EngineX-Hygon/sglang
5
0
Fork 0
Code Issues Pull Requests Actions 7 Projects Releases Wiki Activity

Reorder PD disagg CI tests (#11438)

Browse Source
This commit is contained in:
Lianmin Zheng
2025-10-10 17:56:49 -07:00
committed by GitHub
parent 92777135a0
commit 61055cb309
7 changed files with 402 additions and 152 deletions
Download Patch File Download Diff File Expand all files Collapse all files

82
.github/workflows/pr-test.yml vendored
View File

@@ -693,87 +693,6 @@ jobs:
cd test/srt
python3 run_suite.py --suite per-commit-4-gpu-b200 --auto-partition-id 0 --auto-partition-size 1 --timeout-per-file 3600
unit-test-disaggregation-2-gpu:
needs: [check-changes, sgl-kernel-build-wheels]
if: always() && !failure() && !cancelled() &&
((needs.check-changes.outputs.main_package == 'true') || (needs.check-changes.outputs.sgl_kernel == 'true'))
runs-on: 2-gpu-runner
steps:
- name: Checkout code
uses: actions/checkout@v4
- name: Download artifacts
if: needs.check-changes.outputs.sgl_kernel == 'true'
uses: actions/download-artifact@v4
with:
path: sgl-kernel/dist/
merge-multiple: true
pattern: wheel-python3.10-cuda12.9
- name: Install dependencies
run: |
CUSTOM_BUILD_SGL_KERNEL=${{needs.check-changes.outputs.sgl_kernel}} bash scripts/ci/ci_install_dependency.sh
- name: Run test
timeout-minutes: 20
run: |
cd test/srt
python3 run_suite.py --suite per-commit-2-gpu-disaggregation
unit-test-disaggregation-4-gpu:
needs: [check-changes, unit-test-disaggregation-2-gpu, sgl-kernel-build-wheels]
if: always() && !failure() && !cancelled() &&
((needs.check-changes.outputs.main_package == 'true') || (needs.check-changes.outputs.sgl_kernel == 'true'))
runs-on: 4-gpu-runner
steps:
- name: Checkout code
uses: actions/checkout@v4
- name: Download artifacts
if: needs.check-changes.outputs.sgl_kernel == 'true'
uses: actions/download-artifact@v4
with:
path: sgl-kernel/dist/
merge-multiple: true
pattern: wheel-python3.10-cuda12.9
- name: Install dependencies
run: |
CUSTOM_BUILD_SGL_KERNEL=${{needs.check-changes.outputs.sgl_kernel}} bash scripts/ci/ci_install_dependency.sh
- name: Run test
timeout-minutes: 20
run: |
cd test/srt
python3 run_suite.py --suite per-commit-4-gpu-disaggregation
unit-test-disaggregation-8-gpu:
needs: [check-changes, unit-test-disaggregation-2-gpu, sgl-kernel-build-wheels]
if: always() && !failure() && !cancelled() &&
((needs.check-changes.outputs.main_package == 'true') || (needs.check-changes.outputs.sgl_kernel == 'true'))
runs-on: 8-gpu-h200
steps:
- name: Checkout code
uses: actions/checkout@v4
- name: Download artifacts
if: needs.check-changes.outputs.sgl_kernel == 'true'
uses: actions/download-artifact@v4
with:
path: sgl-kernel/dist/
merge-multiple: true
pattern: wheel-python3.10-cuda12.9
- name: Install dependencies
run: |
CUSTOM_BUILD_SGL_KERNEL=${{needs.check-changes.outputs.sgl_kernel}} bash scripts/ci/ci_install_dependency.sh
- name: Run test
timeout-minutes: 20
run: |
cd test/srt
python3 run_suite.py --suite per-commit-8-gpu-disaggregation
pr-test-finish:
needs: [
check-changes,
@@ -788,7 +707,6 @@ jobs:
accuracy-test-1-gpu, accuracy-test-2-gpu,
unit-test-deepep-4-gpu, unit-test-deepep-8-gpu,
unit-test-backend-4-gpu-b200,
unit-test-disaggregation-2-gpu, unit-test-disaggregation-4-gpu, unit-test-disaggregation-8-gpu,
]
if: always()
runs-on: ubuntu-latest

2
python/sglang/srt/utils/common.py
View File

@@ -523,7 +523,7 @@ def make_layers_non_pp(
layer_fn: LayerFn,
prefix: str = "",
) -> torch.nn.ModuleList:
from sglang.srt.offloader import get_offloader
from sglang.srt.utils.offloader import get_offloader
layers = torch.nn.ModuleList(
get_offloader().wrap_modules(

339
scripts/sort_testcases_alphabetically.py Normal file
View File

@@ -0,0 +1,339 @@
"""
Sort the test case by name alphabetically for run_suite.py
"""
from dataclasses import dataclass
@dataclass
class TestFile:
name: str
estimated_time: float = 60
suites = {
"per-commit": [
TestFile("function_call/test_json_schema_constraint.py", 30),
TestFile("hicache/test_hicache.py", 116),
TestFile("hicache/test_hicache_eagle.py", 150),
TestFile("hicache/test_hicache_mla.py", 127),
TestFile("hicache/test_hicache_storage.py", 127),
TestFile("lora/test_lora.py", 200),
TestFile("lora/test_lora_backend.py", 99),
TestFile("lora/test_lora_eviction.py", 200),
TestFile("lora/test_lora_qwen3.py", 97),
TestFile("lora/test_lora_radix_cache.py", 100),
TestFile("lora/test_lora_update.py", 400),
TestFile("lora/test_multi_lora_backend.py", 60),
TestFile("models/test_embedding_models.py", 73),
TestFile("models/test_encoder_embedding_models.py", 100),
TestFile("models/test_cross_encoder_models.py", 100),
TestFile("models/test_compressed_tensors_models.py", 42),
TestFile("models/test_generation_models.py", 103),
TestFile("models/test_nvidia_nemotron_nano_v2.py", 180),
TestFile("models/test_qwen_models.py", 82),
TestFile("models/test_reward_models.py", 132),
TestFile("models/test_transformers_models.py", 320),
TestFile("models/test_vlm_models.py", 741),
TestFile("openai_server/basic/test_protocol.py", 10),
TestFile("openai_server/basic/test_serving_chat.py", 10),
TestFile("openai_server/basic/test_serving_completions.py", 10),
TestFile("openai_server/basic/test_serving_embedding.py", 10),
TestFile("openai_server/basic/test_openai_embedding.py", 141),
TestFile("openai_server/basic/test_openai_server.py", 149),
TestFile("openai_server/features/test_enable_thinking.py", 70),
TestFile("openai_server/features/test_json_constrained.py", 98),
TestFile("openai_server/features/test_json_mode.py", 90),
TestFile("openai_server/features/test_openai_server_ebnf.py", 95),
TestFile("openai_server/features/test_openai_server_hidden_states.py", 240),
TestFile("openai_server/features/test_reasoning_content.py", 89),
TestFile("openai_server/function_call/test_openai_function_calling.py", 60),
TestFile("openai_server/function_call/test_tool_choice.py", 226),
TestFile("openai_server/validation/test_large_max_new_tokens.py", 41),
TestFile("openai_server/validation/test_matched_stop.py", 60),
TestFile("openai_server/validation/test_openai_server_ignore_eos.py", 85),
TestFile("openai_server/validation/test_request_length_validation.py", 31),
TestFile("quant/test_block_int8.py", 22),
TestFile("quant/test_fp8_kernel.py", 8),
TestFile("quant/test_int8_kernel.py", 8),
TestFile("quant/test_triton_scaled_mm.py", 8),
TestFile("quant/test_w8a8_quantization.py", 46),
TestFile("rl/test_fp32_lm_head.py", 30),
TestFile("rl/test_update_weights_from_disk.py", 114),
TestFile("rl/test_update_weights_from_tensor.py", 48),
TestFile("test_abort.py", 51),
TestFile("test_create_kvindices.py", 2),
TestFile("test_chunked_prefill.py", 313),
TestFile("test_deterministic.py", 300),
TestFile("test_eagle_infer_a.py", 370),
TestFile("test_eagle_infer_b.py", 700),
TestFile("test_ebnf_constrained.py", 108),
TestFile("test_eval_fp8_accuracy.py", 303),
TestFile("test_fa3.py", 376),
# TestFile("test_flashmla.py", 352),
TestFile("test_function_call_parser.py", 10),
TestFile("test_fused_moe.py", 30),
TestFile("test_gpt_oss_1gpu.py", 600),
TestFile("test_harmony_parser.py", 20),
TestFile("test_hidden_states.py", 55),
TestFile("test_hybrid_attn_backend.py", 100),
TestFile("test_input_embeddings.py", 38),
TestFile("test_io_struct.py", 8),
TestFile("test_jinja_template_utils.py", 1),
TestFile("test_logprobs.py", 55),
TestFile("test_metrics.py", 32),
TestFile("test_metrics_utils.py", 1),
TestFile("test_mla.py", 167),
TestFile("test_mla_deepseek_v3.py", 500),
TestFile("test_mla_int8_deepseek_v3.py", 429),
TestFile("test_mla_flashinfer.py", 302),
TestFile("test_mla_fp8.py", 93),
TestFile("test_modelopt_loader.py", 30),
TestFile("test_multi_tokenizer.py", 230),
TestFile("test_ngram_speculative_decoding.py", 250),
TestFile("test_no_chunked_prefill.py", 108),
TestFile("test_no_overlap_scheduler.py", 234),
TestFile("test_original_logprobs.py", 41),
TestFile("test_penalty.py", 41),
TestFile("test_page_size.py", 60),
TestFile("test_priority_scheduling.py", 100),
TestFile("test_pytorch_sampling_backend.py", 66),
TestFile("test_radix_attention.py", 105),
TestFile("test_radix_cache_unit.py", 5),
TestFile("test_regex_constrained.py", 64),
TestFile("test_reasoning_parser.py", 5),
TestFile("test_retract_decode.py", 54),
TestFile("test_request_queue_validation.py", 30),
TestFile("test_score_api.py", 180),
TestFile("test_server_args.py", 1),
TestFile("test_skip_tokenizer_init.py", 117),
TestFile("test_srt_engine.py", 261),
TestFile("test_srt_endpoint.py", 130),
TestFile("test_standalone_speculative_decoding.py", 250),
TestFile("test_start_profile.py", 60),
TestFile("test_swa_unittest.py", 1),
TestFile("test_torch_compile.py", 76),
TestFile("test_torch_compile_moe.py", 172),
TestFile("test_torch_native_attention_backend.py", 123),
TestFile("test_torchao.py", 70),
TestFile("test_triton_attention_kernels.py", 4),
TestFile("test_triton_attention_backend.py", 150),
TestFile("test_triton_moe_channel_fp8_kernel.py", 25),
TestFile("test_triton_sliding_window.py", 250),
TestFile("test_utils_update_weights.py", 48),
TestFile("test_vision_chunked_prefill.py", 175),
TestFile("test_vlm_input_format.py", 300),
TestFile("test_vision_openai_server_a.py", 724),
TestFile("test_vision_openai_server_b.py", 446),
],
"per-commit-2-gpu": [
TestFile("ep/test_moe_ep.py", 140),
TestFile("hicache/test_hicache_storage_file_backend.py", 200),
TestFile("hicache/test_hicache_storage_mooncake_backend.py", 400),
TestFile("hicache/test_hicache_storage_3fs_backend.py", 200),
TestFile("layers/attention/mamba/test_mamba2_mixer.py", 110),
TestFile("lora/test_lora_tp.py", 116),
TestFile("rl/test_update_weights_from_distributed.py", 103),
TestFile("test_data_parallelism.py", 73),
TestFile("test_disaggregation_basic.py", 400),
TestFile("test_dp_attention.py", 594),
TestFile("test_load_weights_from_remote_instance.py", 72),
TestFile("test_patch_torch.py", 19),
TestFile("test_release_memory_occupation.py", 257),
],
"per-commit-4-gpu": [
TestFile("models/test_qwen3_next_models.py", 291),
TestFile("test_disaggregation_dp_attention.py", 155),
TestFile("test_gpt_oss_4gpu.py", 300),
TestFile("test_local_attn.py", 411),
TestFile("test_multi_instance_release_memory_occupation.py", 64),
TestFile("test_pp_single_node.py", 481),
],
"per-commit-8-gpu": [
TestFile("lora/test_lora_llama4.py", 400),
TestFile("test_deepseek_v3_basic.py", 275),
TestFile("test_deepseek_v3_mtp.py", 275),
TestFile("test_disaggregation_different_tp.py", 600),
TestFile("test_disaggregation_pp.py", 140),
],
"per-commit-4-gpu-b200": [
# TestFile("test_gpt_oss_4gpu.py", 600),
# TestFile("test_deepseek_v3_fp4_4gpu.py", 3600),
],
"per-commit-4-gpu-deepep": [
TestFile("ep/test_deepep_small.py", 531),
],
"per-commit-8-gpu-deepep": [
TestFile("ep/test_deepep_large.py", 338),
],
"per-commit-8-gpu-h20": [
TestFile("quant/test_w4a8_deepseek_v3.py", 371),
],
"vllm_dependency_test": [
TestFile("quant/test_awq.py", 163),
TestFile("test_bnb.py", 5),
TestFile("test_gptqmodel_dynamic.py", 102),
TestFile("test_vllm_dependency.py", 185),
# TestFile("test_gguf.py", 96),
],
}
# Add AMD tests
# NOTE: please sort the test cases alphabetically by the test file name
suite_amd = {
"per-commit-amd": [
TestFile("hicache/test_hicache.py", 116),
TestFile("hicache/test_hicache_mla.py", 127),
TestFile("hicache/test_hicache_storage.py", 127),
TestFile("lora/test_lora.py", 200),
TestFile("lora/test_lora_eviction.py", 200),
TestFile("lora/test_lora_backend.py", 99),
TestFile("lora/test_multi_lora_backend.py", 60),
TestFile("lora/test_lora_cuda_graph.py", 250),
TestFile("lora/test_lora_qwen3.py", 97),
# TestFile("models/test_embedding_models.py", 73), # Disabled temporarily, see https://github.com/sgl-project/sglang/issues/11127
TestFile("models/test_compressed_tensors_models.py", 42),
TestFile("models/test_qwen_models.py", 82),
TestFile("models/test_reward_models.py", 132),
TestFile("models/test_transformers_models.py", 320),
TestFile("openai_server/basic/test_protocol.py", 10),
TestFile("openai_server/basic/test_serving_chat.py", 10),
TestFile("openai_server/basic/test_serving_completions.py", 10),
TestFile("openai_server/basic/test_serving_embedding.py", 10),
TestFile("openai_server/basic/test_openai_embedding.py", 141),
TestFile("openai_server/basic/test_openai_server.py", 149),
TestFile("openai_server/features/test_enable_thinking.py", 70),
TestFile("openai_server/features/test_json_constrained.py", 98),
TestFile("openai_server/features/test_json_mode.py", 90),
TestFile("openai_server/features/test_openai_server_ebnf.py", 95),
# TestFile("openai_server/features/test_openai_server_hidden_states.py", 240),
TestFile("openai_server/features/test_reasoning_content.py", 89),
TestFile("openai_server/function_call/test_openai_function_calling.py", 60),
TestFile("openai_server/function_call/test_tool_choice.py", 226),
TestFile("function_call/test_json_schema_constraint.py", 30),
TestFile("openai_server/validation/test_large_max_new_tokens.py", 41),
TestFile("openai_server/validation/test_matched_stop.py", 60),
TestFile("openai_server/validation/test_openai_server_ignore_eos.py", 85),
TestFile("openai_server/validation/test_request_length_validation.py", 31),
TestFile("quant/test_block_int8.py", 22),
TestFile("quant/test_awq_dequant.py", 2),
TestFile("rl/test_update_weights_from_disk.py", 114),
# TestFile("rl/test_update_weights_from_tensor.py", 48),
TestFile("test_abort.py", 51),
TestFile("test_create_kvindices.py", 2),
TestFile("test_chunked_prefill.py", 313),
TestFile("test_ebnf_constrained.py", 108),
TestFile("test_eval_fp8_accuracy.py", 303),
TestFile("test_function_call_parser.py", 10),
TestFile("test_fused_moe.py", 30),
TestFile("test_input_embeddings.py", 38),
TestFile("test_io_struct.py", 8),
TestFile("test_jinja_template_utils.py", 1),
TestFile("test_metrics.py", 32),
TestFile("test_metrics_utils.py", 1),
TestFile("test_mla.py", 242),
TestFile("test_mla_deepseek_v3.py", 221),
TestFile("test_no_chunked_prefill.py", 108),
# TestFile("test_no_overlap_scheduler.py", 234), # Disabled temporarily and track in #7703
TestFile("test_penalty.py", 41),
TestFile("test_page_size.py", 60),
TestFile("test_pytorch_sampling_backend.py", 66),
TestFile("test_radix_attention.py", 105),
TestFile("test_regex_constrained.py", 64),
TestFile("test_retract_decode.py", 54),
TestFile("test_reasoning_parser.py", 5),
TestFile("test_rope_rocm.py", 3),
TestFile("test_server_args.py", 1),
TestFile("test_skip_tokenizer_init.py", 117),
TestFile("test_srt_engine.py", 261),
TestFile("test_srt_endpoint.py", 130),
TestFile("test_torch_compile.py", 169),
TestFile("test_torch_compile_moe.py", 172),
TestFile("test_torch_native_attention_backend.py", 123),
TestFile("test_triton_attention_backend.py", 150),
# TestFile("test_vision_chunked_prefill.py", 175), # Disabled temporarily and track in #7701
TestFile("test_wave_attention_kernels.py", 2),
# TestFile("test_wave_attention_backend.py", 150), # Disabled temporarily, see https://github.com/sgl-project/sglang/issues/11127
],
"per-commit-amd-mi35x": [
TestFile("test_mla.py", 242),
TestFile("test_gpt_oss_1gpu.py", 600),
],
"per-commit-2-gpu-amd": [
TestFile("lora/test_lora_tp.py", 116),
TestFile("rl/test_update_weights_from_distributed.py", 103),
TestFile("test_data_parallelism.py", 73),
TestFile("test_load_weights_from_remote_instance.py", 72),
# TestFile("test_patch_torch.py", 19), # Disabled temporarily, see https://github.com/sgl-project/sglang/issues/11127
],
"per-commit-4-gpu-amd": [
TestFile("test_pp_single_node.py", 150),
],
"per-commit-8-gpu-amd": [
TestFile("test_deepseek_v3_basic.py", 275),
TestFile("test_deepseek_v3_mtp.py", 275),
],
"nightly-amd": [
TestFile("test_nightly_gsm8k_eval_amd.py"),
],
}
# Add Intel Xeon tests
# NOTE: please sort the test cases alphabetically by the test file name
suite_xeon = {
"per-commit-cpu": [
TestFile("cpu/test_activation.py"),
TestFile("cpu/test_binding.py"),
TestFile("cpu/test_decode.py"),
TestFile("cpu/test_extend.py"),
TestFile("cpu/test_gemm.py"),
TestFile("cpu/test_mla.py"),
TestFile("cpu/test_moe.py"),
TestFile("cpu/test_norm.py"),
TestFile("cpu/test_qkv_proj_with_rope.py"),
TestFile("cpu/test_rope.py"),
TestFile("cpu/test_shared_expert.py"),
TestFile("cpu/test_topk.py"),
TestFile("test_intel_amx_attention_backend.py"),
TestFile("test_cpu_graph.py"),
],
}
# Add Ascend NPU tests
# NOTE: please sort the test cases alphabetically by the test file name
suite_ascend = {
"per-commit-1-ascend-npu": [
TestFile("ascend/test_ascend_graph_tp1_bf16.py", 400),
TestFile("ascend/test_ascend_tp1_bf16.py", 400),
],
"per-commit-2-ascend-npu": [
TestFile("ascend/test_ascend_graph_tp2_bf16.py", 400),
TestFile("ascend/test_ascend_mla_fia_w8a8int8.py", 400),
TestFile("ascend/test_ascend_tp2_bf16.py", 400),
TestFile("ascend/test_ascend_tp2_fia_bf16.py", 400),
],
"per-commit-4-ascend-npu": [
TestFile("ascend/test_ascend_mla_w8a8int8.py", 400),
TestFile("ascend/test_ascend_tp4_bf16.py", 400),
],
"per-commit-16-ascend-a3": [
TestFile("ascend/test_ascend_deepep.py", 400),
],
}
suites.update(suite_amd)
suites.update(suite_xeon)
suites.update(suite_ascend)
if __name__ == "__main__":
for key in suites:
cases = suites[key]
names = [x.name for x in cases]
names.sort()
print(f' "{key}": [')
for name in names:
estimated_time = [x.estimated_time for x in cases if x.name == name][0]
print(f' TestFile("{name}", {estimated_time}),')
print(f" ],\n")

5
test/srt/models/test_nvidia_nemotron_nano_v2.py
View File

@@ -1,3 +1,4 @@
import unittest
from types import SimpleNamespace
from sglang.srt.utils import kill_process_tree
@@ -42,3 +43,7 @@ class TestNvidiaNemotronNanoV2(CustomTestCase):
metrics = run_eval(args)
print(f"{metrics=}")
self.assertGreater(metrics["accuracy"], 0.87)
if __name__ == "__main__":
unittest.main()

122
test/srt/run_suite.py
View File

@@ -2,7 +2,6 @@ import argparse
import glob
from dataclasses import dataclass
from sglang.srt.utils import is_hip
from sglang.test.test_utils import run_unittest_files
@@ -12,38 +11,37 @@ class TestFile:
estimated_time: float = 60
# NOTE: please sort the test cases alphabetically by the test file name
suites = {
"per-commit": [
TestFile("function_call/test_json_schema_constraint.py", 30),
TestFile("hicache/test_hicache.py", 116),
TestFile("hicache/test_hicache_eagle.py", 150),
TestFile("hicache/test_hicache_mla.py", 127),
TestFile("hicache/test_hicache_storage.py", 127),
TestFile("hicache/test_hicache_eagle.py", 150),
TestFile("lora/test_lora.py", 200),
TestFile("lora/test_lora_eviction.py", 200),
TestFile("lora/test_lora_backend.py", 99),
TestFile("lora/test_multi_lora_backend.py", 60),
TestFile("lora/test_lora_update.py", 400),
TestFile("lora/test_lora_eviction.py", 200),
TestFile("lora/test_lora_qwen3.py", 97),
TestFile("lora/test_lora_radix_cache.py", 100),
TestFile("models/test_embedding_models.py", 73),
# TestFile("models/test_clip_models.py", 52),
TestFile("models/test_encoder_embedding_models.py", 100),
TestFile("models/test_cross_encoder_models.py", 100),
TestFile("lora/test_lora_update.py", 400),
TestFile("lora/test_multi_lora_backend.py", 60),
TestFile("models/test_compressed_tensors_models.py", 42),
TestFile("models/test_cross_encoder_models.py", 100),
TestFile("models/test_embedding_models.py", 73),
TestFile("models/test_encoder_embedding_models.py", 100),
TestFile("models/test_generation_models.py", 103),
# TestFile("models/test_gme_qwen_models.py", 45),
# TestFile("models/test_grok_models.py", 60), # Disabled due to illegal memory access
TestFile("models/test_nvidia_nemotron_nano_v2.py", 180),
TestFile("models/test_qwen_models.py", 82),
TestFile("models/test_reward_models.py", 132),
TestFile("models/test_vlm_models.py", 741),
TestFile("models/test_transformers_models.py", 320),
TestFile("models/test_vlm_models.py", 741),
TestFile("openai_server/basic/test_openai_embedding.py", 141),
TestFile("openai_server/basic/test_openai_server.py", 149),
TestFile("openai_server/basic/test_protocol.py", 10),
TestFile("openai_server/basic/test_serving_chat.py", 10),
TestFile("openai_server/basic/test_serving_completions.py", 10),
TestFile("openai_server/basic/test_serving_embedding.py", 10),
TestFile("openai_server/basic/test_openai_embedding.py", 141),
TestFile("openai_server/basic/test_openai_server.py", 149),
TestFile("openai_server/features/test_enable_thinking.py", 70),
TestFile("openai_server/features/test_json_constrained.py", 98),
TestFile("openai_server/features/test_json_mode.py", 90),
@@ -65,8 +63,8 @@ suites = {
TestFile("rl/test_update_weights_from_disk.py", 114),
TestFile("rl/test_update_weights_from_tensor.py", 48),
TestFile("test_abort.py", 51),
TestFile("test_create_kvindices.py", 2),
TestFile("test_chunked_prefill.py", 313),
TestFile("test_create_kvindices.py", 2),
TestFile("test_deterministic.py", 300),
TestFile("test_eagle_infer_a.py", 370),
TestFile("test_eagle_infer_b.py", 700),
@@ -80,8 +78,6 @@ suites = {
TestFile("test_harmony_parser.py", 20),
TestFile("test_hidden_states.py", 55),
TestFile("test_hybrid_attn_backend.py", 100),
TestFile("test_standalone_speculative_decoding.py", 250),
TestFile("test_ngram_speculative_decoding.py", 250),
TestFile("test_input_embeddings.py", 38),
TestFile("test_io_struct.py", 8),
TestFile("test_jinja_template_utils.py", 1),
@@ -90,74 +86,76 @@ suites = {
TestFile("test_metrics_utils.py", 1),
TestFile("test_mla.py", 167),
TestFile("test_mla_deepseek_v3.py", 500),
TestFile("test_mla_int8_deepseek_v3.py", 429),
TestFile("test_mla_flashinfer.py", 302),
TestFile("test_mla_fp8.py", 93),
TestFile("test_mla_int8_deepseek_v3.py", 429),
TestFile("test_modelopt_loader.py", 30),
TestFile("test_multi_tokenizer.py", 230),
TestFile("test_ngram_speculative_decoding.py", 250),
TestFile("test_no_chunked_prefill.py", 108),
TestFile("test_no_overlap_scheduler.py", 234),
TestFile("test_original_logprobs.py", 41),
TestFile("test_penalty.py", 41),
TestFile("test_page_size.py", 60),
TestFile("test_penalty.py", 41),
TestFile("test_priority_scheduling.py", 100),
TestFile("test_pytorch_sampling_backend.py", 66),
TestFile("test_radix_attention.py", 105),
TestFile("test_radix_cache_unit.py", 5),
TestFile("test_regex_constrained.py", 64),
TestFile("test_reasoning_parser.py", 5),
TestFile("test_retract_decode.py", 54),
TestFile("test_regex_constrained.py", 64),
TestFile("test_request_queue_validation.py", 30),
TestFile("test_retract_decode.py", 54),
TestFile("test_score_api.py", 180),
TestFile("test_server_args.py", 1),
TestFile("test_skip_tokenizer_init.py", 117),
TestFile("test_srt_engine.py", 261),
TestFile("test_srt_endpoint.py", 130),
TestFile("test_srt_engine.py", 261),
TestFile("test_standalone_speculative_decoding.py", 250),
TestFile("test_start_profile.py", 60),
TestFile("test_swa_unittest.py", 1),
TestFile("test_torch_compile.py", 76),
TestFile("test_torch_compile_moe.py", 172),
TestFile("test_torch_native_attention_backend.py", 123),
TestFile("test_torchao.py", 70),
TestFile("test_triton_attention_kernels.py", 4),
TestFile("test_triton_attention_backend.py", 150),
TestFile("test_triton_attention_kernels.py", 4),
TestFile("test_triton_moe_channel_fp8_kernel.py", 25),
TestFile("test_triton_sliding_window.py", 250),
TestFile("test_utils_update_weights.py", 48),
TestFile("test_vision_chunked_prefill.py", 175),
TestFile("test_vlm_input_format.py", 300),
TestFile("test_vision_openai_server_a.py", 724),
TestFile("test_vision_openai_server_b.py", 446),
TestFile("layers/attention/mamba/test_causal_conv1d.py", 85),
TestFile("layers/attention/mamba/test_mamba_ssm.py", 85),
TestFile("layers/attention/mamba/test_mamba_ssm_ssd.py", 220),
TestFile("models/test_nvidia_nemotron_nano_v2.py", 180),
TestFile("test_modelopt_loader.py", 30),
TestFile("test_vlm_input_format.py", 300),
],
"per-commit-2-gpu": [
TestFile("ep/test_moe_ep.py", 140),
TestFile("hicache/test_hicache_storage_3fs_backend.py", 200),
TestFile("hicache/test_hicache_storage_file_backend.py", 200),
TestFile("hicache/test_hicache_storage_mooncake_backend.py", 400),
TestFile("layers/attention/mamba/test_mamba2_mixer.py", 110),
TestFile("lora/test_lora_tp.py", 116),
TestFile("rl/test_update_weights_from_distributed.py", 103),
TestFile("test_data_parallelism.py", 73),
TestFile("test_disaggregation_basic.py", 400),
TestFile("test_dp_attention.py", 594),
TestFile("test_load_weights_from_remote_instance.py", 72),
TestFile("test_patch_torch.py", 19),
TestFile("test_release_memory_occupation.py", 257),
TestFile("hicache/test_hicache_storage_file_backend.py", 200),
TestFile("hicache/test_hicache_storage_mooncake_backend.py", 400),
TestFile("hicache/test_hicache_storage_3fs_backend.py", 200),
TestFile("layers/attention/mamba/test_mamba2_mixer.py", 110),
],
"per-commit-4-gpu": [
TestFile("models/test_qwen3_next_models.py", 291),
TestFile("test_disaggregation_dp_attention.py", 155),
TestFile("test_gpt_oss_4gpu.py", 300),
TestFile("test_local_attn.py", 411),
TestFile("test_pp_single_node.py", 481),
TestFile("models/test_qwen3_next_models.py", 291),
TestFile("test_multi_instance_release_memory_occupation.py", 64),
TestFile("test_pp_single_node.py", 481),
],
"per-commit-8-gpu": [
TestFile("lora/test_lora_llama4.py", 400),
TestFile("test_deepseek_v3_basic.py", 275),
TestFile("test_deepseek_v3_mtp.py", 275),
TestFile("test_disaggregation_different_tp.py", 600),
TestFile("test_disaggregation_pp.py", 140),
],
"per-commit-4-gpu-b200": [
# TestFile("test_gpt_oss_4gpu.py", 600),
@@ -169,16 +167,6 @@ suites = {
"per-commit-8-gpu-deepep": [
TestFile("ep/test_deepep_large.py", 338),
],
"per-commit-2-gpu-disaggregation": [
TestFile("test_disaggregation_basic.py", 400),
],
"per-commit-4-gpu-disaggregation": [
TestFile("test_disaggregation_dp_attention.py", 155),
],
"per-commit-8-gpu-disaggregation": [
TestFile("test_disaggregation_different_tp.py", 600),
TestFile("test_disaggregation_pp.py", 140),
],
"per-commit-8-gpu-h20": [
TestFile("quant/test_w4a8_deepseek_v3.py", 371),
],
@@ -192,48 +180,46 @@ suites = {
}
# Add AMD tests
# NOTE: please sort the test cases alphabetically by the test file name
suite_amd = {
"per-commit-amd": [
TestFile("function_call/test_json_schema_constraint.py", 30),
TestFile("hicache/test_hicache.py", 116),
TestFile("hicache/test_hicache_mla.py", 127),
TestFile("hicache/test_hicache_storage.py", 127),
TestFile("lora/test_lora.py", 200),
TestFile("lora/test_lora_eviction.py", 200),
TestFile("lora/test_lora_backend.py", 99),
TestFile("lora/test_multi_lora_backend.py", 60),
TestFile("lora/test_lora_cuda_graph.py", 250),
TestFile("lora/test_lora_eviction.py", 200),
TestFile("lora/test_lora_qwen3.py", 97),
# TestFile("models/test_embedding_models.py", 73), # Disabled temporarily, see https://github.com/sgl-project/sglang/issues/11127
TestFile("lora/test_multi_lora_backend.py", 60),
TestFile("models/test_compressed_tensors_models.py", 42),
TestFile("models/test_qwen_models.py", 82),
TestFile("models/test_reward_models.py", 132),
TestFile("models/test_transformers_models.py", 320),
TestFile("openai_server/basic/test_openai_embedding.py", 141),
TestFile("openai_server/basic/test_openai_server.py", 149),
TestFile("openai_server/basic/test_protocol.py", 10),
TestFile("openai_server/basic/test_serving_chat.py", 10),
TestFile("openai_server/basic/test_serving_completions.py", 10),
TestFile("openai_server/basic/test_serving_embedding.py", 10),
TestFile("openai_server/basic/test_openai_embedding.py", 141),
TestFile("openai_server/basic/test_openai_server.py", 149),
TestFile("openai_server/features/test_enable_thinking.py", 70),
TestFile("openai_server/features/test_json_constrained.py", 98),
TestFile("openai_server/features/test_json_mode.py", 90),
TestFile("openai_server/features/test_openai_server_ebnf.py", 95),
# TestFile("openai_server/features/test_openai_server_hidden_states.py", 240),
TestFile("openai_server/features/test_reasoning_content.py", 89),
TestFile("openai_server/function_call/test_openai_function_calling.py", 60),
TestFile("openai_server/function_call/test_tool_choice.py", 226),
TestFile("function_call/test_json_schema_constraint.py", 30),
TestFile("openai_server/validation/test_large_max_new_tokens.py", 41),
TestFile("openai_server/validation/test_matched_stop.py", 60),
TestFile("openai_server/validation/test_openai_server_ignore_eos.py", 85),
TestFile("openai_server/validation/test_request_length_validation.py", 31),
TestFile("quant/test_block_int8.py", 22),
TestFile("quant/test_awq_dequant.py", 2),
TestFile("quant/test_block_int8.py", 22),
TestFile("rl/test_update_weights_from_disk.py", 114),
# TestFile("rl/test_update_weights_from_tensor.py", 48),
TestFile("test_abort.py", 51),
TestFile("test_create_kvindices.py", 2),
TestFile("test_chunked_prefill.py", 313),
TestFile("test_create_kvindices.py", 2),
TestFile("test_ebnf_constrained.py", 108),
TestFile("test_eval_fp8_accuracy.py", 303),
TestFile("test_function_call_parser.py", 10),
@@ -246,30 +232,34 @@ suite_amd = {
TestFile("test_mla.py", 242),
TestFile("test_mla_deepseek_v3.py", 221),
TestFile("test_no_chunked_prefill.py", 108),
# TestFile("test_no_overlap_scheduler.py", 234), # Disabled temporarily and track in #7703
TestFile("test_penalty.py", 41),
TestFile("test_page_size.py", 60),
TestFile("test_penalty.py", 41),
TestFile("test_pytorch_sampling_backend.py", 66),
TestFile("test_radix_attention.py", 105),
TestFile("test_reasoning_parser.py", 5),
TestFile("test_regex_constrained.py", 64),
TestFile("test_retract_decode.py", 54),
TestFile("test_reasoning_parser.py", 5),
TestFile("test_rope_rocm.py", 3),
TestFile("test_server_args.py", 1),
TestFile("test_skip_tokenizer_init.py", 117),
TestFile("test_srt_engine.py", 261),
TestFile("test_srt_endpoint.py", 130),
TestFile("test_srt_engine.py", 261),
TestFile("test_torch_compile.py", 169),
TestFile("test_torch_compile_moe.py", 172),
TestFile("test_torch_native_attention_backend.py", 123),
TestFile("test_triton_attention_backend.py", 150),
# TestFile("test_vision_chunked_prefill.py", 175), # Disabled temporarily and track in #7701
TestFile("test_wave_attention_kernels.py", 2),
# Disabled temporarily
# TestFile("models/test_embedding_models.py", 73), # Disabled temporarily, see https://github.com/sgl-project/sglang/issues/11127
# TestFile("openai_server/features/test_openai_server_hidden_states.py", 240),
# TestFile("rl/test_update_weights_from_tensor.py", 48),
# TestFile("test_no_overlap_scheduler.py", 234), # Disabled temporarily and track in #7703
# TestFile("test_vision_chunked_prefill.py", 175), # Disabled temporarily and track in #7701
# TestFile("test_wave_attention_backend.py", 150), # Disabled temporarily, see https://github.com/sgl-project/sglang/issues/11127
],
"per-commit-amd-mi35x": [
TestFile("test_mla.py", 242),
TestFile("test_gpt_oss_1gpu.py", 600),
TestFile("test_mla.py", 242),
],
"per-commit-2-gpu-amd": [
TestFile("lora/test_lora_tp.py", 116),
@@ -291,6 +281,7 @@ suite_amd = {
}
# Add Intel Xeon tests
# NOTE: please sort the test cases alphabetically by the test file name
suite_xeon = {
"per-commit-cpu": [
TestFile("cpu/test_activation.py"),
@@ -305,22 +296,23 @@ suite_xeon = {
TestFile("cpu/test_rope.py"),
TestFile("cpu/test_shared_expert.py"),
TestFile("cpu/test_topk.py"),
TestFile("test_intel_amx_attention_backend.py"),
TestFile("test_cpu_graph.py"),
TestFile("test_intel_amx_attention_backend.py"),
],
}
# Add Ascend NPU tests
# NOTE: please sort the test cases alphabetically by the test file name
suite_ascend = {
"per-commit-1-ascend-npu": [
TestFile("ascend/test_ascend_tp1_bf16.py", 400),
TestFile("ascend/test_ascend_graph_tp1_bf16.py", 400),
TestFile("ascend/test_ascend_tp1_bf16.py", 400),
],
"per-commit-2-ascend-npu": [
TestFile("ascend/test_ascend_tp2_bf16.py", 400),
TestFile("ascend/test_ascend_graph_tp2_bf16.py", 400),
TestFile("ascend/test_ascend_tp2_fia_bf16.py", 400),
TestFile("ascend/test_ascend_mla_fia_w8a8int8.py", 400),
TestFile("ascend/test_ascend_tp2_bf16.py", 400),
TestFile("ascend/test_ascend_tp2_fia_bf16.py", 400),
],
"per-commit-4-ascend-npu": [
TestFile("ascend/test_ascend_mla_w8a8int8.py", 400),

1
test/srt/test_disaggregation_basic.py
View File

@@ -1,6 +1,5 @@
import json
import os
import time
import unittest
from types import SimpleNamespace

3
test/srt/test_disaggregation_dp_attention.py
View File

@@ -1,15 +1,12 @@
import os
import time
import unittest
from types import SimpleNamespace
from urllib.parse import urlparse
from sglang.test.few_shot_gsm8k import run_eval as run_eval_few_shot_gsm8k
from sglang.test.test_disaggregation_utils import TestDisaggregationBase
from sglang.test.test_utils import (
DEFAULT_MODEL_NAME_FOR_TEST_MLA,
DEFAULT_TIMEOUT_FOR_SERVER_LAUNCH,
DEFAULT_URL_FOR_TEST,
popen_launch_pd_server,
)