ci: migrate quantization kernel tests to test/registered/quant/ (#16323)

test/registered/quant/test_autoround.py

@@ -7,6 +7,7 @@ import unittest
 from types import SimpleNamespace
 
 from sglang.srt.utils import kill_process_tree
+from sglang.test.ci.ci_register import register_cuda_ci
 from sglang.test.run_eval import run_eval
 from sglang.test.test_utils import (
     DEFAULT_AUTOROUND_MODEL_NAME_FOR_TEST,
@@ -16,6 +17,9 @@ from sglang.test.test_utils import (
     popen_launch_server,
 )
 
+# AutoRound quantization tests
+register_cuda_ci(est_time=77, suite="stage-b-test-small-1-gpu")
+
 
 class TestAutoRound(CustomTestCase):
     @classmethod

test/registered/quant/test_awq_dequant.py

@@ -16,8 +16,12 @@ from sglang.srt.layers.quantization.awq_triton import (
     awq_dequantize_triton,
     awq_gemm_triton,
 )
+from sglang.test.ci.ci_register import register_amd_ci
 from sglang.test.test_utils import CustomTestCase
 
+# AWQ dequantization tests (AMD only)
+register_amd_ci(est_time=2, suite="stage-a-test-1")
+
 device = "cuda"
 
 

test/registered/quant/test_block_int8.py

@@ -7,8 +7,13 @@ from sglang.srt.layers.activation import SiluAndMul
 from sglang.srt.layers.moe.fused_moe_triton.fused_moe import fused_moe
 from sglang.srt.layers.moe.topk import TopKConfig, select_experts
 from sglang.srt.server_args import ServerArgs, set_global_server_args_for_scheduler
+from sglang.test.ci.ci_register import register_amd_ci, register_cuda_ci
 from sglang.test.test_utils import CustomTestCase
 
+# Block INT8 quantization kernel tests
+register_cuda_ci(est_time=44, suite="stage-b-test-small-1-gpu")
+register_amd_ci(est_time=22, suite="stage-a-test-1")
+
 
 # For test
 def native_per_token_group_quant_int8(x, group_size, eps=1e-10, dtype=torch.int8):

test/registered/quant/test_fp8_kernel.py

@@ -6,8 +6,12 @@ from sglang.srt.layers.quantization.fp8_kernel import (
     per_token_group_quant_fp8,
     w8a8_block_fp8_matmul,
 )
+from sglang.test.ci.ci_register import register_cuda_ci
 from sglang.test.test_utils import CustomTestCase
 
+# FP8 quantization kernel tests
+register_cuda_ci(est_time=10, suite="stage-b-test-small-1-gpu")
+
 
 class TestFP8Base(CustomTestCase):
     @classmethod

test/registered/quant/test_fused_rms_fp8_group_quant.py

@@ -1,12 +1,15 @@
-# test/srt/quant/test_fused_rms_fp8_group_quant.py
 import itertools
 import unittest
 
 import torch
 import torch.nn.functional as F
 
+from sglang.test.ci.ci_register import register_amd_ci
 from sglang.test.test_utils import CustomTestCase
 
+# Fused RMS FP8 group quantization tests (AMD/ROCm only)
+register_amd_ci(est_time=10, suite="stage-a-test-1")
+
 
 def _fp8_available() -> bool:
     # requirement：1) GPU；2) ROCm；3) torch support float8_e4m3fn

test/registered/quant/test_int8_kernel.py

@@ -8,8 +8,12 @@ from sglang.srt.layers.moe.fused_moe_triton.fused_moe import fused_moe
 from sglang.srt.layers.moe.topk import TopKConfig, select_experts
 from sglang.srt.layers.quantization.int8_kernel import per_token_quant_int8
 from sglang.srt.server_args import ServerArgs, set_global_server_args_for_scheduler
+from sglang.test.ci.ci_register import register_cuda_ci
 from sglang.test.test_utils import CustomTestCase
 
+# INT8 quantization kernel tests
+register_cuda_ci(est_time=8, suite="stage-b-test-small-1-gpu")
+
 
 def native_w8a8_per_token_matmul(A, B, As, Bs, output_dtype=torch.float16):
     """Matrix multiplication function that supports per-token input quantization and per-column weight quantization"""

test/registered/quant/test_triton_scaled_mm.py

@@ -5,8 +5,13 @@ import torch
 import torch.testing
 
 from sglang.srt.layers.quantization.fp8_kernel import triton_scaled_mm
+from sglang.test.ci.ci_register import register_amd_ci, register_cuda_ci
 from sglang.test.test_utils import CustomTestCase
 
+# Triton scaled matrix multiplication tests
+register_cuda_ci(est_time=8, suite="stage-b-test-small-1-gpu")
+register_amd_ci(est_time=12, suite="stage-a-test-1")
+
 
 def torch_scaled_mm(
     a: torch.Tensor,

test/registered/quant/test_w8a8_quantization.py

@@ -5,6 +5,7 @@ from types import SimpleNamespace
 import requests
 
 from sglang.srt.utils import kill_process_tree
+from sglang.test.ci.ci_register import register_cuda_ci
 from sglang.test.few_shot_gsm8k import run_eval
 from sglang.test.test_utils import (
     DEFAULT_TIMEOUT_FOR_SERVER_LAUNCH,
@@ -13,6 +14,9 @@ from sglang.test.test_utils import (
     popen_launch_server,
 )
 
+# W8A8 quantization server integration tests
+register_cuda_ci(est_time=160, suite="stage-b-test-small-1-gpu")
+
 
 class BaseW8A8Test(CustomTestCase):
     model: str = None

test/srt/run_suite.py

@@ -41,12 +41,7 @@ suites = {
         TestFile("openai_server/validation/test_openai_server_ignore_eos.py", 6),
         TestFile("openai_server/validation/test_request_length_validation.py", 38),
         TestFile("ops/test_repeat_interleave.py", 60),
-        TestFile("quant/test_block_int8.py", 44),
+        # quant tests moved to test/registered/quant/
-        TestFile("quant/test_fp8_kernel.py", 10),
-        TestFile("quant/test_int8_kernel.py", 8),
-        TestFile("quant/test_triton_scaled_mm.py", 8),
-        TestFile("quant/test_w8a8_quantization.py", 160),
-        TestFile("quant/test_autoround.py", 77),
         TestFile("rl/test_fp32_lm_head.py", 9),
         # TestFile("rl/test_update_weights_from_disk.py", 210),  # Temporarily disabled, see https://github.com/sgl-project/sglang/pull/13998
         TestFile("rl/test_update_weights_from_tensor.py", 195),
@@ -228,10 +223,7 @@ suite_amd = {
         TestFile("openai_server/validation/test_openai_server_ignore_eos.py", 85),
         TestFile("openai_server/validation/test_request_length_validation.py", 31),
         TestFile("ops/test_repeat_interleave.py", 75),
-        TestFile("quant/test_awq_dequant.py", 2),
+        # quant tests moved to test/registered/quant/
-        TestFile("quant/test_block_int8.py", 22),
-        TestFile("quant/test_fused_rms_fp8_group_quant.py", 10),
-        TestFile("quant/test_triton_scaled_mm.py", 12),
         TestFile("rl/test_fp32_lm_head.py", 15),
         # TestFile("rl/test_update_weights_from_disk.py", 210),  # Temporarily disabled, see https://github.com/sgl-project/sglang/pull/13998
         TestFile("rotary_embedding/test_mrope.py", 15),