| 1 | [ |
| 2 | { |
| 3 | "test_name": "llama8B_tp1_sharegpt", |
| 4 | "qps_list": [4,8,16,32,"inf"], |
| 5 | "common_parameters": { |
| 6 | "model": "meta-llama/Meta-Llama-3-8B-Instruct", |
| 7 | "tp": 1, |
| 8 | "dataset_name": "sharegpt", |
| 9 | "dataset_path": "./ShareGPT_V3_unfiltered_cleaned_split.json", |
| 10 | "num_prompts": 500, |
| 11 | "port": 8000, |
| 12 | "reuse_server": false |
| 13 | }, |
| 14 | "lmdeploy_server_parameters": { |
| 15 | "dtype": "bfloat16" |
| 16 | }, |
| 17 | "lmdeploy_client_parameters": { |
| 18 | }, |
| 19 | "tgi_server_parameters": { |
| 20 | }, |
| 21 | "tgi_client_parameters": { |
| 22 | "endpoint": "/generate_stream" |
| 23 | }, |
| 24 | "trt_server_parameters": { |
| 25 | "model_type": "llama", |
| 26 | "model_dtype": "bfloat16", |
| 27 | "max_batch_size": 2048, |
| 28 | "max_input_len": 4096, |
| 29 | "max_seq_len": 6144, |
| 30 | "max_num_tokens": 16384, |
| 31 | "trt_llm_version": "v0.11.0" |
| 32 | }, |
| 33 | "trt_client_parameters": { |
| 34 | "endpoint": "/v2/models/ensemble/generate_stream" |
| 35 | }, |
| 36 | "vllm_server_parameters": { |
| 37 | "disable_log_stats": "", |
| 38 | "gpu_memory_utilization": 0.9, |
| 39 | "max_num_seqs": 512, |
| 40 | "dtype": "bfloat16" |
| 41 | }, |
| 42 | "vllm_client_parameters": { |
| 43 | }, |
| 44 | "sglang_server_parameters": { |
| 45 | "disable_radix_cache": "", |
| 46 | "enable_torch_compile": "", |
| 47 | "dtype": "bfloat16" |
| 48 | }, |
| 49 | "sglang_client_parameters": { |
| 50 | } |
| 51 | }, |
| 52 | { |
| 53 | "test_name": "llama8B_tp1_sonnet_512_16", |
| 54 | "qps_list": [4,8,16,32,"inf"], |
| 55 | "common_parameters": { |
| 56 | "model": "meta-llama/Meta-Llama-3-8B-Instruct", |
| 57 | "tp": 1, |
| 58 | "dataset_name": "sonnet", |
| 59 | "dataset_path": "./sonnet_4x.txt", |
| 60 | "num_prompts": 500, |
| 61 | "port": 8000, |
| 62 | "sonnet_input_len": 512, |
| 63 | "sonnet_output_len": 16, |
| 64 | "sonnet_prefix_len": 50, |
| 65 | "reuse_server": true |
| 66 | }, |
| 67 | "lmdeploy_server_parameters": { |
| 68 | "dtype": "bfloat16" |
| 69 | }, |
| 70 | "lmdeploy_client_parameters": { |
| 71 | }, |
| 72 | "tgi_server_parameters": { |
| 73 | }, |
| 74 | "tgi_client_parameters": { |
| 75 | "endpoint": "/generate_stream" |
| 76 | }, |
| 77 | "trt_server_parameters": { |
| 78 | "model_type": "llama", |
| 79 | "model_dtype": "bfloat16", |
| 80 | "max_batch_size": 2048, |
| 81 | "max_input_len": 4096, |
| 82 | "max_seq_len": 6144, |
| 83 | "max_num_tokens": 16384, |
| 84 | "trt_llm_version": "v0.11.0" |
| 85 | }, |
| 86 | "trt_client_parameters": { |
| 87 | "endpoint": "/v2/models/ensemble/generate_stream" |
| 88 | }, |
| 89 | "vllm_server_parameters": { |
| 90 | "disable_log_stats": "", |
| 91 | "gpu_memory_utilization": 0.9, |
| 92 | "max_num_seqs": 512, |
| 93 | "dtype": "bfloat16" |
| 94 | }, |
| 95 | "vllm_client_parameters": { |
| 96 | }, |
| 97 | "sglang_server_parameters": { |
| 98 | "disable_radix_cache": "", |
| 99 | "enable_torch_compile": "", |
| 100 | "dtype": "bfloat16" |
| 101 | }, |
| 102 | "sglang_client_parameters": { |
| 103 | } |
| 104 | }, |
| 105 | { |
| 106 | "test_name": "llama8B_tp1_sonnet_512_256", |
| 107 | "qps_list": [4,8,16,32,"inf"], |
| 108 | "common_parameters": { |
| 109 | "model": "meta-llama/Meta-Llama-3-8B-Instruct", |
| 110 | "tp": 1, |
| 111 | "dataset_name": "sonnet", |
| 112 | "dataset_path": "./sonnet_4x.txt", |
| 113 | "num_prompts": 500, |
| 114 | "port": 8000, |
| 115 | "sonnet_input_len": 512, |
| 116 | "sonnet_output_len": 256, |
| 117 | "sonnet_prefix_len": 50, |
| 118 | "reuse_server": true |
| 119 | }, |
| 120 | "lmdeploy_server_parameters": { |
| 121 | "dtype": "bfloat16" |
| 122 | }, |
| 123 | "lmdeploy_client_parameters": { |
| 124 | }, |
| 125 | "tgi_server_parameters": { |
| 126 | }, |
| 127 | "tgi_client_parameters": { |
| 128 | "endpoint": "/generate_stream" |
| 129 | }, |
| 130 | "trt_server_parameters": { |
| 131 | "model_type": "llama", |
| 132 | "model_dtype": "bfloat16", |
| 133 | "max_batch_size": 2048, |
| 134 | "max_input_len": 4096, |
| 135 | "max_seq_len": 6144, |
| 136 | "max_num_tokens": 16384, |
| 137 | "trt_llm_version": "v0.11.0" |
| 138 | }, |
| 139 | "trt_client_parameters": { |
| 140 | "endpoint": "/v2/models/ensemble/generate_stream" |
| 141 | }, |
| 142 | "vllm_server_parameters": { |
| 143 | "disable_log_stats": "", |
| 144 | "gpu_memory_utilization": 0.9, |
| 145 | "max_num_seqs": 512, |
| 146 | "dtype": "bfloat16" |
| 147 | }, |
| 148 | "vllm_client_parameters": { |
| 149 | }, |
| 150 | "sglang_server_parameters": { |
| 151 | "disable_radix_cache": "", |
| 152 | "enable_torch_compile": "", |
| 153 | "dtype": "bfloat16" |
| 154 | }, |
| 155 | "sglang_client_parameters": { |
| 156 | } |
| 157 | }, |
| 158 | { |
| 159 | "test_name": "llama70B_tp4_sharegpt", |
| 160 | "qps_list": [4,8,16,32,"inf"], |
| 161 | "common_parameters": { |
| 162 | "model": "meta-llama/Meta-Llama-3-70B-Instruct", |
| 163 | "tp": 4, |
| 164 | "dataset_name": "sharegpt", |
| 165 | "dataset_path": "./ShareGPT_V3_unfiltered_cleaned_split.json", |
| 166 | "num_prompts": 500, |
| 167 | "port": 8000, |
| 168 | "reuse_server": false |
| 169 | }, |
| 170 | "lmdeploy_server_parameters": { |
| 171 | "dtype": "bfloat16" |
| 172 | }, |
| 173 | "lmdeploy_client_parameters": { |
| 174 | }, |
| 175 | "tgi_server_parameters": { |
| 176 | }, |
| 177 | "tgi_client_parameters": { |
| 178 | "endpoint": "/generate_stream" |
| 179 | }, |
| 180 | "trt_server_parameters": { |
| 181 | "model_type": "llama", |
| 182 | "model_dtype": "bfloat16", |
| 183 | "max_batch_size": 2048, |
| 184 | "max_input_len": 4096, |
| 185 | "max_seq_len": 6144, |
| 186 | "max_num_tokens": 16384, |
| 187 | "trt_llm_version": "v0.11.0" |
| 188 | }, |
| 189 | "trt_client_parameters": { |
| 190 | "endpoint": "/v2/models/ensemble/generate_stream" |
| 191 | }, |
| 192 | "vllm_server_parameters": { |
| 193 | "disable_log_stats": "", |
| 194 | "gpu_memory_utilization": 0.9, |
| 195 | "max_num_seqs": 512, |
| 196 | "dtype": "bfloat16" |
| 197 | }, |
| 198 | "vllm_client_parameters": { |
| 199 | }, |
| 200 | "sglang_server_parameters": { |
| 201 | "disable_radix_cache": "", |
| 202 | "dtype": "bfloat16" |
| 203 | }, |
| 204 | "sglang_client_parameters": { |
| 205 | } |
| 206 | }, |
| 207 | { |
| 208 | "test_name": "llama70B_tp4_sonnet_512_16", |
| 209 | "qps_list": [4,8,16,32,"inf"], |
| 210 | "common_parameters": { |
| 211 | "model": "meta-llama/Meta-Llama-3-70B-Instruct", |
| 212 | "tp": 4, |
| 213 | "dataset_name": "sonnet", |
| 214 | "dataset_path": "./sonnet_4x.txt", |
| 215 | "num_prompts": 500, |
| 216 | "port": 8000, |
| 217 | "sonnet_input_len": 512, |
| 218 | "sonnet_output_len": 16, |
| 219 | "sonnet_prefix_len": 50, |
| 220 | "reuse_server": true |
| 221 | }, |
| 222 | "lmdeploy_server_parameters": { |
| 223 | "dtype": "bfloat16" |
| 224 | }, |
| 225 | "lmdeploy_client_parameters": { |
| 226 | }, |
| 227 | "tgi_server_parameters": { |
| 228 | }, |
| 229 | "tgi_client_parameters": { |
| 230 | "endpoint": "/generate_stream" |
| 231 | }, |
| 232 | "trt_server_parameters": { |
| 233 | "model_type": "llama", |
| 234 | "model_dtype": "bfloat16", |
| 235 | "max_batch_size": 2048, |
| 236 | "max_input_len": 4096, |
| 237 | "max_seq_len": 6144, |
| 238 | "max_num_tokens": 16384, |
| 239 | "trt_llm_version": "v0.11.0" |
| 240 | }, |
| 241 | "trt_client_parameters": { |
| 242 | "endpoint": "/v2/models/ensemble/generate_stream" |
| 243 | }, |
| 244 | "vllm_server_parameters": { |
| 245 | "disable_log_stats": "", |
| 246 | "gpu_memory_utilization": 0.9, |
| 247 | "max_num_seqs": 512, |
| 248 | "dtype": "bfloat16" |
| 249 | }, |
| 250 | "vllm_client_parameters": { |
| 251 | }, |
| 252 | "sglang_server_parameters": { |
| 253 | "disable_radix_cache": "", |
| 254 | "dtype": "bfloat16" |
| 255 | }, |
| 256 | "sglang_client_parameters": { |
| 257 | } |
| 258 | }, |
| 259 | { |
| 260 | "test_name": "llama70B_tp4_sonnet_512_256", |
| 261 | "qps_list": [4,8,16,32,"inf"], |
| 262 | "common_parameters": { |
| 263 | "model": "meta-llama/Meta-Llama-3-70B-Instruct", |
| 264 | "tp": 4, |
| 265 | "dataset_name": "sonnet", |
| 266 | "dataset_path": "./sonnet_4x.txt", |
| 267 | "num_prompts": 500, |
| 268 | "port": 8000, |
| 269 | "sonnet_input_len": 512, |
| 270 | "sonnet_output_len": 256, |
| 271 | "sonnet_prefix_len": 50, |
| 272 | "reuse_server": true |
| 273 | }, |
| 274 | "lmdeploy_server_parameters": { |
| 275 | "dtype": "bfloat16" |
| 276 | }, |
| 277 | "lmdeploy_client_parameters": { |
| 278 | }, |
| 279 | "tgi_server_parameters": { |
| 280 | }, |
| 281 | "tgi_client_parameters": { |
| 282 | "endpoint": "/generate_stream" |
| 283 | }, |
| 284 | "trt_server_parameters": { |
| 285 | "model_type": "llama", |
| 286 | "model_dtype": "bfloat16", |
| 287 | "max_batch_size": 2048, |
| 288 | "max_input_len": 4096, |
| 289 | "max_seq_len": 6144, |
| 290 | "max_num_tokens": 16384, |
| 291 | "trt_llm_version": "v0.11.0" |
| 292 | }, |
| 293 | "trt_client_parameters": { |
| 294 | "endpoint": "/v2/models/ensemble/generate_stream" |
| 295 | }, |
| 296 | "vllm_server_parameters": { |
| 297 | "disable_log_stats": "", |
| 298 | "gpu_memory_utilization": 0.9, |
| 299 | "max_num_seqs": 512, |
| 300 | "dtype": "bfloat16" |
| 301 | }, |
| 302 | "vllm_client_parameters": { |
| 303 | }, |
| 304 | "sglang_server_parameters": { |
| 305 | "disable_radix_cache": "", |
| 306 | "dtype": "bfloat16" |
| 307 | }, |
| 308 | "sglang_client_parameters": { |
| 309 | } |
| 310 | } |
| 311 | ] |