diff --git a/dev/bench/data.js b/dev/bench/data.js index ac69d16..1127571 100644 --- a/dev/bench/data.js +++ b/dev/bench/data.js @@ -1,5 +1,5 @@ window.BENCHMARK_DATA = { - "lastUpdate": 1723152082686, + "lastUpdate": 1723170521476, "repoUrl": "https://github.com/neuralmagic/nm-vllm-ent", "entries": { "smaller_is_better": [ @@ -1980,6 +1980,52 @@ window.BENCHMARK_DATA = { "extra": "{\n \"description\": \"VLLM Serving - Dense\\nmodel - meta-llama/Meta-Llama-3-8B-Instruct\\nmax-model-len - 4096\\nsparsity - None\\nbenchmark_serving {\\n \\\"nr-qps-pair_\\\": \\\"300,1\\\",\\n \\\"dataset\\\": \\\"sharegpt\\\"\\n}\",\n \"benchmarking_context\": {\n \"vllm_version\": \"0.5.2.0\",\n \"python_version\": \"3.10.12 (main, Jun 7 2023, 13:43:11) [GCC 11.3.0]\",\n \"torch_version\": \"2.3.1+cu121\",\n \"torch_cuda_version\": \"12.1\",\n \"cuda_devices\": \"[_CudaDeviceProperties(name='NVIDIA H100 80GB HBM3', major=9, minor=0, total_memory=81116MB, multi_processor_count=132)]\",\n \"cuda_device_names\": [\n \"NVIDIA H100 80GB HBM3\"\n ]\n },\n \"gpu_description\": \"NVIDIA H100 80GB HBM3 x 1\",\n \"script_name\": \"benchmark_serving.py\",\n \"script_args\": {\n \"description\": \"VLLM Serving - Dense\\nmodel - meta-llama/Meta-Llama-3-8B-Instruct\\nmax-model-len - 4096\\nsparsity - None\\nbenchmark_serving {\\n \\\"nr-qps-pair_\\\": \\\"300,1\\\",\\n \\\"dataset\\\": \\\"sharegpt\\\"\\n}\",\n \"backend\": \"vllm\",\n \"version\": \"N/A\",\n \"base_url\": null,\n \"host\": \"127.0.0.1\",\n \"port\": 9000,\n \"endpoint\": \"/generate\",\n \"dataset\": \"sharegpt\",\n \"num_input_tokens\": null,\n \"num_output_tokens\": null,\n \"model\": \"meta-llama/Meta-Llama-3-8B-Instruct\",\n \"tokenizer\": \"meta-llama/Meta-Llama-3-8B-Instruct\",\n \"best_of\": 1,\n \"use_beam_search\": false,\n \"log_model_io\": false,\n \"seed\": 0,\n \"trust_remote_code\": false,\n \"disable_tqdm\": false,\n \"save_directory\": \"benchmark-results\",\n \"num_prompts_\": null,\n \"request_rate_\": null,\n \"nr_qps_pair_\": [\n 300,\n \"1.0\"\n ],\n \"server_tensor_parallel_size\": 1,\n \"server_args\": \"{'model': 'meta-llama/Meta-Llama-3-8B-Instruct', 'tokenizer': 'meta-llama/Meta-Llama-3-8B-Instruct', 'max-model-len': 4096, 'host': '127.0.0.1', 'port': 9000, 'tensor-parallel-size': 1, 'disable-log-requests': ''}\"\n },\n \"date\": \"2024-08-08 21:20:08 UTC\",\n \"model\": \"meta-llama/Meta-Llama-3-8B-Instruct\",\n \"dataset\": \"sharegpt\"\n}" } ] + }, + { + "commit": { + "author": { + "name": "Derek Kozikowski", + "username": "derekk-nm", + "email": "106621615+derekk-nm@users.noreply.github.com" + }, + "committer": { + "name": "GitHub", + "username": "web-flow", + "email": "noreply@github.com" + }, + "id": "63734105d31ab6019461319dc49e42b6a9579c15", + "message": "relax relative tolerance for lm-eval (#47)\n\nSUMMARY:\r\n\r\nthe observation is that on gcp-k8s-l4-duo instances, the gsm8k task\r\nmetric values have a wider range than seen on gcp-k8s-l4-solo instances.\r\ntemporarily relax the tolerance for the `isclose` check until we\r\nimplement custom settings.\r\n\r\nTEST PLAN:\r\nusing results from various runs yesterday and today, I calculated the\r\nRTOL necessary to get the measured value of any metric to fall w/in a\r\ntolerance of the ground truth. 0.040 was enough to capture the widest\r\ndifference.", + "timestamp": "2024-08-08T20:03:33Z", + "url": "https://github.com/neuralmagic/nm-vllm-ent/commit/63734105d31ab6019461319dc49e42b6a9579c15" + }, + "date": 1723170521274, + "tool": "customSmallerIsBetter", + "benches": [ + { + "name": "{\"name\": \"mean_ttft_ms\", \"description\": \"VLLM Serving - Dense\\nmodel - meta-llama/Meta-Llama-3-8B-Instruct\\nmax-model-len - 4096\\nsparsity - None\\nbenchmark_serving {\\n \\\"nr-qps-pair_\\\": \\\"300,1\\\",\\n \\\"dataset\\\": \\\"sharegpt\\\"\\n}\", \"gpu_description\": \"NVIDIA L4 x 1\", \"vllm_version\": \"0.5.2.0\", \"python_version\": \"3.10.12 (main, Jun 7 2023, 13:43:11) [GCC 11.3.0]\", \"torch_version\": \"2.3.1+cu121\"}", + "value": 173.77672133667755, + "unit": "ms", + "extra": "{\n \"description\": \"VLLM Serving - Dense\\nmodel - meta-llama/Meta-Llama-3-8B-Instruct\\nmax-model-len - 4096\\nsparsity - None\\nbenchmark_serving {\\n \\\"nr-qps-pair_\\\": \\\"300,1\\\",\\n \\\"dataset\\\": \\\"sharegpt\\\"\\n}\",\n \"benchmarking_context\": {\n \"vllm_version\": \"0.5.2.0\",\n \"python_version\": \"3.10.12 (main, Jun 7 2023, 13:43:11) [GCC 11.3.0]\",\n \"torch_version\": \"2.3.1+cu121\",\n \"torch_cuda_version\": \"12.1\",\n \"cuda_devices\": \"[_CudaDeviceProperties(name='NVIDIA L4', major=8, minor=9, total_memory=22478MB, multi_processor_count=58)]\",\n \"cuda_device_names\": [\n \"NVIDIA L4\"\n ]\n },\n \"gpu_description\": \"NVIDIA L4 x 1\",\n \"script_name\": \"benchmark_serving.py\",\n \"script_args\": {\n \"description\": \"VLLM Serving - Dense\\nmodel - meta-llama/Meta-Llama-3-8B-Instruct\\nmax-model-len - 4096\\nsparsity - None\\nbenchmark_serving {\\n \\\"nr-qps-pair_\\\": \\\"300,1\\\",\\n \\\"dataset\\\": \\\"sharegpt\\\"\\n}\",\n \"backend\": \"vllm\",\n \"version\": \"N/A\",\n \"base_url\": null,\n \"host\": \"127.0.0.1\",\n \"port\": 9000,\n \"endpoint\": \"/generate\",\n \"dataset\": \"sharegpt\",\n \"num_input_tokens\": null,\n \"num_output_tokens\": null,\n \"model\": \"meta-llama/Meta-Llama-3-8B-Instruct\",\n \"tokenizer\": \"meta-llama/Meta-Llama-3-8B-Instruct\",\n \"best_of\": 1,\n \"use_beam_search\": false,\n \"log_model_io\": false,\n \"seed\": 0,\n \"trust_remote_code\": false,\n \"disable_tqdm\": false,\n \"save_directory\": \"benchmark-results\",\n \"num_prompts_\": null,\n \"request_rate_\": null,\n \"nr_qps_pair_\": [\n 300,\n \"1.0\"\n ],\n \"server_tensor_parallel_size\": 1,\n \"server_args\": \"{'model': 'meta-llama/Meta-Llama-3-8B-Instruct', 'tokenizer': 'meta-llama/Meta-Llama-3-8B-Instruct', 'max-model-len': 4096, 'host': '127.0.0.1', 'port': 9000, 'tensor-parallel-size': 1, 'disable-log-requests': ''}\"\n },\n \"date\": \"2024-08-09 02:27:34 UTC\",\n \"model\": \"meta-llama/Meta-Llama-3-8B-Instruct\",\n \"dataset\": \"sharegpt\"\n}" + }, + { + "name": "{\"name\": \"mean_tpot_ms\", \"description\": \"VLLM Serving - Dense\\nmodel - meta-llama/Meta-Llama-3-8B-Instruct\\nmax-model-len - 4096\\nsparsity - None\\nbenchmark_serving {\\n \\\"nr-qps-pair_\\\": \\\"300,1\\\",\\n \\\"dataset\\\": \\\"sharegpt\\\"\\n}\", \"gpu_description\": \"NVIDIA L4 x 1\", \"vllm_version\": \"0.5.2.0\", \"python_version\": \"3.10.12 (main, Jun 7 2023, 13:43:11) [GCC 11.3.0]\", \"torch_version\": \"2.3.1+cu121\"}", + "value": 83.68109886580959, + "unit": "ms", + "extra": "{\n \"description\": \"VLLM Serving - Dense\\nmodel - meta-llama/Meta-Llama-3-8B-Instruct\\nmax-model-len - 4096\\nsparsity - None\\nbenchmark_serving {\\n \\\"nr-qps-pair_\\\": \\\"300,1\\\",\\n \\\"dataset\\\": \\\"sharegpt\\\"\\n}\",\n \"benchmarking_context\": {\n \"vllm_version\": \"0.5.2.0\",\n \"python_version\": \"3.10.12 (main, Jun 7 2023, 13:43:11) [GCC 11.3.0]\",\n \"torch_version\": \"2.3.1+cu121\",\n \"torch_cuda_version\": \"12.1\",\n \"cuda_devices\": \"[_CudaDeviceProperties(name='NVIDIA L4', major=8, minor=9, total_memory=22478MB, multi_processor_count=58)]\",\n \"cuda_device_names\": [\n \"NVIDIA L4\"\n ]\n },\n \"gpu_description\": \"NVIDIA L4 x 1\",\n \"script_name\": \"benchmark_serving.py\",\n \"script_args\": {\n \"description\": \"VLLM Serving - Dense\\nmodel - meta-llama/Meta-Llama-3-8B-Instruct\\nmax-model-len - 4096\\nsparsity - None\\nbenchmark_serving {\\n \\\"nr-qps-pair_\\\": \\\"300,1\\\",\\n \\\"dataset\\\": \\\"sharegpt\\\"\\n}\",\n \"backend\": \"vllm\",\n \"version\": \"N/A\",\n \"base_url\": null,\n \"host\": \"127.0.0.1\",\n \"port\": 9000,\n \"endpoint\": \"/generate\",\n \"dataset\": \"sharegpt\",\n \"num_input_tokens\": null,\n \"num_output_tokens\": null,\n \"model\": \"meta-llama/Meta-Llama-3-8B-Instruct\",\n \"tokenizer\": \"meta-llama/Meta-Llama-3-8B-Instruct\",\n \"best_of\": 1,\n \"use_beam_search\": false,\n \"log_model_io\": false,\n \"seed\": 0,\n \"trust_remote_code\": false,\n \"disable_tqdm\": false,\n \"save_directory\": \"benchmark-results\",\n \"num_prompts_\": null,\n \"request_rate_\": null,\n \"nr_qps_pair_\": [\n 300,\n \"1.0\"\n ],\n \"server_tensor_parallel_size\": 1,\n \"server_args\": \"{'model': 'meta-llama/Meta-Llama-3-8B-Instruct', 'tokenizer': 'meta-llama/Meta-Llama-3-8B-Instruct', 'max-model-len': 4096, 'host': '127.0.0.1', 'port': 9000, 'tensor-parallel-size': 1, 'disable-log-requests': ''}\"\n },\n \"date\": \"2024-08-09 02:27:34 UTC\",\n \"model\": \"meta-llama/Meta-Llama-3-8B-Instruct\",\n \"dataset\": \"sharegpt\"\n}" + }, + { + "name": "{\"name\": \"mean_ttft_ms\", \"description\": \"VLLM Serving - Dense\\nmodel - facebook/opt-350m\\nmax-model-len - 2048\\nsparsity - None\\nbenchmark_serving {\\n \\\"nr-qps-pair_\\\": \\\"300,1\\\",\\n \\\"dataset\\\": \\\"sharegpt\\\"\\n}\", \"gpu_description\": \"NVIDIA L4 x 1\", \"vllm_version\": \"0.5.2.0\", \"python_version\": \"3.10.12 (main, Jun 7 2023, 13:43:11) [GCC 11.3.0]\", \"torch_version\": \"2.3.1+cu121\"}", + "value": 23.218442933333183, + "unit": "ms", + "extra": "{\n \"description\": \"VLLM Serving - Dense\\nmodel - facebook/opt-350m\\nmax-model-len - 2048\\nsparsity - None\\nbenchmark_serving {\\n \\\"nr-qps-pair_\\\": \\\"300,1\\\",\\n \\\"dataset\\\": \\\"sharegpt\\\"\\n}\",\n \"benchmarking_context\": {\n \"vllm_version\": \"0.5.2.0\",\n \"python_version\": \"3.10.12 (main, Jun 7 2023, 13:43:11) [GCC 11.3.0]\",\n \"torch_version\": \"2.3.1+cu121\",\n \"torch_cuda_version\": \"12.1\",\n \"cuda_devices\": \"[_CudaDeviceProperties(name='NVIDIA L4', major=8, minor=9, total_memory=22478MB, multi_processor_count=58)]\",\n \"cuda_device_names\": [\n \"NVIDIA L4\"\n ]\n },\n \"gpu_description\": \"NVIDIA L4 x 1\",\n \"script_name\": \"benchmark_serving.py\",\n \"script_args\": {\n \"description\": \"VLLM Serving - Dense\\nmodel - facebook/opt-350m\\nmax-model-len - 2048\\nsparsity - None\\nbenchmark_serving {\\n \\\"nr-qps-pair_\\\": \\\"300,1\\\",\\n \\\"dataset\\\": \\\"sharegpt\\\"\\n}\",\n \"backend\": \"vllm\",\n \"version\": \"N/A\",\n \"base_url\": null,\n \"host\": \"127.0.0.1\",\n \"port\": 9000,\n \"endpoint\": \"/generate\",\n \"dataset\": \"sharegpt\",\n \"num_input_tokens\": null,\n \"num_output_tokens\": null,\n \"model\": \"facebook/opt-350m\",\n \"tokenizer\": \"facebook/opt-350m\",\n \"best_of\": 1,\n \"use_beam_search\": false,\n \"log_model_io\": false,\n \"seed\": 0,\n \"trust_remote_code\": false,\n \"disable_tqdm\": false,\n \"save_directory\": \"benchmark-results\",\n \"num_prompts_\": null,\n \"request_rate_\": null,\n \"nr_qps_pair_\": [\n 300,\n \"1.0\"\n ],\n \"server_tensor_parallel_size\": 1,\n \"server_args\": \"{'model': 'facebook/opt-350m', 'tokenizer': 'facebook/opt-350m', 'max-model-len': 2048, 'host': '127.0.0.1', 'port': 9000, 'tensor-parallel-size': 1, 'disable-log-requests': ''}\"\n },\n \"date\": \"2024-08-09 02:18:26 UTC\",\n \"model\": \"facebook/opt-350m\",\n \"dataset\": \"sharegpt\"\n}" + }, + { + "name": "{\"name\": \"mean_tpot_ms\", \"description\": \"VLLM Serving - Dense\\nmodel - facebook/opt-350m\\nmax-model-len - 2048\\nsparsity - None\\nbenchmark_serving {\\n \\\"nr-qps-pair_\\\": \\\"300,1\\\",\\n \\\"dataset\\\": \\\"sharegpt\\\"\\n}\", \"gpu_description\": \"NVIDIA L4 x 1\", \"vllm_version\": \"0.5.2.0\", \"python_version\": \"3.10.12 (main, Jun 7 2023, 13:43:11) [GCC 11.3.0]\", \"torch_version\": \"2.3.1+cu121\"}", + "value": 6.069495508058525, + "unit": "ms", + "extra": "{\n \"description\": \"VLLM Serving - Dense\\nmodel - facebook/opt-350m\\nmax-model-len - 2048\\nsparsity - None\\nbenchmark_serving {\\n \\\"nr-qps-pair_\\\": \\\"300,1\\\",\\n \\\"dataset\\\": \\\"sharegpt\\\"\\n}\",\n \"benchmarking_context\": {\n \"vllm_version\": \"0.5.2.0\",\n \"python_version\": \"3.10.12 (main, Jun 7 2023, 13:43:11) [GCC 11.3.0]\",\n \"torch_version\": \"2.3.1+cu121\",\n \"torch_cuda_version\": \"12.1\",\n \"cuda_devices\": \"[_CudaDeviceProperties(name='NVIDIA L4', major=8, minor=9, total_memory=22478MB, multi_processor_count=58)]\",\n \"cuda_device_names\": [\n \"NVIDIA L4\"\n ]\n },\n \"gpu_description\": \"NVIDIA L4 x 1\",\n \"script_name\": \"benchmark_serving.py\",\n \"script_args\": {\n \"description\": \"VLLM Serving - Dense\\nmodel - facebook/opt-350m\\nmax-model-len - 2048\\nsparsity - None\\nbenchmark_serving {\\n \\\"nr-qps-pair_\\\": \\\"300,1\\\",\\n \\\"dataset\\\": \\\"sharegpt\\\"\\n}\",\n \"backend\": \"vllm\",\n \"version\": \"N/A\",\n \"base_url\": null,\n \"host\": \"127.0.0.1\",\n \"port\": 9000,\n \"endpoint\": \"/generate\",\n \"dataset\": \"sharegpt\",\n \"num_input_tokens\": null,\n \"num_output_tokens\": null,\n \"model\": \"facebook/opt-350m\",\n \"tokenizer\": \"facebook/opt-350m\",\n \"best_of\": 1,\n \"use_beam_search\": false,\n \"log_model_io\": false,\n \"seed\": 0,\n \"trust_remote_code\": false,\n \"disable_tqdm\": false,\n \"save_directory\": \"benchmark-results\",\n \"num_prompts_\": null,\n \"request_rate_\": null,\n \"nr_qps_pair_\": [\n 300,\n \"1.0\"\n ],\n \"server_tensor_parallel_size\": 1,\n \"server_args\": \"{'model': 'facebook/opt-350m', 'tokenizer': 'facebook/opt-350m', 'max-model-len': 2048, 'host': '127.0.0.1', 'port': 9000, 'tensor-parallel-size': 1, 'disable-log-requests': ''}\"\n },\n \"date\": \"2024-08-09 02:18:26 UTC\",\n \"model\": \"facebook/opt-350m\",\n \"dataset\": \"sharegpt\"\n}" + } + ] } ] }