Skip to content

GitLab

Unverified Commit 548a57b1 authored by Lianmin Zheng's avatar Lianmin Zheng Committed by GitHub
Browse files

Fix port conflicts in CI (#11497)

parent 88e73ed0
Hide whitespace changes
Inline Side-by-side
Showing with 34 additions and 336 deletions
.github/workflows/pr-test.yml
View file @ 548a57b1
...@@ -327,7 +327,7 @@ jobs: ...@@ -327,7 +327,7 @@ jobs:
strategy: strategy:
fail-fast: false fail-fast: false
matrix: matrix:
part: [0, 1, 2] part: [0, 1]
steps: steps:
- name: Checkout code - name: Checkout code
uses: actions/checkout@v4 uses: actions/checkout@v4
...@@ -348,7 +348,7 @@ jobs: ...@@ -348,7 +348,7 @@ jobs:
timeout-minutes: 20 timeout-minutes: 20
run: | run: |
cd test/srt cd test/srt
python3 run_suite.py --suite per-commit-8-gpu --auto-partition-id ${{ matrix.part }} --auto-partition-size 3 python3 run_suite.py --suite per-commit-8-gpu --auto-partition-id ${{ matrix.part }} --auto-partition-size 2
performance-test-1-gpu-part-1: performance-test-1-gpu-part-1:
needs: [check-changes, sgl-kernel-build-wheels] needs: [check-changes, sgl-kernel-build-wheels]
......
python/pyproject.toml
View file @ 548a57b1
...@@ -76,6 +76,7 @@ decord = ["decord"] ...@@ -76,6 +76,7 @@ decord = ["decord"]
test = [ test = [
"accelerate", "accelerate",
"expecttest", "expecttest",
"gguf",
"jsonlines", "jsonlines",
"matplotlib", "matplotlib",
"pandas", "pandas",
...@@ -85,10 +86,10 @@ test = [ ...@@ -85,10 +86,10 @@ test = [
"tabulate", "tabulate",
] ]
tracing = [ tracing = [
"opentelemetry-api", "opentelemetry-api",
"opentelemetry-exporter-otlp", "opentelemetry-exporter-otlp",
"opentelemetry-exporter-otlp-proto-grpc", "opentelemetry-exporter-otlp-proto-grpc",
"opentelemetry-sdk", "opentelemetry-sdk",
] ]
all = ["sglang[test]", "sglang[decord]"] all = ["sglang[test]", "sglang[decord]"]
all_aarch64 = ["sglang[test]"] all_aarch64 = ["sglang[test]"]
......
python/sglang/test/test_utils.py
View file @ 548a57b1
...@@ -135,11 +135,11 @@ def _use_cached_default_models(model_repo: str): ...@@ -135,11 +135,11 @@ def _use_cached_default_models(model_repo: str):
if is_in_ci(): if is_in_ci():
DEFAULT_PORT_FOR_SRT_TEST_RUNNER = ( DEFAULT_PORT_FOR_SRT_TEST_RUNNER = (
5000 + int(os.environ.get("CUDA_VISIBLE_DEVICES", "0")[0]) * 100 10000 + int(os.environ.get("CUDA_VISIBLE_DEVICES", "0")[0]) * 1000
) )
else: else:
DEFAULT_PORT_FOR_SRT_TEST_RUNNER = ( DEFAULT_PORT_FOR_SRT_TEST_RUNNER = (
7000 + int(os.environ.get("CUDA_VISIBLE_DEVICES", "0")[0]) * 100 20000 + int(os.environ.get("CUDA_VISIBLE_DEVICES", "0")[0]) * 1000
) )
DEFAULT_URL_FOR_TEST = f"http://127.0.0.1:{DEFAULT_PORT_FOR_SRT_TEST_RUNNER + 1000}" DEFAULT_URL_FOR_TEST = f"http://127.0.0.1:{DEFAULT_PORT_FOR_SRT_TEST_RUNNER + 1000}"
...@@ -396,8 +396,6 @@ def _get_call_generate(args: argparse.Namespace): ...@@ -396,8 +396,6 @@ def _get_call_generate(args: argparse.Namespace):
return partial(call_generate_vllm, url=f"{args.host}:{args.port}/generate") return partial(call_generate_vllm, url=f"{args.host}:{args.port}/generate")
elif args.backend == "srt-raw": elif args.backend == "srt-raw":
return partial(call_generate_srt_raw, url=f"{args.host}:{args.port}/generate") return partial(call_generate_srt_raw, url=f"{args.host}:{args.port}/generate")
elif args.backend == "gserver":
return partial(call_generate_gserver, url=f"{args.host}:{args.port}")
elif args.backend == "outlines": elif args.backend == "outlines":
return partial(call_generate_outlines, url=f"{args.host}:{args.port}/generate") return partial(call_generate_outlines, url=f"{args.host}:{args.port}/generate")
elif args.backend == "guidance": elif args.backend == "guidance":
...@@ -521,7 +519,6 @@ def popen_launch_server( ...@@ -521,7 +519,6 @@ def popen_launch_server(
# Auto-detect device if needed # Auto-detect device if needed
if device == "auto": if device == "auto":
device = auto_config_device() device = auto_config_device()
print(f"Auto-configed device: {device}", flush=True)
other_args = list(other_args) other_args = list(other_args)
other_args += ["--device", str(device)] other_args += ["--device", str(device)]
......
scripts/sort_testcases_alphabetically.py
View file @ 548a57b1
...@@ -11,320 +11,7 @@ class TestFile: ...@@ -11,320 +11,7 @@ class TestFile:
estimated_time: float = 60 estimated_time: float = 60
suites = { 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_eagle_infer_beta.py", 300),
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__": if __name__ == "__main__":
......
test/srt/run_suite.py
View file @ 548a57b1
...@@ -79,7 +79,7 @@ suites = { ...@@ -79,7 +79,7 @@ suites = {
TestFile("test_gpt_oss_1gpu.py", 600), TestFile("test_gpt_oss_1gpu.py", 600),
TestFile("test_harmony_parser.py", 20), TestFile("test_harmony_parser.py", 20),
TestFile("test_hidden_states.py", 55), TestFile("test_hidden_states.py", 55),
TestFile("test_hybrid_attn_backend.py", 100), TestFile("test_hybrid_attn_backend.py", 379),
TestFile("test_input_embeddings.py", 38), TestFile("test_input_embeddings.py", 38),
TestFile("test_io_struct.py", 8), TestFile("test_io_struct.py", 8),
TestFile("test_jinja_template_utils.py", 1), TestFile("test_jinja_template_utils.py", 1),
......
test/srt/test_mla_int8_deepseek_v3.py
View file @ 548a57b1
...@@ -22,7 +22,15 @@ class TestMLADeepseekV3ChannelInt8(CustomTestCase): ...@@ -22,7 +22,15 @@ class TestMLADeepseekV3ChannelInt8(CustomTestCase):
cls.base_url = DEFAULT_URL_FOR_TEST cls.base_url = DEFAULT_URL_FOR_TEST
other_args = ["--trust-remote-code"] other_args = ["--trust-remote-code"]
if torch.cuda.is_available() and torch.version.cuda: if torch.cuda.is_available() and torch.version.cuda:
other_args.extend(["--enable-torch-compile", "--cuda-graph-max-bs", "2"]) other_args.extend(
[
"--cuda-graph-max-bs",
"16",
"--enable-torch-compile",
"--torch-compile-max-bs",
"2",
]
)
cls.process = popen_launch_server( cls.process = popen_launch_server(
cls.model, cls.model,
cls.base_url, cls.base_url,
...@@ -50,6 +58,7 @@ class TestMLADeepseekV3ChannelInt8(CustomTestCase): ...@@ -50,6 +58,7 @@ class TestMLADeepseekV3ChannelInt8(CustomTestCase):
self.assertGreaterEqual(metrics["accuracy"], 0.61) self.assertGreaterEqual(metrics["accuracy"], 0.61)
@unittest.skipIf(is_in_ci(), "To reduce the CI execution time.")
class TestDeepseekV3MTPChannelInt8(CustomTestCase): class TestDeepseekV3MTPChannelInt8(CustomTestCase):
@classmethod @classmethod
def setUpClass(cls): def setUpClass(cls):
...@@ -60,12 +69,10 @@ class TestDeepseekV3MTPChannelInt8(CustomTestCase): ...@@ -60,12 +69,10 @@ class TestDeepseekV3MTPChannelInt8(CustomTestCase):
other_args.extend( other_args.extend(
[ [
"--cuda-graph-max-bs", "--cuda-graph-max-bs",
"2", "16",
"--disable-radix",
"--enable-torch-compile", "--enable-torch-compile",
"--torch-compile-max-bs", "--torch-compile-max-bs",
"1", "2" "--speculative-algorithm",
"--speculative-algorithm",
"EAGLE", "EAGLE",
"--speculative-draft-model-path", "--speculative-draft-model-path",
"sgl-project/sglang-ci-dsv3-channel-int8-test-NextN", "sgl-project/sglang-ci-dsv3-channel-int8-test-NextN",
...@@ -121,7 +128,15 @@ class TestMLADeepseekV3BlockInt8(CustomTestCase): ...@@ -121,7 +128,15 @@ class TestMLADeepseekV3BlockInt8(CustomTestCase):
cls.base_url = DEFAULT_URL_FOR_TEST cls.base_url = DEFAULT_URL_FOR_TEST
other_args = ["--trust-remote-code"] other_args = ["--trust-remote-code"]
if torch.cuda.is_available() and torch.version.cuda: if torch.cuda.is_available() and torch.version.cuda:
other_args.extend(["--enable-torch-compile", "--cuda-graph-max-bs", "2"]) other_args.extend(
[
"--cuda-graph-max-bs",
"16",
"--enable-torch-compile",
"--torch-compile-max-bs",
"2",
]
)
cls.process = popen_launch_server( cls.process = popen_launch_server(
cls.model, cls.model,
cls.base_url, cls.base_url,
...@@ -159,12 +174,10 @@ class TestDeepseekV3MTPBlockInt8(CustomTestCase): ...@@ -159,12 +174,10 @@ class TestDeepseekV3MTPBlockInt8(CustomTestCase):
other_args.extend( other_args.extend(
[ [
"--cuda-graph-max-bs", "--cuda-graph-max-bs",
"2", "16",
"--disable-radix",
"--enable-torch-compile", "--enable-torch-compile",
"--torch-compile-max-bs", "--torch-compile-max-bs",
"1", "2" "--speculative-algorithm",
"--speculative-algorithm",
"EAGLE", "EAGLE",
"--speculative-num-steps", "--speculative-num-steps",
"2", "2",
......
Markdown is supported
0% or .
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment