# ==========================================
# ACTUAL TEST CASES
# ==========================================
test_cases:
  - name: "Kimi-K2.7-W4A8-in3.5k-out1.5k-TPOT50-0-128-32"
    model: "Eco-Tech/Kimi-K2.7-Code-w4a8"
    envs:
      HCCL_OP_EXPANSION_MODE: "AIV"
      PYTORCH_NPU_ALLOC_CONF: "expandable_segments:True"
      OMP_PROC_BIND: "false"
      OMP_NUM_THREADS: "1"
      TASK_QUEUE_ENABLE: "1"
      HCCL_BUFFSIZE: "800"
      VLLM_ASCEND_ENABLE_FLASHCOMM1: "1"
      VLLM_ASCEND_ENABLE_MLAPO: "1"
      VLLM_ASCEND_BALANCE_SCHEDULING: "1"
      VLLM_ASCEND_ENABLE_FUSED_MC2: "1"
      DYNAMIC_EPLB: "true"
      SERVER_PORT: "DEFAULT_PORT"
    server_cmd:
      - "--quantization"
      - "ascend"
      - "--port"
      - "$SERVER_PORT"
      - "--allowed-local-media-path"
      - "/"
      - "--trust-remote-code"
      - "--safetensors-load-strategy"
      - 'prefetch'
      - "--tensor-parallel-size"
      - "8"
      - "--data-parallel-size"
      - "2"
      - "--no-enable-prefix-caching"
      - "--enable-expert-parallel"
      - "--max-num-seqs"
      - "24"
      - "--max-model-len"
      - "6144"
      - "--max-num-batched-tokens"
      - "4096"
      - "--gpu-memory-utilization"
      - "0.85"
      - "--seed"
      - "42"
      - "--async-scheduling"
      - "--compilation-config"
      - '{"cudagraph_mode":"FULL_DECODE_ONLY"}'
      - "--additional-config"
      - '{"eplb_config": {"dynamic_eplb": true},"ascend_compilation_config": {"enable_static_kernel": false}}'
      - "--profiler-config"
      - '{"profiler": "torch", "torch_profiler_dir": "./vllm_profile", "torch_profiler_with_stack": true}'
      - "--mm-processor-cache-gb"
      - "0"
      - "--mm-encoder-tp-mode"
      - "data"
      - "--speculative-config"
      - '{"method": "dflash","model": "z-lab/Kimi-K2.5-DFlash", "num_speculative_tokens": 7}'
    benchmarks:
      perf:
        case_type: performance
        dataset_path: vllm-ascend/GSM8K-in3500-bs4096-prefix0-kimi
        request_conf: vllm_api_stream_chat
        dataset_conf: gsm8k/gsm8k_gen_0_shot_cot_str_perf
        num_prompts: 128
        max_out_len: 1500
        batch_size: 32
        request_rate: 0
        baseline: 1585.1248
        threshold: 0.97
  - name: "Kimi-K2.7-W4A8-in3.5k-out1.5k-TPOT20-0-128-32"
    model: "Eco-Tech/Kimi-K2.7-Code-w4a8"
    envs:
      HCCL_OP_EXPANSION_MODE: "AIV"
      PYTORCH_NPU_ALLOC_CONF: "expandable_segments:True"
      OMP_PROC_BIND: "false"
      OMP_NUM_THREADS: "1"
      TASK_QUEUE_ENABLE: "1"
      HCCL_BUFFSIZE: "800"
      VLLM_ASCEND_ENABLE_FLASHCOMM1: "1"
      VLLM_ASCEND_ENABLE_MLAPO: "1"
      VLLM_ASCEND_BALANCE_SCHEDULING: "1"
      VLLM_ASCEND_ENABLE_FUSED_MC2: "1"
      DYNAMIC_EPLB: "true"
      SERVER_PORT: "DEFAULT_PORT"
    server_cmd:
      - "--quantization"
      - "ascend"
      - "--port"
      - "$SERVER_PORT"
      - "--allowed-local-media-path"
      - "/"
      - "--trust-remote-code"
      - "--safetensors-load-strategy"
      - 'prefetch'
      - "--tensor-parallel-size"
      - "8"
      - "--data-parallel-size"
      - "2"
      - "--no-enable-prefix-caching"
      - "--enable-expert-parallel"
      - "--max-num-seqs"
      - "24"
      - "--max-model-len"
      - "6144"
      - "--max-num-batched-tokens"
      - "4096"
      - "--gpu-memory-utilization"
      - "0.85"
      - "--seed"
      - "42"
      - "--async-scheduling"
      - "--compilation-config"
      - '{"cudagraph_mode":"FULL_DECODE_ONLY"}'
      - "--additional-config"
      - '{"eplb_config": {"dynamic_eplb": true},"ascend_compilation_config": {"enable_static_kernel": false}}'
      - "--profiler-config"
      - '{"profiler": "torch", "torch_profiler_dir": "./vllm_profile", "torch_profiler_with_stack": true}'
      - "--mm-processor-cache-gb"
      - "0"
      - "--mm-encoder-tp-mode"
      - "data"
      - "--speculative-config"
      - '{"method": "dflash","model": "z-lab/Kimi-K2.5-DFlash", "num_speculative_tokens": 7}'
    benchmarks:
      perf:
        case_type: performance
        dataset_path: vllm-ascend/GSM8K-in3500-bs4096-prefix0-kimi
        request_conf: vllm_api_stream_chat
        dataset_conf: gsm8k/gsm8k_gen_0_shot_cot_str_perf
        num_prompts: 128
        max_out_len: 1500
        batch_size: 32
        request_rate: 0
        baseline: 1585.1248
        threshold: 0.97
  - name: "Kimi-K2.7-W4A8-in64k-out1k-TPOT50-90-24-6"
    model: "Eco-Tech/Kimi-K2.7-Code-w4a8"
    envs:
      HCCL_OP_EXPANSION_MODE: "AIV"
      PYTORCH_NPU_ALLOC_CONF: "expandable_segments:True"
      OMP_PROC_BIND: "false"
      OMP_NUM_THREADS: "1"
      TASK_QUEUE_ENABLE: "1"
      HCCL_BUFFSIZE: "1300"
      VLLM_ASCEND_ENABLE_FLASHCOMM1: "1"
      VLLM_ASCEND_ENABLE_MLAPO: "1"
      VLLM_ASCEND_BALANCE_SCHEDULING: "1"
      VLLM_ASCEND_ENABLE_FUSED_MC2: "1"
      DYNAMIC_EPLB: "true"
      SERVER_PORT: "DEFAULT_PORT"
    server_cmd:
      - "--quantization"
      - "ascend"
      - "--port"
      - "$SERVER_PORT"
      - "--allowed-local-media-path"
      - "/"
      - "--trust-remote-code"
      - "--safetensors-load-strategy"
      - 'prefetch'
      - "--tensor-parallel-size"
      - "8"
      - "--data-parallel-size"
      - "2"
      - "--enable-prefix-caching"
      - "--enable-expert-parallel"
      - "--max-num-seqs"
      - "4"
      - "--max-model-len"
      - "67000"
      - "--max-num-batched-tokens"
      - "8192"
      - "--gpu-memory-utilization"
      - "0.87"
      - "--seed"
      - "42"
      - "--async-scheduling"
      - "--compilation-config"
      - '{"cudagraph_mode":"FULL_DECODE_ONLY"}'
      - "--additional-config"
      - '{"eplb_config": {"dynamic_eplb": true},"ascend_compilation_config": {"enable_static_kernel": false}}'
      - "--profiler-config"
      - '{"profiler": "torch", "torch_profiler_dir": "./vllm_profile", "torch_profiler_with_stack": true}'
      - "--mm-processor-cache-gb"
      - "0"
      - "--mm-encoder-tp-mode"
      - "data"
      - "--speculative-config"
      - '{"method": "dflash","model": "z-lab/Kimi-K2.5-DFlash", "num_speculative_tokens": 7}'
    benchmarks:
      perf:
        case_type: performance
        dataset_path: vllm-ascend/GSM8K_prefix90_in65536_bs1000_kimi
        request_conf: vllm_api_stream_chat
        dataset_conf: gsm8k/gsm8k_gen_0_shot_cot_str_perf
        num_prompts: 24
        max_out_len: 1024
        batch_size: 6
        request_rate: 0
        baseline: 167.6767
        threshold: 0.97
  - name: "Kimi-K2.7-W4A8-in128k-out1k-TPOT50-90-24-6"
    model: "Eco-Tech/Kimi-K2.7-Code-w4a8"
    envs:
      HCCL_OP_EXPANSION_MODE: "AIV"
      PYTORCH_NPU_ALLOC_CONF: "expandable_segments:True"
      OMP_PROC_BIND: "false"
      OMP_NUM_THREADS: "1"
      TASK_QUEUE_ENABLE: "1"
      VLLM_ASCEND_ENABLE_MLAPO: "1"
      HCCL_BUFFSIZE: "1500"
      VLLM_ASCEND_ENABLE_FLASHCOMM1: "1"
      VLLM_ASCEND_BALANCE_SCHEDULING: "0"
      VLLM_ASCEND_ENABLE_FUSED_MC2: "1"
      SERVER_PORT: "DEFAULT_PORT"
    server_cmd:
      - "--quantization"
      - "ascend"
      - "--port"
      - "$SERVER_PORT"
      - "--allowed-local-media-path"
      - "/"
      - "--trust-remote-code"
      - "--tensor-parallel-size"
      - "8"
      - "--data-parallel-size"
      - "2"
      - "--enable-expert-parallel"
      - "--max-num-seqs"
      - "10"
      - "--max-model-len"
      - "133120"
      - "--max-num-batched-tokens"
      - "16384"
      - "--gpu-memory-utilization"
      - "0.92"
      - "--seed"
      - "42"
      - "--async-scheduling"
      - "--compilation-config"
      - '{"cudagraph_mode":"FULL_DECODE_ONLY"}'
      - "--profiler-config"
      - '{"profiler": "torch", "torch_profiler_dir": "./vllm_profile", "torch_profiler_with_stack": true}'
      - "--mm-processor-cache-gb"
      - "0"
      - "--mm-encoder-tp-mode"
      - "data"
      - "--speculative-config"
      - '{"method": "dflash","model": "z-lab/Kimi-K2.5-DFlash", "num_speculative_tokens": 15}'
    benchmarks:
      acc:
        case_type: accuracy
        dataset_path: vllm-ascend/aime2025
        request_conf: vllm_api_general_chat
        dataset_conf: aime2025/aime2025_gen_0_shot_chat_prompt
        max_out_len: 32768
        batch_size: 8
        baseline: 96.67
        threshold: 10
        temperature: 1.0
        top_p: 1
      perf:
        case_type: performance
        dataset_path: vllm-ascend/GSM8K_prefix90_in131072_bs1000_kimi
        request_conf: vllm_api_stream_chat
        dataset_conf: gsm8k/gsm8k_gen_0_shot_cot_str_perf
        num_prompts: 24
        max_out_len: 1024
        batch_size: 6
        request_rate: 0
        baseline: 95.3761
        threshold: 0.97
  - name: "Kimi-K2.7-W4A8-in254k-out1k"
    model: "Eco-Tech/Kimi-K2.7-Code-w4a8"
    envs:
      HCCL_OP_EXPANSION_MODE: "AIV"
      PYTORCH_NPU_ALLOC_CONF: "expandable_segments:True"
      OMP_PROC_BIND: "false"
      OMP_NUM_THREADS: "1"
      TASK_QUEUE_ENABLE: "1"
      VLLM_ASCEND_ENABLE_MLAPO: "1"
      HCCL_BUFFSIZE: "300"
      VLLM_ASCEND_BALANCE_SCHEDULING: "1"
      SERVER_PORT: "DEFAULT_PORT"
    server_cmd:
      - "--quantization"
      - "ascend"
      - "--port"
      - "$SERVER_PORT"
      - "--allowed-local-media-path"
      - "/"
      - "--trust-remote-code"
      - "--tensor-parallel-size"
      - "16"
      - "--data-parallel-size"
      - "1"
      - "--no-enable-prefix-caching"
      - "--enable-expert-parallel"
      - "--max-num-seqs"
      - "1"
      - "--max-model-len"
      - "262144"
      - "--max-num-batched-tokens"
      - "8192"
      - "--gpu-memory-utilization"
      - "0.95"
      - "--seed"
      - "42"
      - "--async-scheduling"
      - "--compilation-config"
      - '{"cudagraph_mode":"FULL_DECODE_ONLY"}'
      - "--profiler-config"
      - '{"profiler": "torch", "torch_profiler_dir": "./vllm_profile", "torch_profiler_with_stack": true}'
      - "--mm-processor-cache-gb"
      - "0"
      - "--mm-encoder-tp-mode"
      - "data"
    benchmarks:
      perf:
        case_type: performance
        dataset_path: vllm-ascend/GSM8K-in254k-bs16-prefix99-kimi
        request_conf: vllm_api_stream_chat
        dataset_conf: gsm8k/gsm8k_gen_0_shot_cot_str_perf
        num_prompts: 1
        max_out_len: 1024
        batch_size: 1
        request_rate: 0
        baseline: 7.5611
        threshold: 0.97