOKEN_VAL" \
+              -e SGLANG_CPU_OMP_THREADS_BIND="$cpus" \
+              --name "$cname" \
+              sglang_xeon
+            # AMX sanity on this socket, then the partition it owns.
+            docker exec -w /sglang-checkout/ "$cname" bash -c \
+              "source /opt/.venv/bin/activate && python3 -c 'import torch; import sgl_kernel; assert torch.cpu._is_amx_tile_supported(); assert hasattr(torch.ops.sgl_kernel, \"convert_weight_packed\")' && \
+               cd ./test && python3 run_suite.py --hw cpu --suite ${{ matrix.suite }} --auto-partition-id $pid --auto-partition-size ${{ matrix.part_size }} --timeout-from-est-time"
+          }
+
+          # socket0 -> part_lo, socket1 -> part_hi; run concurrently, wait both.
+          # Runners have hyper-threading OFF: 64 physical cores = CPUs 0-63,
+          # NUMA0=0-31, NUMA1=32-63 (no sibling range).
+          run_socket 0 "0-31"  0 ${{ matrix.part_lo }} > s0.log 2>&1 &
+          P0=$!
+          run_socket 1 "32-63" 1 ${{ matrix.part_hi }} > s1.log 2>&1 &
+          P1=$!
+
+          rc=0
+          wait $P0 || rc=1
+          wait $P1 || rc=1
+
+          echo "==================== socket0 (partition ${{ matrix.part_lo }}) ===================="
+          cat s0.log || true
+          echo "==================== socket1 (partition ${{ matrix.part_hi }}) ===================="
+          cat s1.log || true
+          exit $rc
+
+      # ============================ Common ================================= #
      - name: Change permission
+        if: always()
        timeout-minutes: 2
        run: |
-          docker exec -u root ci_sglang_${{ matrix.runner }} bash -c "
-            rm -rf /tmp/ci-home  &&
-            chown -R  $(id -u):$(id -g) /sglang-checkout/ 2>/dev/null || true
-          "
+          for c in ci_sglang_${{ matrix.name }} ci_sglang_${{ matrix.name }}_s0 ci_sglang_${{ matrix.name }}_s1; do
+            docker exec -u root "$c" bash -c "rm -rf /tmp/ci-home && chown -R $(id -u):$(id -g) /sglang-checkout/ 2>/dev/null || true" 2>/dev/null || true
+          done

      - name: Cleanup container
        if: always()
        run: |
-          docker rm -f ci_sglang_${{ matrix.runner }} || true
+          for c in ci_sglang_${{ matrix.name }} ci_sglang_${{ matrix.name }}_s0 ci_sglang_${{ matrix.name }}_s1; do
+            docker rm -f "$c" || true
+          done
@@ -39,7 +39,7 @@ RUN source /opt/.venv/bin/activate && \
    git checkout ${VER_SGLANG} && \
    cd python && \
    cp pyproject_cpu.toml pyproject.toml && \
-    uv pip install . && \
+    uv pip install ".[diffusion]" && \
    cd sglang/kernels/aot && \
    cp pyproject_cpu.toml pyproject.toml && \
    uv pip install . && \
@@ -60,7 +60,7 @@ from sglang.test.ci.ci_register import register_cpu_ci
 from sglang.test.test_utils import CustomTestCase

 register_cpu_ci(est_time=40, suite="base-a-test-cpu")
-register_cpu_ci(est_time=7, suite="base-c-test-cpu")
+register_cpu_ci(est_time=46, suite="base-c-test-cpu")


 class _DummyTokenTensor:
@@ -24,7 +24,7 @@ from sglang.test.test_utils import (
    popen_launch_server,
 )

-register_cpu_ci(est_time=330, suite="base-b-test-cpu")
+register_cpu_ci(est_time=183, suite="base-b-tp-test-cpu")


 class TestAutoRoundCPUConfig(CustomTestCase):
@@ -8,7 +8,7 @@ from sglang.srt.server_args import ServerArgs, set_global_server_args_for_schedu
 from sglang.test.ci.ci_register import register_cpu_ci
 from sglang.test.cpu_test_utils import GeluAndMul, SiluAndMul, precision

-register_cpu_ci(est_time=10, suite="base-b-test-cpu")
+register_cpu_ci(est_time=9, suite="base-b-test-cpu")
 register_cpu_ci(est_time=10, suite="base-b-test-cpu-arm64")

 torch.manual_seed(1234)
@@ -8,7 +8,7 @@ kernel = torch.ops.sgl_kernel
 from sglang.test.ci.ci_register import register_cpu_ci
 from sglang.test.test_utils import CustomTestCase

-register_cpu_ci(est_time=10, suite="base-b-test-cpu")
+register_cpu_ci(est_time=6, suite="base-b-tp-test-cpu")


 class TestBinding(CustomTestCase):
@@ -11,7 +11,7 @@ from sglang.test.ci.ci_register import register_cpu_ci
 from sglang.test.cpu_test_utils import precision
 from sglang.test.test_utils import CustomTestCase

-register_cpu_ci(est_time=10, suite="base-b-test-cpu")
+register_cpu_ci(est_time=13, suite="base-b-test-cpu")

 torch.manual_seed(1234)

@@ -9,7 +9,7 @@ from sglang.test.ci.ci_register import register_cpu_ci
 from sglang.test.cpu_test_utils import parametrize, precision
 from sglang.test.test_utils import CustomTestCase

-register_cpu_ci(est_time=10, suite="base-b-test-cpu")
+register_cpu_ci(est_time=7, suite="base-b-test-cpu")

 causal_conv1d_weight_pack = torch.ops.sgl_kernel.causal_conv1d_weight_pack
 causal_conv1d_fwd = torch.ops.sgl_kernel.causal_conv1d_fwd_cpu
@@ -13,7 +13,7 @@ import torch.multiprocessing as mp
 from sglang.test.ci.ci_register import register_cpu_ci
 from sglang.test.test_utils import CustomTestCase, find_available_port

-register_cpu_ci(est_time=30, suite="base-b-test-cpu")
+register_cpu_ci(est_time=43, suite="base-b-test-cpu")


 def run_distributed_test(rank, world_size, master_port, output_writer, fn):
@@ -21,7 +21,7 @@ from sglang.test.test_utils import (
    popen_launch_server,
 )

-register_cpu_ci(est_time=10, suite="base-b-test-cpu")
+register_cpu_ci(est_time=315, suite="base-b-tp-test-cpu")


 class TestCPUGraph(CustomTestCase):
@@ -7,7 +7,7 @@ from torch.nn.functional import scaled_dot_product_attention
 from sglang.test.ci.ci_register import register_cpu_ci
 from sglang.test.test_utils import CustomTestCase

-register_cpu_ci(est_time=10, suite="base-b-test-cpu")
+register_cpu_ci(est_time=7, suite="base-b-test-cpu")

 torch.manual_seed(1234)

@@ -8,7 +8,7 @@ from sglang.test.ci.ci_register import register_cpu_ci
 from sglang.test.cpu_test_utils import parametrize, precision
 from sglang.test.test_utils import CustomTestCase

-register_cpu_ci(est_time=10, suite="base-b-test-cpu")
+register_cpu_ci(est_time=54, suite="base-b-test-cpu")

 flash_attn_varlen_func = torch.ops.sgl_kernel.flash_attn_varlen_func

@@ -18,7 +18,7 @@ from sglang.test.cpu_test_utils import (
 )
 from sglang.test.test_utils import CustomTestCase

-register_cpu_ci(est_time=10, suite="base-b-test-cpu")
+register_cpu_ci(est_time=9, suite="base-b-test-cpu")

 torch.manual_seed(1234)

@@ -21,7 +21,7 @@ from sglang.test.test_utils import (
    popen_launch_server,
 )

-register_cpu_ci(est_time=10, suite="base-b-test-cpu")
+register_cpu_ci(est_time=685, suite="base-b-tp-test-cpu")


 class TestIntelAMXAttnBackend(CustomTestCase):
@@ -14,7 +14,7 @@ from sglang.test.test_utils import (
    intel_amx_benchmark,
 )

-register_cpu_ci(est_time=10, suite="base-b-test-cpu")
+register_cpu_ci(est_time=47, suite="base-b-test-cpu")


 class TestIntelAMXAttnBackendQuant(CustomTestCase):
@@ -14,7 +14,7 @@ from sglang.test.test_utils import (
    intel_amx_benchmark,
 )

-register_cpu_ci(est_time=10, suite="base-b-test-cpu")
+register_cpu_ci(est_time=477, suite="base-b-tp-test-cpu")


 class TestIntelAMXAttnBackendQuant(CustomTestCase):
@@ -9,7 +9,7 @@ from torch.nn.functional import softplus
 from sglang.test.ci.ci_register import register_cpu_ci
 from sglang.test.cpu_test_utils import precision

-register_cpu_ci(est_time=10, suite="base-b-test-cpu")
+register_cpu_ci(est_time=7, suite="base-b-test-cpu")

 torch.manual_seed(1234)

@@ -8,7 +8,7 @@ from sglang.test.ci.ci_register import register_cpu_ci
 from sglang.test.cpu_test_utils import precision
 from sglang.test.test_utils import CustomTestCase

-register_cpu_ci(est_time=10, suite="base-b-test-cpu")
+register_cpu_ci(est_time=8, suite="base-b-test-cpu")

 torch.manual_seed(1234)

@@ -34,7 +34,7 @@ from sglang.test.cpu_test_utils import (
    unpack_and_dequant_awq,
 )

-register_cpu_ci(est_time=10, suite="base-b-test-cpu")
+register_cpu_ci(est_time=7, suite="base-b-test-cpu")


 def run_fused_experts(
@@ -8,7 +8,7 @@ import torch
 from sglang.test.ci.ci_register import register_cpu_ci
 from sglang.test.cpu_test_utils import make_non_contiguous, precision

-register_cpu_ci(est_time=10, suite="base-b-test-cpu")
+register_cpu_ci(est_time=5, suite="base-b-test-cpu")
 register_cpu_ci(est_time=10, suite="base-b-test-cpu-arm64")

 torch.manual_seed(1234)
@@ -14,7 +14,7 @@ from sglang.test.cpu_test_utils import (
 )
 from sglang.test.test_utils import CustomTestCase

-register_cpu_ci(est_time=10, suite="base-b-test-cpu")
+register_cpu_ci(est_time=7, suite="base-b-test-cpu")

 convert_weight_packed = torch.ops.sgl_kernel.convert_weight_packed
 qkv_proj_with_rope = torch.ops.sgl_kernel.qkv_proj_with_rope
@@ -8,7 +8,7 @@ from sglang.srt.utils import is_host_cpu_arm64
 from sglang.test.ci.ci_register import register_cpu_ci
 from sglang.test.cpu_test_utils import precision

-register_cpu_ci(est_time=10, suite="base-b-test-cpu")
+register_cpu_ci(est_time=5, suite="base-b-test-cpu")
 register_cpu_ci(est_time=10, suite="base-b-test-cpu-arm64")

 torch.manual_seed(1234)
@@ -25,7 +25,7 @@ from sglang.srt.utils.rank_consensus_checker import (
 from sglang.test.ci.ci_register import register_cpu_ci
 from sglang.test.test_utils import CustomTestCase, find_available_port

-register_cpu_ci(est_time=30, suite="base-b-test-cpu")
+register_cpu_ci(est_time=193, suite="base-b-test-cpu")


 def run_distributed_test(
@@ -8,7 +8,7 @@ from sglang.srt.entrypoints.http_request_decompression import (
 )
 from sglang.test.ci.ci_register import register_cpu_ci

-register_cpu_ci(est_time=10, suite="base-b-test-cpu")
+register_cpu_ci(est_time=5, suite="base-b-test-cpu")

 PAYLOAD = b'{"text":"hello world","n":7}'
 COMPRESSED = zstandard.ZstdCompressor().compress(PAYLOAD)
@@ -7,7 +7,7 @@ from starlette.datastructures import Headers
 from sglang.srt.entrypoints.request_headers import apply_header_overrides
 from sglang.test.ci.ci_register import register_cpu_ci

-register_cpu_ci(est_time=10, suite="base-b-test-cpu")
+register_cpu_ci(est_time=4, suite="base-b-test-cpu")


 def _obj():
@@ -17,7 +17,7 @@ from sglang.test.ci.ci_register import register_cpu_ci
 from sglang.test.cpu_test_utils import precision
 from sglang.test.test_utils import CustomTestCase

-register_cpu_ci(est_time=10, suite="base-b-test-cpu")
+register_cpu_ci(est_time=33, suite="base-b-test-cpu")
 register_cpu_ci(est_time=10, suite="base-b-test-cpu-arm64")

 torch.manual_seed(1234)
@@ -7,7 +7,7 @@ from unittest.mock import patch
 from sglang.srt.server_args import ServerArgs
 from sglang.test.ci.ci_register import register_cpu_ci

-register_cpu_ci(est_time=10, suite="base-b-test-cpu")
+register_cpu_ci(est_time=6, suite="base-b-test-cpu")
 register_cpu_ci(est_time=10, suite="base-b-test-cpu-arm64")


@@ -20,7 +20,7 @@ from sglang.test.cpu_test_utils import (
 )
 from sglang.test.test_utils import CustomTestCase

-register_cpu_ci(est_time=10, suite="base-b-test-cpu")
+register_cpu_ci(est_time=8, suite="base-b-test-cpu")

 torch.manual_seed(1234)

@@ -9,7 +9,7 @@ from sglang.test.ci.ci_register import register_cpu_ci
 from sglang.test.cpu_test_utils import precision
 from sglang.test.test_utils import CustomTestCase

-register_cpu_ci(est_time=20, suite="base-b-test-cpu")
+register_cpu_ci(est_time=6, suite="base-b-test-cpu")


 def _topk1_chain_inputs(bs, num_steps):
@@ -6,7 +6,7 @@ import torch

 from sglang.test.ci.ci_register import register_cpu_ci

-register_cpu_ci(est_time=25, suite="base-b-test-cpu")
+register_cpu_ci(est_time=14, suite="base-b-test-cpu")

 torch.manual_seed(42)

@@ -13,7 +13,7 @@ from sglang.multimodal_gen.runtime.layers.attention.backends.subblock_sparse_att
 from sglang.test.ci.ci_register import register_cpu_ci
 from sglang.test.test_utils import CustomTestCase

-register_cpu_ci(est_time=10, suite="base-b-test-cpu")
+register_cpu_ci(est_time=9, suite="base-b-test-cpu")


 class TestSubBlockSparseAttentionDispatch(CustomTestCase):
@@ -15,7 +15,7 @@ from sglang.srt.models.llama4 import Llama4MoE
 from sglang.test.ci.ci_register import register_cpu_ci
 from sglang.test.test_utils import CustomTestCase

-register_cpu_ci(est_time=10, suite="base-b-test-cpu")
+register_cpu_ci(est_time=30, suite="base-b-test-cpu")
 register_cpu_ci(est_time=10, suite="base-b-test-cpu-arm64")


@@ -34,7 +34,7 @@ from sglang.srt.debug_utils.comparator.utils import Pair
 from sglang.test.ci.ci_register import register_cpu_ci

 register_cpu_ci(est_time=15, suite="base-a-test-cpu", nightly=True)
-register_cpu_ci(est_time=1, suite="base-c-test-cpu")
+register_cpu_ci(est_time=7, suite="base-c-test-cpu")


 class TestExecuteSubPlans:
@@ -26,7 +26,7 @@ from sglang.srt.debug_utils.comparator.utils import Pair
 from sglang.test.ci.ci_register import register_cpu_ci

 register_cpu_ci(est_time=15, suite="base-a-test-cpu", nightly=True)
-register_cpu_ci(est_time=1, suite="base-c-test-cpu")
+register_cpu_ci(est_time=8, suite="base-c-test-cpu")


 def _make_meta(
@@ -27,7 +27,7 @@ from sglang.srt.debug_utils.comparator.dims_spec import (
 from sglang.test.ci.ci_register import register_cpu_ci

 register_cpu_ci(est_time=10, suite="base-a-test-cpu", nightly=True)
-register_cpu_ci(est_time=1, suite="base-c-test-cpu")
+register_cpu_ci(est_time=6, suite="base-c-test-cpu")


 def _zigzag_order(cp_size: int) -> list[int]:
@@ -27,7 +27,7 @@ from sglang.srt.debug_utils.comparator.dims_spec import (
 from sglang.test.ci.ci_register import register_cpu_ci

 register_cpu_ci(est_time=10, suite="base-a-test-cpu", nightly=True)
-register_cpu_ci(est_time=1, suite="base-c-test-cpu")
+register_cpu_ci(est_time=7, suite="base-c-test-cpu")


 class TestComputeReordererPlans:
@@ -18,7 +18,7 @@ from sglang.srt.debug_utils.comparator.utils import Pair
 from sglang.test.ci.ci_register import register_cpu_ci

 register_cpu_ci(est_time=15, suite="base-a-test-cpu", nightly=True)
-register_cpu_ci(est_time=1, suite="base-c-test-cpu")
+register_cpu_ci(est_time=8, suite="base-c-test-cpu")


 class TestComputeAxisAlignerPlan:
@@ -17,7 +17,7 @@ from sglang.srt.debug_utils.comparator.dims_spec import TokenLayout
 from sglang.test.ci.ci_register import register_cpu_ci

 register_cpu_ci(est_time=15, suite="base-a-test-cpu", nightly=True)
-register_cpu_ci(est_time=1, suite="base-c-test-cpu")
+register_cpu_ci(est_time=6, suite="base-c-test-cpu")

 _sglang_plugin = _SGLangPlugin()
 _megatron_plugin = _MegatronPlugin()
@@ -11,7 +11,7 @@ from sglang.srt.debug_utils.comparator.utils import Pair
 from sglang.test.ci.ci_register import register_cpu_ci

 register_cpu_ci(est_time=15, suite="base-a-test-cpu", nightly=True)
-register_cpu_ci(est_time=1, suite="base-c-test-cpu")
+register_cpu_ci(est_time=6, suite="base-c-test-cpu")


 class TestExecuteConcat:
@@ -30,7 +30,7 @@ from sglang.srt.debug_utils.comparator.utils import Pair
 from sglang.test.ci.ci_register import register_cpu_ci

 register_cpu_ci(est_time=15, suite="base-a-test-cpu", nightly=True)
-register_cpu_ci(est_time=1, suite="base-c-test-cpu")
+register_cpu_ci(est_time=8, suite="base-c-test-cpu")


 def _named(tensor: torch.Tensor, names: list[str]) -> torch.Tensor:
@@ -24,7 +24,7 @@ from sglang.srt.debug_utils.comparator.utils import Pair
 from sglang.test.ci.ci_register import register_cpu_ci

 register_cpu_ci(est_time=30, suite="base-a-test-cpu", nightly=True)
-register_cpu_ci(est_time=1, suite="base-c-test-cpu")
+register_cpu_ci(est_time=6, suite="base-c-test-cpu")


 class TestBuildTokenIndexSGLangThd:
@@ -31,7 +31,7 @@ from sglang.srt.debug_utils.comparator.output_types import ReplicatedCheckResult
 from sglang.test.ci.ci_register import register_cpu_ci

 register_cpu_ci(est_time=10, suite="base-a-test-cpu", nightly=True)
-register_cpu_ci(est_time=1, suite="base-c-test-cpu")
+register_cpu_ci(est_time=8, suite="base-c-test-cpu")


 def _name_tensors(
@@ -10,7 +10,7 @@ from sglang.srt.debug_utils.comparator.dims_spec import ParallelAxis
 from sglang.test.ci.ci_register import register_cpu_ci

 register_cpu_ci(est_time=10, suite="base-a-test-cpu", nightly=True)
-register_cpu_ci(est_time=1, suite="base-c-test-cpu")
+register_cpu_ci(est_time=6, suite="base-c-test-cpu")


 class TestNormalizeParallelInfo:
@@ -20,7 +20,7 @@ from sglang.srt.debug_utils.comparator.dims_spec import ParallelAxis, parse_dims
 from sglang.test.ci.ci_register import register_cpu_ci

 register_cpu_ci(est_time=10, suite="base-a-test-cpu", nightly=True)
-register_cpu_ci(est_time=1, suite="base-c-test-cpu")
+register_cpu_ci(est_time=8, suite="base-c-test-cpu")


 class TestComputeUnsharderPlan:
@@ -13,7 +13,7 @@ from sglang.srt.debug_utils.comparator.dims_spec import (
 from sglang.test.ci.ci_register import register_cpu_ci

 register_cpu_ci(est_time=5, suite="base-a-test-cpu", nightly=True)
-register_cpu_ci(est_time=1, suite="base-c-test-cpu")
+register_cpu_ci(est_time=6, suite="base-c-test-cpu")


 class TestParseDim:
@@ -16,7 +16,7 @@ from sglang.srt.debug_utils.comparator.dims_spec import (
 from sglang.test.ci.ci_register import register_cpu_ci

 register_cpu_ci(est_time=5, suite="base-a-test-cpu", nightly=True)
-register_cpu_ci(est_time=1, suite="base-c-test-cpu")
+register_cpu_ci(est_time=8, suite="base-c-test-cpu")


 class TestSingletonDimUtilFilterOut:
@@ -15,7 +15,7 @@ from sglang.srt.debug_utils.comparator.dims_spec import (
 from sglang.test.ci.ci_register import register_cpu_ci

 register_cpu_ci(est_time=5, suite="base-a-test-cpu", nightly=True)
-register_cpu_ci(est_time=1, suite="base-c-test-cpu")
+register_cpu_ci(est_time=7, suite="base-c-test-cpu")


 class TestFindDimIndex:
@@ -10,7 +10,7 @@ from sglang.srt.debug_utils.comparator.dims_spec import (
 from sglang.test.ci.ci_register import register_cpu_ci

 register_cpu_ci(est_time=5, suite="base-a-test-cpu", nightly=True)
-register_cpu_ci(est_time=1, suite="base-c-test-cpu")
+register_cpu_ci(est_time=8, suite="base-c-test-cpu")


 class TestDimConstants:
@@ -17,7 +17,7 @@ from sglang.srt.debug_utils.comparator.threshold_dsl import DiffThresholdRule
 from sglang.test.ci.ci_register import register_cpu_ci

 register_cpu_ci(est_time=20, suite="base-a-test-cpu", nightly=True)
-register_cpu_ci(est_time=1, suite="base-c-test-cpu")
+register_cpu_ci(est_time=7, suite="base-c-test-cpu")


 class TestComputeTensorInfo:
@@ -22,7 +22,7 @@ from sglang.srt.debug_utils.comparator.tensor_comparator.types import (
 from sglang.test.ci.ci_register import register_cpu_ci

 register_cpu_ci(est_time=10, suite="base-a-test-cpu", nightly=True)
-register_cpu_ci(est_time=1, suite="base-c-test-cpu")
+register_cpu_ci(est_time=8, suite="base-c-test-cpu")


 def _make_stats(**overrides) -> TensorStats:
@@ -12,7 +12,7 @@ from sglang.srt.debug_utils.comparator.report_sink import report_sink
 from sglang.test.ci.ci_register import register_cpu_ci

 register_cpu_ci(est_time=10, suite="base-a-test-cpu", nightly=True)
-register_cpu_ci(est_time=1, suite="base-c-test-cpu")
+register_cpu_ci(est_time=7, suite="base-c-test-cpu")


 def _make_error_log(**overrides) -> ErrorLog:
@@ -17,7 +17,7 @@ from sglang.srt.debug_utils.comparator.meta_overrider import (
 from sglang.test.ci.ci_register import register_cpu_ci

 register_cpu_ci(est_time=10, suite="base-a-test-cpu", nightly=True)
-register_cpu_ci(est_time=1, suite="base-c-test-cpu")
+register_cpu_ci(est_time=8, suite="base-c-test-cpu")


 # ───────────────────── Unit: MetaOverrideRule ─────────────────────
@@ -45,7 +45,7 @@ from sglang.srt.debug_utils.comparator.utils import Pair, _check_equal_lengths
 from sglang.test.ci.ci_register import register_cpu_ci

 register_cpu_ci(est_time=10, suite="base-a-test-cpu", nightly=True)
-register_cpu_ci(est_time=1, suite="base-c-test-cpu")
+register_cpu_ci(est_time=7, suite="base-c-test-cpu")


 class TestCheckEqualLengths:
@@ -4,7 +4,7 @@ from sglang.srt.debug_utils.comparator.preset import PRESETS, expand_preset
 from sglang.test.ci.ci_register import register_cpu_ci

 register_cpu_ci(est_time=5, suite="base-a-test-cpu", nightly=True)
-register_cpu_ci(est_time=1, suite="base-c-test-cpu")
+register_cpu_ci(est_time=8, suite="base-c-test-cpu")


 class TestExpandPreset:
@@ -18,7 +18,7 @@ from sglang.srt.debug_utils.comparator.utils import (
 from sglang.test.ci.ci_register import register_cpu_ci

 register_cpu_ci(est_time=10, suite="base-a-test-cpu", nightly=True)
-register_cpu_ci(est_time=1, suite="base-c-test-cpu")
+register_cpu_ci(est_time=7, suite="base-c-test-cpu")


 class TestCalcRelDiff:
@@ -11,7 +11,7 @@ from sglang.srt.debug_utils.source_patcher.types import EditSpec, PatchSpec
 from sglang.test.ci.ci_register import register_cpu_ci

 register_cpu_ci(est_time=10, suite="base-a-test-cpu", nightly=True)
-register_cpu_ci(est_time=7, suite="base-c-test-cpu")
+register_cpu_ci(est_time=5, suite="base-c-test-cpu")

 SAMPLE_MODULE_NAME = "_source_patcher_test_fixtures.sample_module"

@@ -9,7 +9,7 @@ from sglang.srt.debug_utils.dumper import DumperConfig, _Dumper
 from sglang.test.ci.ci_register import register_cpu_ci

 register_cpu_ci(est_time=10, suite="base-a-test-cpu", nightly=True)
-register_cpu_ci(est_time=8, suite="base-c-test-cpu")
+register_cpu_ci(est_time=6, suite="base-c-test-cpu")

 SAMPLE_MODULE_NAME = "_source_patcher_test_fixtures.sample_module"

@@ -6,7 +6,7 @@ from sglang.srt.debug_utils.source_patcher.types import EditSpec, PatchApplicati
 from sglang.test.ci.ci_register import register_cpu_ci

 register_cpu_ci(est_time=10, suite="base-a-test-cpu", nightly=True)
-register_cpu_ci(est_time=7, suite="base-c-test-cpu")
+register_cpu_ci(est_time=4, suite="base-c-test-cpu")


 class TestApplyEdits:
@@ -23,7 +23,7 @@ from sglang.test.test_utils import (

 register_cuda_ci(est_time=70, stage="nightly", runner_config="1-gpu-large")
 register_amd_ci(est_time=40, suite="nightly-amd-1-gpu", nightly=True)
-register_cpu_ci(est_time=225, suite="base-c-test-cpu")
+register_cpu_ci(est_time=48, suite="base-c-test-cpu")


 class TestCrashDump(CustomTestCase):
@@ -26,7 +26,7 @@ from sglang.test.ci.ci_register import register_cpu_ci
 from sglang.test.test_utils import CustomTestCase

 register_cpu_ci(est_time=120, suite="base-a-test-cpu", nightly=True)
-register_cpu_ci(est_time=338, suite="base-c-test-cpu")
+register_cpu_ci(est_time=357, suite="base-c-test-cpu")

 # ==================== Non-E2E Tests ====================

@@ -19,7 +19,7 @@ from sglang.test.test_utils import (

 register_cuda_ci(est_time=240, stage="nightly", runner_config="1-gpu-large")
 register_amd_ci(est_time=120, suite="nightly-amd-1-gpu", nightly=True)
-register_cpu_ci(est_time=622, suite="base-c-test-cpu")
+register_cpu_ci(est_time=237, suite="base-c-test-cpu")


 class BaseTestSoftWatchdog:
@@ -12,7 +12,7 @@ from sglang.srt.parser.reasoning_parser import KimiK2Detector as KimiK2Reasoning
 from sglang.test.ci.ci_register import register_cpu_ci

 register_cpu_ci(5, "base-a-test-cpu")
-register_cpu_ci(est_time=7, suite="base-c-test-cpu")
+register_cpu_ci(est_time=4, suite="base-c-test-cpu")


 def _make_tool(name, parameters=None):
@@ -7,7 +7,11 @@ import requests
 from transformers import AutoModelForCausalLM, AutoTokenizer

 from sglang.srt.utils import kill_process_tree
-from sglang.test.ci.ci_register import register_amd_ci, register_cuda_ci
+from sglang.test.ci.ci_register import (
+    register_amd_ci,
+    register_cpu_ci,
+    register_cuda_ci,
+)
 from sglang.test.test_utils import (
    DEFAULT_SMALL_MODEL_NAME_FOR_TEST,
    DEFAULT_TIMEOUT_FOR_SERVER_LAUNCH,
@@ -18,6 +22,7 @@ from sglang.test.test_utils import (

 register_cuda_ci(est_time=42, stage="base-b", runner_config="1-gpu-small")
 register_amd_ci(est_time=38, suite="stage-b-test-1-gpu-small-amd")
+register_cpu_ci(est_time=58, suite="base-c-test-cpu")


 class TestInputEmbeds(CustomTestCase):
@@ -21,7 +21,11 @@ from transformers import AutoModelForCausalLM, AutoTokenizer

 from sglang.srt.environ import envs
 from sglang.srt.utils import kill_process_tree
-from sglang.test.ci.ci_register import register_amd_ci, register_cuda_ci
+from sglang.test.ci.ci_register import (
+    register_amd_ci,
+    register_cpu_ci,
+    register_cuda_ci,
+)
 from sglang.test.test_utils import (
    DEFAULT_SMALL_MODEL_NAME_FOR_TEST,
    DEFAULT_TIMEOUT_FOR_SERVER_LAUNCH,
@@ -32,6 +36,7 @@ from sglang.test.test_utils import (

 register_cuda_ci(est_time=43, stage="base-b", runner_config="1-gpu-small")
 register_amd_ci(est_time=43, suite="stage-b-test-1-gpu-small-amd")
+register_cpu_ci(est_time=54, suite="base-c-test-cpu")

 CHUNKED_PREFILL_SIZE = 256

@@ -8,7 +8,7 @@ from sglang.test.kv_canary.e2e_base import (
 )
 from sglang.test.test_utils import CustomTestCase

-register_cpu_ci(est_time=1, suite="base-c-test-cpu")
+register_cpu_ci(est_time=8, suite="base-c-test-cpu")


 class TestCanaryE2EBase(CustomTestCase):
@@ -8,7 +8,7 @@ from sglang.test.ci.ci_register import register_cpu_ci
 from sglang.test.kv_canary.e2e_base import CanaryE2EBase
 from sglang.test.test_utils import CustomTestCase

-register_cpu_ci(est_time=5, suite="base-c-test-cpu")
+register_cpu_ci(est_time=6, suite="base-c-test-cpu")


 _GOOD_LINE: str = SwaDivergenceLog(
@@ -13,7 +13,7 @@ from sglang.srt.server_args import ServerArgs
 from sglang.test.ci.ci_register import register_amd_ci, register_cpu_ci

 register_amd_ci(est_time=30, suite="nightly-amd-1-gpu", nightly=True)
-register_cpu_ci(est_time=8, suite="base-c-test-cpu")
+register_cpu_ci(est_time=7, suite="base-c-test-cpu")


 class MockTokenizerManager:
@@ -11,7 +11,7 @@ from sglang.test.test_utils import CustomTestCase

 register_cuda_ci(est_time=29, stage="base-b", runner_config="1-gpu-small")
 register_amd_ci(est_time=45, suite="stage-b-test-1-gpu-small-amd")
-register_cpu_ci(est_time=203, suite="base-c-test-cpu")
+register_cpu_ci(est_time=32, suite="base-c-test-cpu")


 class TestExternalModels(CustomTestCase):
@@ -13,7 +13,7 @@ from sglang.test.server_fixtures.default_fixture import DefaultServerBase

 register_cuda_ci(est_time=48, stage="base-b", runner_config="1-gpu-small")
 register_amd_ci(est_time=48, suite="stage-b-test-1-gpu-small-amd")
-register_cpu_ci(est_time=320, suite="base-c-test-cpu")
+register_cpu_ci(est_time=206, suite="base-c-test-cpu")


 class TestTransformersBackendEval(DefaultServerBase):
@@ -15,7 +15,7 @@ from sglang.srt.runtime_context import get_parallel
 from sglang.srt.server_args import ServerArgs, set_global_server_args_for_scheduler
 from sglang.test.ci.ci_register import register_cpu_ci

-register_cpu_ci(est_time=5, suite="base-b-test-cpu")
+register_cpu_ci(est_time=6, suite="base-b-test-cpu")


@pytest.fixture(autouse=True)
@@ -26,7 +26,7 @@ register_cuda_ci(
    runner_config="1-gpu-small",
 )
 register_amd_ci(est_time=60, suite="stage-b-test-1-gpu-small-amd")
-register_cpu_ci(est_time=179, suite="base-c-test-cpu")
+register_cpu_ci(est_time=49, suite="base-c-test-cpu")

 _MODEL_NAME = "Qwen/Qwen3-0.6B"

@@ -16,7 +16,11 @@ import unittest
 import requests

 from sglang.srt.utils import kill_process_tree
-from sglang.test.ci.ci_register import register_amd_ci, register_cuda_ci
+from sglang.test.ci.ci_register import (
+    register_amd_ci,
+    register_cpu_ci,
+    register_cuda_ci,
+)
 from sglang.test.test_utils import (
    DEFAULT_SMALL_MODEL_NAME_FOR_TEST,
    DEFAULT_TIMEOUT_FOR_SERVER_LAUNCH,
@@ -27,6 +31,7 @@ from sglang.test.test_utils import (

 register_cuda_ci(est_time=50, stage="base-b", runner_config="1-gpu-large")
 register_amd_ci(est_time=140, suite="stage-b-test-1-gpu-small-amd")
+register_cpu_ci(est_time=54, suite="base-c-test-cpu")

 # System message to guide Llama3.2 to produce proper tool call format
 SYSTEM_MESSAGE = (
@@ -11,7 +11,11 @@ import openai

 from sglang.srt.utils import kill_process_tree
 from sglang.srt.utils.hf_transformers_utils import get_tokenizer
-from sglang.test.ci.ci_register import register_amd_ci, register_cuda_ci
+from sglang.test.ci.ci_register import (
+    register_amd_ci,
+    register_cpu_ci,
+    register_cuda_ci,
+)
 from sglang.test.test_utils import (
    DEFAULT_SMALL_MODEL_NAME_FOR_TEST,
    DEFAULT_TIMEOUT_FOR_SERVER_LAUNCH,
@@ -24,6 +28,7 @@ from sglang.test.test_utils import (

 register_cuda_ci(est_time=58, stage="base-b", runner_config="1-gpu-large")
 register_amd_ci(est_time=41, suite="stage-b-test-1-gpu-small-amd")
+register_cpu_ci(est_time=101, suite="base-c-test-cpu")


 class TestLargeMaxNewTokens(CustomTestCase):
@@ -1,7 +1,11 @@
 import unittest

 from sglang.srt.utils import kill_process_tree
-from sglang.test.ci.ci_register import register_amd_ci, register_cuda_ci
+from sglang.test.ci.ci_register import (
+    register_amd_ci,
+    register_cpu_ci,
+    register_cuda_ci,
+)
 from sglang.test.kits.matched_stop_kit import MatchedStopMixin
 from sglang.test.test_utils import (
    DEFAULT_MODEL_NAME_FOR_TEST,
@@ -12,6 +16,7 @@ from sglang.test.test_utils import (

 register_cuda_ci(est_time=52, stage="base-b", runner_config="1-gpu-small")
 register_amd_ci(est_time=60, suite="stage-b-test-1-gpu-small-amd")
+register_cpu_ci(est_time=83, suite="base-c-test-cpu")


 class TestMatchedStop(CustomTestCase, MatchedStopMixin):
@@ -4,7 +4,11 @@ import unittest
 import openai

 from sglang.srt.utils import kill_process_tree
-from sglang.test.ci.ci_register import register_amd_ci, register_cuda_ci
+from sglang.test.ci.ci_register import (
+    register_amd_ci,
+    register_cpu_ci,
+    register_cuda_ci,
+)
 from sglang.test.test_utils import (
    DEFAULT_SMALL_EMBEDDING_MODEL_NAME_FOR_TEST,
    DEFAULT_TIMEOUT_FOR_SERVER_LAUNCH,
@@ -15,6 +19,7 @@ from sglang.test.test_utils import (

 register_cuda_ci(est_time=91, stage="base-b", runner_config="1-gpu-small")
 register_amd_ci(est_time=141, suite="stage-b-test-1-gpu-small-amd")
+register_cpu_ci(est_time=91, suite="base-c-test-cpu")


 class TestOpenAIEmbedding(CustomTestCase):
@@ -1,7 +1,11 @@
 import unittest

 from sglang.srt.environ import envs
-from sglang.test.ci.ci_register import register_amd_ci, register_cuda_ci
+from sglang.test.ci.ci_register import (
+    register_amd_ci,
+    register_cpu_ci,
+    register_cuda_ci,
+)
 from sglang.test.kits.radix_cache_server_kit import run_radix_attention_test
 from sglang.test.test_utils import (
    DEFAULT_SMALL_MODEL_NAME_FOR_TEST,
@@ -16,6 +20,7 @@ from sglang.test.test_utils import (
 # RadixAttention server integration tests
 register_cuda_ci(est_time=100, stage="base-b", runner_config="1-gpu-small")
 register_amd_ci(est_time=100, suite="stage-b-test-1-gpu-small-amd")
+register_cpu_ci(est_time=405, suite="base-c-test-cpu")


 class TestRadixCacheFCFS(CustomTestCase):
@@ -1,7 +1,11 @@
 import unittest

 from sglang.srt.utils import kill_process_tree
-from sglang.test.ci.ci_register import register_amd_ci, register_cuda_ci
+from sglang.test.ci.ci_register import (
+    register_amd_ci,
+    register_cpu_ci,
+    register_cuda_ci,
+)
 from sglang.test.kits.cache_hit_kit import run_multiturn_cache_hit_test
 from sglang.test.test_utils import (
    DEFAULT_SMALL_MODEL_NAME_FOR_TEST,
@@ -13,6 +17,7 @@ from sglang.test.test_utils import (

 register_cuda_ci(est_time=55, stage="base-b", runner_config="1-gpu-small")
 register_amd_ci(est_time=55, suite="stage-b-test-1-gpu-small-amd")
+register_cpu_ci(est_time=113, suite="base-c-test-cpu")

 MODEL = DEFAULT_SMALL_MODEL_NAME_FOR_TEST

@@ -1,12 +1,13 @@
 import unittest

-from sglang.test.ci.ci_register import register_cuda_ci
+from sglang.test.ci.ci_register import register_cpu_ci, register_cuda_ci
 from sglang.test.kits.kl_divergence_kit import KLDivergenceMixin
 from sglang.test.server_fixtures.default_fixture import DefaultServerBase

 MODEL = "openai/gpt-oss-20b"

 register_cuda_ci(est_time=151, stage="base-b", runner_config="1-gpu-large")
+register_cpu_ci(est_time=1602, suite="base-c-test-cpu")


 class TestSWARadixCacheKL(KLDivergenceMixin, DefaultServerBase):
@@ -10,7 +10,7 @@ from sglang.test.ci.ci_register import register_cpu_ci
 from sglang.test.test_utils import CustomTestCase

 register_cpu_ci(est_time=7, suite="base-a-test-cpu")
-register_cpu_ci(est_time=7, suite="base-c-test-cpu")
+register_cpu_ci(est_time=8, suite="base-c-test-cpu")


 # ---------------------------------------------------------------------------
@@ -19,7 +19,7 @@ from sglang.test.ci.ci_register import register_cpu_ci
 from sglang.test.test_utils import CustomTestCase

 register_cpu_ci(est_time=7, suite="base-a-test-cpu")
-register_cpu_ci(est_time=7, suite="base-c-test-cpu")
+register_cpu_ci(est_time=8, suite="base-c-test-cpu")

 _HTTP_SCOPE = {
    "type": "http",
@@ -22,7 +22,7 @@ from sglang.test.test_utils import (

 register_cuda_ci(est_time=50, stage="nightly", runner_config="1-gpu-large")
 register_amd_ci(est_time=120, suite="nightly-amd-1-gpu", nightly=True)
-register_cpu_ci(est_time=184, suite="base-c-test-cpu")
+register_cpu_ci(est_time=55, suite="base-c-test-cpu")


 class TestRoutingKeyScheduling(CustomTestCase):
@@ -5,7 +5,7 @@ CUDA-only: gpt-oss-20b is not part of the AMD streaming-session coverage.

 import unittest

-from sglang.test.ci.ci_register import register_cuda_ci
+from sglang.test.ci.ci_register import register_cpu_ci, register_cuda_ci
 from sglang.test.kits.streaming_session_kit import StreamingSessionKitMixin
 from sglang.test.server_fixtures.streaming_session_fixture import (
    SWA_COMMON_ARGS,
@@ -14,6 +14,7 @@ from sglang.test.server_fixtures.streaming_session_fixture import (
 )

 register_cuda_ci(est_time=130, stage="extra-a", runner_config="1-gpu-large")
+register_cpu_ci(est_time=449, suite="base-c-test-cpu")


 class TestStreamingSessionSWARetractMixedChunk(
@@ -11,7 +11,7 @@ from sglang.test.simple_eval_mixed_prefix_gsm8k import (
 )
 from sglang.test.test_utils import CustomTestCase

-register_cpu_ci(est_time=5, suite="base-c-test-cpu")
+register_cpu_ci(est_time=8, suite="base-c-test-cpu")


 def _write_synthetic_dataset(path: str, n: int) -> None:
@@ -10,7 +10,7 @@ from sglang.test.ci.ci_register import register_cpu_ci
 from sglang.test.run_eval import _run_sgl_eval
 from sglang.test.test_utils import CustomTestCase

-register_cpu_ci(est_time=5, suite="base-b-test-cpu")
+register_cpu_ci(est_time=6, suite="base-b-test-cpu")


 def _write_fake_metrics(out_parent: Path, eval_name: str, payload: dict) -> None:
@@ -31,7 +31,7 @@ from sglang.test.ci.ci_register import register_cpu_ci
 register_cpu_ci(2.0, "base-a-test-cpu")


-register_cpu_ci(est_time=7, suite="base-c-test-cpu")
+register_cpu_ci(est_time=5, suite="base-c-test-cpu")


 def _make_scheduler(grammar_backend_name="none", skip_tokenizer=False):
@@ -18,7 +18,7 @@ from sglang.srt.parser.reasoning_parser import KimiK3Detector as KimiK3Reasoning
 from sglang.test.ci.ci_register import register_cpu_ci

 register_cpu_ci(2.0, "base-a-test-cpu")
-register_cpu_ci(est_time=7, suite="base-c-test-cpu")
+register_cpu_ci(est_time=5, suite="base-c-test-cpu")


 class _DummyTokenizer:
@@ -15,7 +15,7 @@ from sglang.srt.constrained.utils import is_legacy_structural_tag
 from sglang.test.ci.ci_register import register_cpu_ci

 register_cpu_ci(1.0, "base-a-test-cpu")
-register_cpu_ci(est_time=7, suite="base-c-test-cpu")
+register_cpu_ci(est_time=5, suite="base-c-test-cpu")


 class TestIsLegacyStructuralTag(unittest.TestCase):
@@ -59,7 +59,7 @@ from sglang.srt.managers.io_struct import EmbeddingReqInput
 from sglang.test.ci.ci_register import register_cpu_ci

 register_cpu_ci(est_time=10, suite="base-a-test-cpu")
-register_cpu_ci(est_time=8, suite="base-c-test-cpu")
+register_cpu_ci(est_time=6, suite="base-c-test-cpu")


 # Mock TokenizerManager for embedding tests
@@ -34,7 +34,7 @@ from sglang.srt.function_call.qwen3_coder_detector import Qwen3CoderDetector
 from sglang.test.ci.ci_register import register_cpu_ci

 register_cpu_ci(est_time=15, suite="base-a-test-cpu")
-register_cpu_ci(est_time=61, suite="base-c-test-cpu")
+register_cpu_ci(est_time=70, suite="base-c-test-cpu")


 class TestInklingDetector(unittest.TestCase):
@@ -19,7 +19,7 @@ from sglang.srt.function_call.utils import (
 from sglang.test.ci.ci_register import register_cpu_ci

 register_cpu_ci(5, "base-a-test-cpu")
-register_cpu_ci(est_time=7, suite="base-c-test-cpu")
+register_cpu_ci(est_time=6, suite="base-c-test-cpu")


 class TestJsonSchemaConstraint(unittest.TestCase):
@@ -24,7 +24,7 @@ from sglang.srt.function_call.json_array_parser import JsonArrayParser
 from sglang.test.ci.ci_register import register_cpu_ci

 register_cpu_ci(5, "base-a-test-cpu")
-register_cpu_ci(est_time=7, suite="base-c-test-cpu")
+register_cpu_ci(est_time=4, suite="base-c-test-cpu")


 class TestParallelToolCalls(unittest.TestCase):
@@ -10,7 +10,7 @@ from sglang.srt.function_call.core_types import StreamingParseResult
 from sglang.test.ci.ci_register import register_cpu_ci

 register_cpu_ci(5, "base-a-test-cpu")
-register_cpu_ci(est_time=7, suite="base-c-test-cpu")
+register_cpu_ci(est_time=5, suite="base-c-test-cpu")


 class DummyDetector(BaseFormatDetector):
@@ -14,7 +14,7 @@ from sglang.srt.layers.moe.moe_runner.aiter import (
 from sglang.srt.layers.moe.moe_runner.base import MoeRunnerConfig
 from sglang.test.ci.ci_register import register_cpu_ci

-register_cpu_ci(est_time=5, suite="base-c-test-cpu")
+register_cpu_ci(est_time=7, suite="base-c-test-cpu")


 def _runner_input():
@@ -14,7 +14,7 @@ from sglang.srt.layers.moe.utils import MoeRunnerBackend
 from sglang.srt.runtime_context import get_flags
 from sglang.test.ci.ci_register import register_cpu_ci

-register_cpu_ci(est_time=5, suite="base-c-test-cpu")
+register_cpu_ci(est_time=6, suite="base-c-test-cpu")


@pytest.fixture
@@ -37,7 +37,7 @@ from sglang.test.test_utils import (

 register_cuda_ci(est_time=8, stage="base-b", runner_config="1-gpu-large")
 register_amd_ci(est_time=8, suite="stage-b-test-1-gpu-small-amd")
-register_cpu_ci(est_time=8, suite="base-c-test-cpu")
+register_cpu_ci(est_time=6, suite="base-c-test-cpu")


 class TestTokenizedReqInputMsgpack(unittest.TestCase):
@@ -7,7 +7,7 @@ from sglang.srt.environ import envs
 from sglang.srt.managers import mm_schedule as mm_utils
 from sglang.test.ci.ci_register import register_cpu_ci

-register_cpu_ci(est_time=5, suite="base-b-test-cpu")
+register_cpu_ci(est_time=8, suite="base-b-test-cpu")


@pytest.mark.parametrize(
@@ -40,7 +40,7 @@ from sglang.srt.utils.weight_checker import ParallelismInfo as PydanticParalleli
 from sglang.test.ci.ci_register import register_cpu_ci
 from sglang.test.test_utils import CustomTestCase

-register_cpu_ci(est_time=10, suite="base-c-test-cpu")
+register_cpu_ci(est_time=8, suite="base-c-test-cpu")


 def _round_trip(obj):
@@ -24,7 +24,7 @@ from sglang.srt.runtime_context import publish, reset_context
 from sglang.srt.sampling.sampling_params import SamplingParams

 register_cpu_ci(est_time=15, suite="base-a-test-cpu")
-register_cpu_ci(est_time=9, suite="base-c-test-cpu")
+register_cpu_ci(est_time=8, suite="base-c-test-cpu")


 class TestSchedulerPauseGeneration(unittest.TestCase):
@@ -3,7 +3,7 @@
 from sglang.test.ci.ci_register import register_cpu_ci

 register_cpu_ci(est_time=6, suite="base-a-test-cpu")
-register_cpu_ci(est_time=7, suite="base-c-test-cpu")
+register_cpu_ci(est_time=5, suite="base-c-test-cpu")

 import unittest
 from unittest.mock import MagicMock
@@ -7,7 +7,7 @@ from unittest.mock import MagicMock, patch
 from sglang.test.ci.ci_register import register_cpu_ci

 register_cpu_ci(est_time=60, suite="base-a-test-cpu", nightly=True)
-register_cpu_ci(est_time=7, suite="base-c-test-cpu")
+register_cpu_ci(est_time=5, suite="base-c-test-cpu")


 class TestCpuMonitor(unittest.TestCase):
--- a/Show More
+++ b/Show More