diff --git a/test/srt/ascend/test_ascend_w8a8_quantization.py b/test/manual/ascend/test_ascend_w8a8_quantization.py similarity index 100% rename from test/srt/ascend/test_ascend_w8a8_quantization.py rename to test/manual/ascend/test_ascend_w8a8_quantization.py diff --git a/test/srt/ascend/test_mindspore_models.py b/test/manual/ascend/test_mindspore_models.py similarity index 100% rename from test/srt/ascend/test_mindspore_models.py rename to test/manual/ascend/test_mindspore_models.py diff --git a/test/srt/cpu/test_comm.py b/test/manual/cpu/test_comm.py similarity index 100% rename from test/srt/cpu/test_comm.py rename to test/manual/cpu/test_comm.py diff --git a/test/srt/debug_utils/test_log_parser.py b/test/manual/debug_utils/test_log_parser.py similarity index 100% rename from test/srt/debug_utils/test_log_parser.py rename to test/manual/debug_utils/test_log_parser.py diff --git a/test/srt/entrypoints/http_server/test_abort_request.py b/test/manual/entrypoints/http_server/test_abort_request.py similarity index 100% rename from test/srt/entrypoints/http_server/test_abort_request.py rename to test/manual/entrypoints/http_server/test_abort_request.py diff --git a/test/srt/hicache/test_disaggregation_hicache.py b/test/manual/hicache/test_disaggregation_hicache.py similarity index 100% rename from test/srt/hicache/test_disaggregation_hicache.py rename to test/manual/hicache/test_disaggregation_hicache.py diff --git a/test/srt/layers/attention/nsa/test_act_quant_triton.py b/test/manual/layers/attention/nsa/test_act_quant_triton.py similarity index 100% rename from test/srt/layers/attention/nsa/test_act_quant_triton.py rename to test/manual/layers/attention/nsa/test_act_quant_triton.py diff --git a/test/srt/layers/moe/test_moe_runners.py b/test/manual/layers/moe/test_moe_runners.py similarity index 100% rename from test/srt/layers/moe/test_moe_runners.py rename to test/manual/layers/moe/test_moe_runners.py diff --git a/test/srt/lora/test_chunked_sgmv_backend.py b/test/manual/lora/test_chunked_sgmv_backend.py similarity index 100% rename from test/srt/lora/test_chunked_sgmv_backend.py rename to test/manual/lora/test_chunked_sgmv_backend.py diff --git a/test/srt/lora/test_lora_cuda_graph.py b/test/manual/lora/test_lora_cuda_graph.py similarity index 100% rename from test/srt/lora/test_lora_cuda_graph.py rename to test/manual/lora/test_lora_cuda_graph.py diff --git a/test/srt/lora/test_lora_llama4.py b/test/manual/lora/test_lora_llama4.py similarity index 100% rename from test/srt/lora/test_lora_llama4.py rename to test/manual/lora/test_lora_llama4.py diff --git a/test/srt/lora/test_lora_qwen3_vl.py b/test/manual/lora/test_lora_qwen3_vl.py similarity index 100% rename from test/srt/lora/test_lora_qwen3_vl.py rename to test/manual/lora/test_lora_qwen3_vl.py diff --git a/test/srt/models/test_clip_models.py b/test/manual/models/test_clip_models.py similarity index 100% rename from test/srt/models/test_clip_models.py rename to test/manual/models/test_clip_models.py diff --git a/test/srt/models/test_dummy_grok_models.py b/test/manual/models/test_dummy_grok_models.py similarity index 100% rename from test/srt/models/test_dummy_grok_models.py rename to test/manual/models/test_dummy_grok_models.py diff --git a/test/srt/models/test_falcon_h1_models.py b/test/manual/models/test_falcon_h1_models.py similarity index 100% rename from test/srt/models/test_falcon_h1_models.py rename to test/manual/models/test_falcon_h1_models.py diff --git a/test/srt/models/test_gme_qwen_models.py b/test/manual/models/test_gme_qwen_models.py similarity index 100% rename from test/srt/models/test_gme_qwen_models.py rename to test/manual/models/test_gme_qwen_models.py diff --git a/test/srt/models/test_grok_models.py b/test/manual/models/test_grok_models.py similarity index 100% rename from test/srt/models/test_grok_models.py rename to test/manual/models/test_grok_models.py diff --git a/test/srt/models/test_llama4_models.py b/test/manual/models/test_llama4_models.py similarity index 100% rename from test/srt/models/test_llama4_models.py rename to test/manual/models/test_llama4_models.py diff --git a/test/srt/models/test_mtp_models.py b/test/manual/models/test_mtp_models.py similarity index 100% rename from test/srt/models/test_mtp_models.py rename to test/manual/models/test_mtp_models.py diff --git a/test/srt/models/test_unsloth_models.py b/test/manual/models/test_unsloth_models.py similarity index 100% rename from test/srt/models/test_unsloth_models.py rename to test/manual/models/test_unsloth_models.py diff --git a/test/srt/nightly/test_deepseek_v31_perf.py b/test/manual/nightly/test_deepseek_v31_perf.py similarity index 100% rename from test/srt/nightly/test_deepseek_v31_perf.py rename to test/manual/nightly/test_deepseek_v31_perf.py diff --git a/test/srt/nightly/test_deepseek_v32_perf.py b/test/manual/nightly/test_deepseek_v32_perf.py similarity index 100% rename from test/srt/nightly/test_deepseek_v32_perf.py rename to test/manual/nightly/test_deepseek_v32_perf.py diff --git a/test/srt/nightly/test_text_models_gsm8k_eval.py b/test/manual/nightly/test_text_models_gsm8k_eval.py similarity index 100% rename from test/srt/nightly/test_text_models_gsm8k_eval.py rename to test/manual/nightly/test_text_models_gsm8k_eval.py diff --git a/test/srt/nightly/test_text_models_perf.py b/test/manual/nightly/test_text_models_perf.py similarity index 100% rename from test/srt/nightly/test_text_models_perf.py rename to test/manual/nightly/test_text_models_perf.py diff --git a/test/srt/nightly/test_vlms_mmmu_eval.py b/test/manual/nightly/test_vlms_mmmu_eval.py similarity index 100% rename from test/srt/nightly/test_vlms_mmmu_eval.py rename to test/manual/nightly/test_vlms_mmmu_eval.py diff --git a/test/srt/nightly/test_vlms_perf.py b/test/manual/nightly/test_vlms_perf.py similarity index 100% rename from test/srt/nightly/test_vlms_perf.py rename to test/manual/nightly/test_vlms_perf.py diff --git a/test/srt/nightly/test_vlms_piecewise_cuda_graph.py b/test/manual/nightly/test_vlms_piecewise_cuda_graph.py similarity index 100% rename from test/srt/nightly/test_vlms_piecewise_cuda_graph.py rename to test/manual/nightly/test_vlms_piecewise_cuda_graph.py diff --git a/test/srt/openai_server/features/test_cache_report.py b/test/manual/openai_server/features/test_cache_report.py similarity index 100% rename from test/srt/openai_server/features/test_cache_report.py rename to test/manual/openai_server/features/test_cache_report.py diff --git a/test/srt/openai_server/features/test_continuous_usage_stats.py b/test/manual/openai_server/features/test_continuous_usage_stats.py similarity index 100% rename from test/srt/openai_server/features/test_continuous_usage_stats.py rename to test/manual/openai_server/features/test_continuous_usage_stats.py diff --git a/test/srt/openai_server/features/test_structural_tag.py b/test/manual/openai_server/features/test_structural_tag.py similarity index 100% rename from test/srt/openai_server/features/test_structural_tag.py rename to test/manual/openai_server/features/test_structural_tag.py diff --git a/test/srt/quant/test_fp8_kvcache.py b/test/manual/quant/test_fp8_kvcache.py similarity index 100% rename from test/srt/quant/test_fp8_kvcache.py rename to test/manual/quant/test_fp8_kvcache.py diff --git a/test/srt/rl/test_verl_engine_2_gpu.py b/test/manual/rl/test_verl_engine_2_gpu.py similarity index 100% rename from test/srt/rl/test_verl_engine_2_gpu.py rename to test/manual/rl/test_verl_engine_2_gpu.py diff --git a/test/srt/rl/test_verl_engine_4_gpu.py b/test/manual/rl/test_verl_engine_4_gpu.py similarity index 100% rename from test/srt/rl/test_verl_engine_4_gpu.py rename to test/manual/rl/test_verl_engine_4_gpu.py diff --git a/test/srt/test_async_dynamic_batch_tokenizer.py b/test/manual/test_async_dynamic_batch_tokenizer.py similarity index 100% rename from test/srt/test_async_dynamic_batch_tokenizer.py rename to test/manual/test_async_dynamic_batch_tokenizer.py diff --git a/test/srt/test_async_mm_data_processor.py b/test/manual/test_async_mm_data_processor.py similarity index 100% rename from test/srt/test_async_mm_data_processor.py rename to test/manual/test_async_mm_data_processor.py diff --git a/test/srt/test_config_integration.py b/test/manual/test_config_integration.py similarity index 100% rename from test/srt/test_config_integration.py rename to test/manual/test_config_integration.py diff --git a/test/srt/test_custom_allreduce.py b/test/manual/test_custom_allreduce.py similarity index 100% rename from test/srt/test_custom_allreduce.py rename to test/manual/test_custom_allreduce.py diff --git a/test/srt/test_deepseek_chat_templates.py b/test/manual/test_deepseek_chat_templates.py similarity index 100% rename from test/srt/test_deepseek_chat_templates.py rename to test/manual/test_deepseek_chat_templates.py diff --git a/test/srt/test_deepseek_v32_cp_single_node.py b/test/manual/test_deepseek_v32_cp_single_node.py similarity index 100% rename from test/srt/test_deepseek_v32_cp_single_node.py rename to test/manual/test_deepseek_v32_cp_single_node.py diff --git a/test/srt/test_deepseek_v3_cutedsl_4gpu.py b/test/manual/test_deepseek_v3_cutedsl_4gpu.py similarity index 100% rename from test/srt/test_deepseek_v3_cutedsl_4gpu.py rename to test/manual/test_deepseek_v3_cutedsl_4gpu.py diff --git a/test/srt/test_double_sparsity.py b/test/manual/test_double_sparsity.py similarity index 100% rename from test/srt/test_double_sparsity.py rename to test/manual/test_double_sparsity.py diff --git a/test/srt/test_eagle_infer_beta_dp_attention.py b/test/manual/test_eagle_infer_beta_dp_attention.py similarity index 100% rename from test/srt/test_eagle_infer_beta_dp_attention.py rename to test/manual/test_eagle_infer_beta_dp_attention.py diff --git a/test/srt/test_expert_distribution.py b/test/manual/test_expert_distribution.py similarity index 100% rename from test/srt/test_expert_distribution.py rename to test/manual/test_expert_distribution.py diff --git a/test/srt/test_expert_location_updater.py b/test/manual/test_expert_location_updater.py similarity index 100% rename from test/srt/test_expert_location_updater.py rename to test/manual/test_expert_location_updater.py diff --git a/test/srt/test_fim_completion.py b/test/manual/test_fim_completion.py similarity index 100% rename from test/srt/test_fim_completion.py rename to test/manual/test_fim_completion.py diff --git a/test/srt/test_forward_split_prefill.py b/test/manual/test_forward_split_prefill.py similarity index 100% rename from test/srt/test_forward_split_prefill.py rename to test/manual/test_forward_split_prefill.py diff --git a/test/srt/test_get_weights_by_name.py b/test/manual/test_get_weights_by_name.py similarity index 100% rename from test/srt/test_get_weights_by_name.py rename to test/manual/test_get_weights_by_name.py diff --git a/test/srt/test_health_check.py b/test/manual/test_health_check.py similarity index 100% rename from test/srt/test_health_check.py rename to test/manual/test_health_check.py diff --git a/test/srt/test_kv_events.py b/test/manual/test_kv_events.py similarity index 100% rename from test/srt/test_kv_events.py rename to test/manual/test_kv_events.py diff --git a/test/srt/test_logprobs.py b/test/manual/test_logprobs.py similarity index 100% rename from test/srt/test_logprobs.py rename to test/manual/test_logprobs.py diff --git a/test/srt/test_mla_tp.py b/test/manual/test_mla_tp.py similarity index 100% rename from test/srt/test_mla_tp.py rename to test/manual/test_mla_tp.py diff --git a/test/srt/test_modelopt.py b/test/manual/test_modelopt.py similarity index 100% rename from test/srt/test_modelopt.py rename to test/manual/test_modelopt.py diff --git a/test/srt/test_modelopt_fp8kvcache.py b/test/manual/test_modelopt_fp8kvcache.py similarity index 100% rename from test/srt/test_modelopt_fp8kvcache.py rename to test/manual/test_modelopt_fp8kvcache.py diff --git a/test/srt/test_models_from_modelscope.py b/test/manual/test_models_from_modelscope.py similarity index 100% rename from test/srt/test_models_from_modelscope.py rename to test/manual/test_models_from_modelscope.py diff --git a/test/srt/test_mscclpp.py b/test/manual/test_mscclpp.py similarity index 100% rename from test/srt/test_mscclpp.py rename to test/manual/test_mscclpp.py diff --git a/test/srt/test_quick_allreduce.py b/test/manual/test_quick_allreduce.py similarity index 100% rename from test/srt/test_quick_allreduce.py rename to test/manual/test_quick_allreduce.py diff --git a/test/srt/test_sagemaker_server.py b/test/manual/test_sagemaker_server.py similarity index 100% rename from test/srt/test_sagemaker_server.py rename to test/manual/test_sagemaker_server.py diff --git a/test/srt/test_schedule_policy.py b/test/manual/test_schedule_policy.py similarity index 100% rename from test/srt/test_schedule_policy.py rename to test/manual/test_schedule_policy.py diff --git a/test/srt/test_session_control.py b/test/manual/test_session_control.py similarity index 100% rename from test/srt/test_session_control.py rename to test/manual/test_session_control.py diff --git a/test/srt/test_srt_engine_with_quant_args.py b/test/manual/test_srt_engine_with_quant_args.py similarity index 100% rename from test/srt/test_srt_engine_with_quant_args.py rename to test/manual/test_srt_engine_with_quant_args.py diff --git a/test/srt/test_tokenizer_batch_encode.py b/test/manual/test_tokenizer_batch_encode.py similarity index 100% rename from test/srt/test_tokenizer_batch_encode.py rename to test/manual/test_tokenizer_batch_encode.py diff --git a/test/srt/test_tokenizer_manager.py b/test/manual/test_tokenizer_manager.py similarity index 100% rename from test/srt/test_tokenizer_manager.py rename to test/manual/test_tokenizer_manager.py diff --git a/test/srt/test_torch_flex_attention_backend.py b/test/manual/test_torch_flex_attention_backend.py similarity index 100% rename from test/srt/test_torch_flex_attention_backend.py rename to test/manual/test_torch_flex_attention_backend.py diff --git a/test/srt/test_torch_tp.py b/test/manual/test_torch_tp.py similarity index 100% rename from test/srt/test_torch_tp.py rename to test/manual/test_torch_tp.py diff --git a/test/srt/test_tracing.py b/test/manual/test_tracing.py similarity index 100% rename from test/srt/test_tracing.py rename to test/manual/test_tracing.py diff --git a/test/srt/test_triton_attention_rocm_mla.py b/test/manual/test_triton_attention_rocm_mla.py similarity index 100% rename from test/srt/test_triton_attention_rocm_mla.py rename to test/manual/test_triton_attention_rocm_mla.py diff --git a/test/srt/test_triton_fused_moe.py b/test/manual/test_triton_fused_moe.py similarity index 100% rename from test/srt/test_triton_fused_moe.py rename to test/manual/test_triton_fused_moe.py diff --git a/test/srt/test_triton_moe_wna16.py b/test/manual/test_triton_moe_wna16.py similarity index 100% rename from test/srt/test_triton_moe_wna16.py rename to test/manual/test_triton_moe_wna16.py diff --git a/test/srt/test_two_batch_overlap.py b/test/manual/test_two_batch_overlap.py similarity index 100% rename from test/srt/test_two_batch_overlap.py rename to test/manual/test_two_batch_overlap.py diff --git a/test/srt/test_vertex_endpoint.py b/test/manual/test_vertex_endpoint.py similarity index 100% rename from test/srt/test_vertex_endpoint.py rename to test/manual/test_vertex_endpoint.py diff --git a/test/srt/test_vlm_accuracy.py b/test/manual/test_vlm_accuracy.py similarity index 100% rename from test/srt/test_vlm_accuracy.py rename to test/manual/test_vlm_accuracy.py diff --git a/test/srt/test_wave_attention_backend.py b/test/manual/test_wave_attention_backend.py similarity index 100% rename from test/srt/test_wave_attention_backend.py rename to test/manual/test_wave_attention_backend.py diff --git a/test/srt/test_weight_version.py b/test/manual/test_weight_version.py similarity index 100% rename from test/srt/test_weight_version.py rename to test/manual/test_weight_version.py diff --git a/test/srt/run_suite.py b/test/srt/run_suite.py index 1bcde43a1..964b06ec3 100644 --- a/test/srt/run_suite.py +++ b/test/srt/run_suite.py @@ -230,86 +230,12 @@ suites = { ], "nightly-8-gpu-h20": [], "__not_in_ci__": [ - TestFile("ascend/test_ascend_w8a8_quantization.py"), - TestFile("ascend/test_mindspore_models.py"), - TestFile("cpu/test_comm.py"), - TestFile("debug_utils/test_log_parser.py", 5), - TestFile("test_deepseek_v3_cutedsl_4gpu.py"), - TestFile("entrypoints/http_server/test_abort_request.py"), - TestFile("hicache/test_disaggregation_hicache.py"), - TestFile("layers/attention/nsa/test_act_quant_triton.py"), - TestFile("layers/moe/test_moe_runners.py"), - TestFile("lora/test_chunked_sgmv_backend.py"), - TestFile("lora/test_lora_llama4.py"), - TestFile("lora/test_lora_cuda_graph.py"), - TestFile("lora/test_lora_qwen3_vl.py"), - TestFile("models/test_clip_models.py"), - TestFile("models/test_dummy_grok_models.py"), - TestFile("models/test_falcon_h1_models.py"), - TestFile("models/test_gme_qwen_models.py"), - TestFile("models/test_grok_models.py"), - TestFile("models/test_llama4_models.py"), - TestFile("models/test_mtp_models.py"), - TestFile("models/test_unsloth_models.py"), - TestFile("openai_server/features/test_cache_report.py"), - TestFile("openai_server/features/test_continuous_usage_stats.py"), - TestFile("openai_server/features/test_structural_tag.py"), - TestFile("quant/test_fp8_kvcache.py"), - TestFile("rl/test_verl_engine_2_gpu.py"), - TestFile("rl/test_verl_engine_4_gpu.py"), - TestFile("test_async_dynamic_batch_tokenizer.py"), - TestFile("test_async_mm_data_processor.py"), TestFile("test_bench_one_batch.py"), TestFile("test_bench_serving.py"), - TestFile("test_config_integration.py"), - TestFile("test_custom_allreduce.py"), - TestFile("test_deepseek_chat_templates.py"), - TestFile("test_double_sparsity.py"), - TestFile("test_eagle_infer_beta_dp_attention.py"), TestFile("test_eval_accuracy_large.py"), - TestFile("test_expert_distribution.py"), - TestFile("test_expert_location_updater.py"), - TestFile("test_fim_completion.py"), - TestFile("test_forward_split_prefill.py"), - TestFile("test_get_weights_by_name.py"), TestFile("test_gpt_oss_common.py"), - TestFile("test_health_check.py"), - TestFile("test_kv_events.py"), - TestFile("test_logprobs.py"), - TestFile("test_mla_tp.py"), - TestFile("test_modelopt.py"), - TestFile("test_modelopt_fp8kvcache.py"), - TestFile("test_models_from_modelscope.py"), TestFile("test_moe_eval_accuracy_large.py"), - TestFile("test_mscclpp.py"), - TestFile("nightly/test_deepseek_v31_perf.py"), - TestFile("nightly/test_deepseek_v32_perf.py"), - TestFile("nightly/test_text_models_gsm8k_eval.py"), - TestFile("nightly/test_text_models_perf.py"), - TestFile("nightly/test_vlms_mmmu_eval.py"), - TestFile("nightly/test_vlms_perf.py"), - TestFile("nightly/test_vlms_piecewise_cuda_graph.py"), - TestFile("test_quick_allreduce.py"), - TestFile("test_sagemaker_server.py"), - TestFile("test_schedule_policy.py"), - TestFile("test_session_control.py"), - TestFile("test_srt_engine_with_quant_args.py"), - TestFile("test_tokenizer_batch_encode.py"), - TestFile("test_tokenizer_manager.py"), - TestFile("test_torch_flex_attention_backend.py"), - TestFile("test_torch_tp.py"), - TestFile("test_tracing.py"), - TestFile("test_triton_attention_rocm_mla.py"), - TestFile("test_triton_fused_moe.py"), - TestFile("test_triton_moe_wna16.py"), - TestFile("test_two_batch_overlap.py"), - TestFile("test_vertex_endpoint.py"), - # TestFile("test_vision_openai_server_a.py"), # TODO: Fix timeout TestFile("test_vision_openai_server_common.py"), - TestFile("test_vlm_accuracy.py"), - TestFile("test_wave_attention_backend.py"), - TestFile("test_weight_version.py"), - TestFile("test_deepseek_v32_cp_single_node.py", 275), ], }