| 1 | [ |
| 2 | { |
| 3 | "test_name": "serving_llama8B_tp1_sharegpt", |
| 4 | "qps_list": [1, 4, 16, "inf"], |
| 5 | "server_environment_variables": { |
| 6 | "PT_HPU_LAZY_MODE": 1, |
| 7 | "VLLM_CONTIGUOUS_PA": 1, |
| 8 | "VLLM_DEFRAG": 1 |
| 9 | }, |
| 10 | "server_parameters": { |
| 11 | "model": "meta-llama/Meta-Llama-3.1-8B-Instruct", |
| 12 | "tensor_parallel_size": 1, |
| 13 | "disable_log_stats": "", |
| 14 | "load_format": "dummy", |
| 15 | "max-model-len": 2048, |
| 16 | "max-num-seqs": 256, |
| 17 | "async-scheduling": "" |
| 18 | }, |
| 19 | "client_parameters": { |
| 20 | "model": "meta-llama/Meta-Llama-3.1-8B-Instruct", |
| 21 | "backend": "vllm", |
| 22 | "dataset_name": "sharegpt", |
| 23 | "dataset_path": "./ShareGPT_V3_unfiltered_cleaned_split.json", |
| 24 | "temperature": 0, |
| 25 | "num_prompts": 200 |
| 26 | } |
| 27 | }, |
| 28 | { |
| 29 | "test_name": "serving_llama70B_tp4_sharegpt", |
| 30 | "qps_list": [1, 4, 16, "inf"], |
| 31 | "server_environment_variables": { |
| 32 | "PT_HPU_LAZY_MODE": 1, |
| 33 | "PT_HPU_ENABLE_LAZY_COLLECTIVES": 1, |
| 34 | "VLLM_CONTIGUOUS_PA": 1, |
| 35 | "VLLM_DEFRAG": 1 |
| 36 | }, |
| 37 | "server_parameters": { |
| 38 | "model": "meta-llama/Meta-Llama-3.1-70B-Instruct", |
| 39 | "tensor_parallel_size": 4, |
| 40 | "disable_log_stats": "", |
| 41 | "load_format": "dummy", |
| 42 | "max-model-len": 2048, |
| 43 | "max-num-seqs": 256, |
| 44 | "async-scheduling": "" |
| 45 | }, |
| 46 | "client_parameters": { |
| 47 | "model": "meta-llama/Meta-Llama-3.1-70B-Instruct", |
| 48 | "backend": "vllm", |
| 49 | "dataset_name": "sharegpt", |
| 50 | "dataset_path": "./ShareGPT_V3_unfiltered_cleaned_split.json", |
| 51 | "temperature": 0, |
| 52 | "num_prompts": 200 |
| 53 | } |
| 54 | }, |
| 55 | { |
| 56 | "test_name": "serving_mixtral8x7B_tp2_sharegpt", |
| 57 | "qps_list": [1, 4, 16, "inf"], |
| 58 | "server_environment_variables": { |
| 59 | "PT_HPU_LAZY_MODE": 1, |
| 60 | "PT_HPU_ENABLE_LAZY_COLLECTIVES": 1, |
| 61 | "VLLM_CONTIGUOUS_PA": 1, |
| 62 | "VLLM_DEFRAG": 1 |
| 63 | }, |
| 64 | "server_parameters": { |
| 65 | "model": "mistralai/Mixtral-8x7B-Instruct-v0.1", |
| 66 | "tensor_parallel_size": 2, |
| 67 | "disable_log_stats": "", |
| 68 | "load_format": "dummy", |
| 69 | "max-model-len": 2048, |
| 70 | "max-num-seqs": 256, |
| 71 | "async-scheduling": "" |
| 72 | }, |
| 73 | "client_parameters": { |
| 74 | "model": "mistralai/Mixtral-8x7B-Instruct-v0.1", |
| 75 | "backend": "vllm", |
| 76 | "dataset_name": "sharegpt", |
| 77 | "dataset_path": "./ShareGPT_V3_unfiltered_cleaned_split.json", |
| 78 | "temperature": 0, |
| 79 | "num_prompts": 200 |
| 80 | } |
| 81 | }, |
| 82 | { |
| 83 | "test_name": "serving_deepseek_r1", |
| 84 | "qps_list": [1, 4, 16, "inf"], |
| 85 | "server_environment_variables": { |
| 86 | "PT_HPU_LAZY_MODE": 1, |
| 87 | "PT_HPU_ENABLE_LAZY_COLLECTIVES": 1, |
| 88 | "VLLM_CONTIGUOUS_PA": 1, |
| 89 | "VLLM_DEFRAG": 1 |
| 90 | }, |
| 91 | "server_parameters": { |
| 92 | "model": "deepseek-ai/DeepSeek-R1", |
| 93 | "tensor_parallel_size": 8, |
| 94 | "disable_log_stats": "", |
| 95 | "load_format": "dummy", |
| 96 | "max-model-len": 2048, |
| 97 | "max-num-seqs": 200, |
| 98 | "async-scheduling": "", |
| 99 | "dtype": "bfloat16" |
| 100 | }, |
| 101 | "client_parameters": { |
| 102 | "model": "deepseek-ai/DeepSeek-R1", |
| 103 | "backend": "vllm", |
| 104 | "dataset_name": "sharegpt", |
| 105 | "dataset_path": "./ShareGPT_V3_unfiltered_cleaned_split.json", |
| 106 | "temperature": 0, |
| 107 | "num_prompts": 200 |
| 108 | } |
| 109 | }, |
| 110 | { |
| 111 | "test_name": "serving_llama4_maverick_17b128e_instruct_fp8", |
| 112 | "qps_list": [1, 4, 16, "inf"], |
| 113 | "server_environment_variables": { |
| 114 | "PT_HPU_LAZY_MODE": 1, |
| 115 | "PT_HPU_ENABLE_LAZY_COLLECTIVES": 1, |
| 116 | "VLLM_CONTIGUOUS_PA": 1, |
| 117 | "VLLM_DEFRAG": 1 |
| 118 | }, |
| 119 | "server_parameters": { |
| 120 | "model": "meta-llama/Llama-4-Maverick-17B-128E-Instruct-FP8", |
| 121 | "tensor_parallel_size": 8, |
| 122 | "disable_log_stats": "", |
| 123 | "max-model-len": 2048, |
| 124 | "max-num-seqs": 128, |
| 125 | "async-scheduling": "", |
| 126 | "enable_expert_parallel": "", |
| 127 | "max-num-batched-tokens": 4096 |
| 128 | }, |
| 129 | "client_parameters": { |
| 130 | "model": "meta-llama/Llama-4-Maverick-17B-128E-Instruct-FP8", |
| 131 | "backend": "vllm", |
| 132 | "dataset_name": "sharegpt", |
| 133 | "dataset_path": "./ShareGPT_V3_unfiltered_cleaned_split.json", |
| 134 | "temperature": 0, |
| 135 | "num_prompts": 200 |
| 136 | } |
| 137 | }, |
| 138 | { |
| 139 | "test_name": "serving_qwen3_8b", |
| 140 | "qps_list": [1, 4, 10, "inf"], |
| 141 | "server_environment_variables": { |
| 142 | "PT_HPU_LAZY_MODE": 1, |
| 143 | "PT_HPU_ENABLE_LAZY_COLLECTIVES": 1, |
| 144 | "VLLM_CONTIGUOUS_PA": 1, |
| 145 | "VLLM_DEFRAG": 1 |
| 146 | }, |
| 147 | "server_parameters": { |
| 148 | "model": "Qwen/Qwen-3-8B", |
| 149 | "tensor_parallel_size": 1, |
| 150 | "dtype": "bfloat16", |
| 151 | "disable_log_stats": "", |
| 152 | "async-scheduling": "" |
| 153 | }, |
| 154 | "client_parameters": { |
| 155 | "model": "Qwen/Qwen-3-8B", |
| 156 | "backend": "vllm", |
| 157 | "dataset_name": "sharegpt", |
| 158 | "dataset_path": "./ShareGPT_V3_unfiltered_cleaned_split.json", |
| 159 | "temperature": 0, |
| 160 | "num_prompts": 200 |
| 161 | } |
| 162 | } |
| 163 | ] |