diff --git a/dev/bench/data.js b/dev/bench/data.js index 35f82e0..43ad4c6 100644 --- a/dev/bench/data.js +++ b/dev/bench/data.js @@ -1,54 +1,8 @@ window.BENCHMARK_DATA = { - "lastUpdate": 1728379716970, + "lastUpdate": 1728382178568, "repoUrl": "https://github.com/neuralmagic/nm-vllm-ent", "entries": { "smaller_is_better": [ - { - "commit": { - "author": { - "name": "Andy Linfoot", - "username": "andy-neuma", - "email": "78757007+andy-neuma@users.noreply.github.com" - }, - "committer": { - "name": "GitHub", - "username": "web-flow", - "email": "noreply@github.com" - }, - "id": "d31523d6c981fea0dc356a394ba30de57318fa95", - "message": "have \"install whl action\" return whl info (#78)\n\nSUMMARY:\r\n* have the \"install whl action\" return whl info about vllm and\r\nmagic_wand\r\n* update \"nm test\" to use the new values in the workflow summary\r\n\r\nTEST PLAN:\r\nruns on remote push\r\n\r\nCo-authored-by: andy-neuma ", - "timestamp": "2024-09-24T18:27:57Z", - "url": "https://github.com/neuralmagic/nm-vllm-ent/commit/d31523d6c981fea0dc356a394ba30de57318fa95" - }, - "date": 1727489824796, - "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 H100 80GB HBM3 x 1\", \"vllm_version\": \"0.5.4.0\", \"python_version\": \"3.10.12 (main, Jun 7 2023, 13:43:11) [GCC 11.3.0]\", \"torch_version\": \"2.3.1+cu121\"}", - "value": 33.68554593529552, - "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\": \"0.5.4.0\",\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\": \"2024-09-28 02:15:53 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 H100 80GB HBM3 x 1\", \"vllm_version\": \"0.5.4.0\", \"python_version\": \"3.10.12 (main, Jun 7 2023, 13:43:11) [GCC 11.3.0]\", \"torch_version\": \"2.3.1+cu121\"}", - "value": 11.868966910670112, - "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\": \"0.5.4.0\",\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\": \"2024-09-28 02:15:53 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 H100 80GB HBM3 x 1\", \"vllm_version\": \"0.5.4.0\", \"python_version\": \"3.10.12 (main, Jun 7 2023, 13:43:11) [GCC 11.3.0]\", \"torch_version\": \"2.3.1+cu121\"}", - "value": 41.427275429790214, - "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\": \"0.5.4.0\",\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 - 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\": \"2024-09-28 02:07:33 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 H100 80GB HBM3 x 1\", \"vllm_version\": \"0.5.4.0\", \"python_version\": \"3.10.12 (main, Jun 7 2023, 13:43:11) [GCC 11.3.0]\", \"torch_version\": \"2.3.1+cu121\"}", - "value": 7.500525367565746, - "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\": \"0.5.4.0\",\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 - 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\": \"2024-09-28 02:07:33 UTC\",\n \"model\": \"facebook/opt-350m\",\n \"dataset\": \"sharegpt\"\n}" - } - ] - }, { "commit": { "author": { @@ -2302,6 +2256,52 @@ window.BENCHMARK_DATA = { "extra": "{\n \"description\": \"VLLM Serving - Dense\\nmodel - meta-llama/Meta-Llama-3-70B-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\": \"0.5.3.1.dev16+g77db673c\",\n \"python_version\": \"3.10.12 (main, Sep 30 2024, 21:31:31) [GCC 11.4.0]\",\n \"torch_version\": \"2.4.0+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), _CudaDeviceProperties(name='NVIDIA H100 80GB HBM3', major=9, minor=0, total_memory=81116MB, multi_processor_count=132), _CudaDeviceProperties(name='NVIDIA H100 80GB HBM3', major=9, minor=0, total_memory=81116MB, multi_processor_count=132), _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 \"NVIDIA H100 80GB HBM3\",\n \"NVIDIA H100 80GB HBM3\",\n \"NVIDIA H100 80GB HBM3\"\n ]\n },\n \"gpu_description\": \"NVIDIA H100 80GB HBM3 x 4\",\n \"script_name\": \"benchmark_serving.py\",\n \"script_args\": {\n \"description\": \"VLLM Serving - Dense\\nmodel - meta-llama/Meta-Llama-3-70B-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-3-70B-Instruct\",\n \"tokenizer\": \"meta-llama/Meta-Llama-3-70B-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\": 4,\n \"server_args\": \"{'model': 'meta-llama/Meta-Llama-3-70B-Instruct', 'tokenizer': 'meta-llama/Meta-Llama-3-70B-Instruct', 'max-model-len': 4096, 'host': '127.0.0.1', 'port': 9000, 'tensor-parallel-size': 4, 'disable-log-requests': ''}\"\n },\n \"date\": \"2024-10-08 09:06:17 UTC\",\n \"model\": \"meta-llama/Meta-Llama-3-70B-Instruct\",\n \"dataset\": \"sharegpt\"\n}" } ] + }, + { + "commit": { + "author": { + "name": "Andy Linfoot", + "username": "andy-neuma", + "email": "78757007+andy-neuma@users.noreply.github.com" + }, + "committer": { + "name": "GitHub", + "username": "web-flow", + "email": "noreply@github.com" + }, + "id": "77db673c67ce6410e68939fd1ff8db6065988524", + "message": "enable more debugging in automation (#84)\n\nSUMMARY:\r\n* \"build\", \"test\", \"benchmark\", and \"lm-eval\" workflows have been\r\nupdated to make them callable. this will allow us to run through a\r\nsingle stage in automation provided we have assets from a build job. in\r\nparticular, making \"test\" workflow callable enables not having to sit\r\nthrough a build. this along with the skip mechanism shortens the\r\ndeveloper loop while we work through patches in `tests` directory.\r\n* adding helper script to invoke: build, test, benchmark, and lm-evel.\r\nthe scripts should be sufficient for running larger experiments via\r\nscripting.\r\n* install python via \"set python\" in \"debug workflow\" and add option to\r\nspecify python installed via \"set python\"\r\n\r\nTEST PLAN:\r\ndogfooding and runs on remote push\r\n\r\nnm-build.yml ...\r\nhttps://github.com/neuralmagic/nm-vllm-ent/actions/runs/11182839300\r\n\r\nthe following use the `run_id` from the above build\r\n\r\nnm-benchmark.yml ...\r\nhttps://github.com/neuralmagic/nm-vllm-ent/actions/runs/11183464277\r\n\r\nnm-lm-eval.yml ...\r\nhttps://github.com/neuralmagic/nm-vllm-ent/actions/runs/11184027633\r\n\r\nnm-test.yml ...\r\nhttps://github.com/neuralmagic/nm-vllm-ent/actions/runs/11193109513\r\n\r\nhere is the usage from each script\r\n\r\n```\r\n> ./.github/scripts/invoke-build -h\r\nUsage: ./.github/scripts/invoke-build \r\n\r\nall have standard defaults\r\n\r\n -h - this list of options\r\n -w - wf_category\r\n -l - runner label to use\r\n -t - timeout in minutes\r\n -r - gitref\r\n -m - GI to reserve per thread\r\n -n - number of 'nvcc' threads\r\n -p - python version to run while building\r\n -b - branch to use in '--ref '\r\n\r\n```\r\n\r\n```\r\n> ./.github/scripts/invoke-test -h\r\nUsage: ./.github/scripts/invoke-test \r\n\r\nunless noted have standard defaults. triggering\r\nnm-test with this script will not push results to Testmo\r\n\r\n -h - this list of options\r\n -l - runner label to use\r\n -t - timeout in minutes\r\n -r - gitref\r\n -p - python version to run while building\r\n -i - GHA run_id (required)\r\n -c - test skip env vars config file\r\n -v - enable code coverage report (true or false)\r\n -b - branch to use in '--ref '\r\n\r\n```\r\n\r\n```\r\n> ./.github/scripts/invoke-benchmark -h\r\nUsage: ./.github/scripts/invoke-benchmark \r\n\r\nunless noted have standard defaults. triggering\r\nbenchmarks with this script will not push results to 'gh_pages'\r\n\r\n -h - this list of options\r\n -l - runner label to use\r\n -c - benchmark config list file\r\n -t - timeout in minutes\r\n -r - gitref\r\n -p - python version to run while building\r\n -i - GHA run_id (required)\r\n -b - branch to use in '--ref '\r\n\r\n```\r\n\r\n```\r\n> ./.github/scripts/invoke-lm-eval -h\r\nUsage: ./.github/scripts/invoke-lm-eval \r\n\r\nunless noted have standard defaults.\r\n\r\n -h - this list of options\r\n -l - runner label to use\r\n -c - lm-eval config file\r\n -t - timeout in minutes\r\n -r - gitref\r\n -p - python version to run while building\r\n -i - GHA run_id (required)\r\n -b - branch to use in '--ref '\r\n\r\n```\r\n\r\n---------\r\n\r\nCo-authored-by: andy-neuma ", + "timestamp": "2024-10-05T16:35:23Z", + "url": "https://github.com/neuralmagic/nm-vllm-ent/commit/77db673c67ce6410e68939fd1ff8db6065988524" + }, + "date": 1728382177873, + "tool": "customSmallerIsBetter", + "benches": [ + { + "name": "{\"name\": \"mean_ttft_ms\", \"description\": \"VLLM Serving - Dense\\nmodel - mistralai/Mixtral-8x7B-Instruct-v0.1\\nmax-model-len - 4096\\nsparsity - None\\nbenchmark_serving {\\n \\\"nr-qps-pair_\\\": \\\"300,1\\\",\\n \\\"dataset\\\": \\\"sharegpt\\\"\\n}\", \"gpu_description\": \"NVIDIA A100-SXM4-80GB x 4\", \"vllm_version\": \"0.5.3.1.dev16+g77db673c\", \"python_version\": \"3.10.12 (main, Sep 30 2024, 21:31:31) [GCC 11.4.0]\", \"torch_version\": \"2.4.0+cu121\"}", + "value": 408.93799188857275, + "unit": "ms", + "extra": "{\n \"description\": \"VLLM Serving - Dense\\nmodel - mistralai/Mixtral-8x7B-Instruct-v0.1\\nmax-model-len - 4096\\nsparsity - None\\nbenchmark_serving {\\n \\\"nr-qps-pair_\\\": \\\"300,1\\\",\\n \\\"dataset\\\": \\\"sharegpt\\\"\\n}\",\n \"benchmarking_context\": {\n \"vllm_version\": \"0.5.3.1.dev16+g77db673c\",\n \"python_version\": \"3.10.12 (main, Sep 30 2024, 21:31:31) [GCC 11.4.0]\",\n \"torch_version\": \"2.4.0+cu121\",\n \"torch_cuda_version\": \"12.1\",\n \"cuda_devices\": \"[_CudaDeviceProperties(name='NVIDIA A100-SXM4-80GB', major=8, minor=0, total_memory=81049MB, multi_processor_count=108), _CudaDeviceProperties(name='NVIDIA A100-SXM4-80GB', major=8, minor=0, total_memory=81049MB, multi_processor_count=108), _CudaDeviceProperties(name='NVIDIA A100-SXM4-80GB', major=8, minor=0, total_memory=81049MB, multi_processor_count=108), _CudaDeviceProperties(name='NVIDIA A100-SXM4-80GB', major=8, minor=0, total_memory=81049MB, multi_processor_count=108)]\",\n \"cuda_device_names\": [\n \"NVIDIA A100-SXM4-80GB\",\n \"NVIDIA A100-SXM4-80GB\",\n \"NVIDIA A100-SXM4-80GB\",\n \"NVIDIA A100-SXM4-80GB\"\n ]\n },\n \"gpu_description\": \"NVIDIA A100-SXM4-80GB x 4\",\n \"script_name\": \"benchmark_serving.py\",\n \"script_args\": {\n \"description\": \"VLLM Serving - Dense\\nmodel - mistralai/Mixtral-8x7B-Instruct-v0.1\\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\": \"mistralai/Mixtral-8x7B-Instruct-v0.1\",\n \"tokenizer\": \"mistralai/Mixtral-8x7B-Instruct-v0.1\",\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\": 4,\n \"server_args\": \"{'model': 'mistralai/Mixtral-8x7B-Instruct-v0.1', 'tokenizer': 'mistralai/Mixtral-8x7B-Instruct-v0.1', 'max-model-len': 4096, 'host': '127.0.0.1', 'port': 9000, 'tensor-parallel-size': 4, 'disable-log-requests': ''}\"\n },\n \"date\": \"2024-10-08 10:08:17 UTC\",\n \"model\": \"mistralai/Mixtral-8x7B-Instruct-v0.1\",\n \"dataset\": \"sharegpt\"\n}" + }, + { + "name": "{\"name\": \"mean_tpot_ms\", \"description\": \"VLLM Serving - Dense\\nmodel - mistralai/Mixtral-8x7B-Instruct-v0.1\\nmax-model-len - 4096\\nsparsity - None\\nbenchmark_serving {\\n \\\"nr-qps-pair_\\\": \\\"300,1\\\",\\n \\\"dataset\\\": \\\"sharegpt\\\"\\n}\", \"gpu_description\": \"NVIDIA A100-SXM4-80GB x 4\", \"vllm_version\": \"0.5.3.1.dev16+g77db673c\", \"python_version\": \"3.10.12 (main, Sep 30 2024, 21:31:31) [GCC 11.4.0]\", \"torch_version\": \"2.4.0+cu121\"}", + "value": 36.847519373137715, + "unit": "ms", + "extra": "{\n \"description\": \"VLLM Serving - Dense\\nmodel - mistralai/Mixtral-8x7B-Instruct-v0.1\\nmax-model-len - 4096\\nsparsity - None\\nbenchmark_serving {\\n \\\"nr-qps-pair_\\\": \\\"300,1\\\",\\n \\\"dataset\\\": \\\"sharegpt\\\"\\n}\",\n \"benchmarking_context\": {\n \"vllm_version\": \"0.5.3.1.dev16+g77db673c\",\n \"python_version\": \"3.10.12 (main, Sep 30 2024, 21:31:31) [GCC 11.4.0]\",\n \"torch_version\": \"2.4.0+cu121\",\n \"torch_cuda_version\": \"12.1\",\n \"cuda_devices\": \"[_CudaDeviceProperties(name='NVIDIA A100-SXM4-80GB', major=8, minor=0, total_memory=81049MB, multi_processor_count=108), _CudaDeviceProperties(name='NVIDIA A100-SXM4-80GB', major=8, minor=0, total_memory=81049MB, multi_processor_count=108), _CudaDeviceProperties(name='NVIDIA A100-SXM4-80GB', major=8, minor=0, total_memory=81049MB, multi_processor_count=108), _CudaDeviceProperties(name='NVIDIA A100-SXM4-80GB', major=8, minor=0, total_memory=81049MB, multi_processor_count=108)]\",\n \"cuda_device_names\": [\n \"NVIDIA A100-SXM4-80GB\",\n \"NVIDIA A100-SXM4-80GB\",\n \"NVIDIA A100-SXM4-80GB\",\n \"NVIDIA A100-SXM4-80GB\"\n ]\n },\n \"gpu_description\": \"NVIDIA A100-SXM4-80GB x 4\",\n \"script_name\": \"benchmark_serving.py\",\n \"script_args\": {\n \"description\": \"VLLM Serving - Dense\\nmodel - mistralai/Mixtral-8x7B-Instruct-v0.1\\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\": \"mistralai/Mixtral-8x7B-Instruct-v0.1\",\n \"tokenizer\": \"mistralai/Mixtral-8x7B-Instruct-v0.1\",\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\": 4,\n \"server_args\": \"{'model': 'mistralai/Mixtral-8x7B-Instruct-v0.1', 'tokenizer': 'mistralai/Mixtral-8x7B-Instruct-v0.1', 'max-model-len': 4096, 'host': '127.0.0.1', 'port': 9000, 'tensor-parallel-size': 4, 'disable-log-requests': ''}\"\n },\n \"date\": \"2024-10-08 10:08:17 UTC\",\n \"model\": \"mistralai/Mixtral-8x7B-Instruct-v0.1\",\n \"dataset\": \"sharegpt\"\n}" + }, + { + "name": "{\"name\": \"mean_ttft_ms\", \"description\": \"VLLM Serving - Dense\\nmodel - meta-llama/Meta-Llama-3-70B-Instruct\\nmax-model-len - 4096\\nsparsity - None\\nbenchmark_serving {\\n \\\"nr-qps-pair_\\\": \\\"300,1\\\",\\n \\\"dataset\\\": \\\"sharegpt\\\"\\n}\", \"gpu_description\": \"NVIDIA A100-SXM4-80GB x 4\", \"vllm_version\": \"0.5.3.1.dev16+g77db673c\", \"python_version\": \"3.10.12 (main, Sep 30 2024, 21:31:31) [GCC 11.4.0]\", \"torch_version\": \"2.4.0+cu121\"}", + "value": 315.1896276464686, + "unit": "ms", + "extra": "{\n \"description\": \"VLLM Serving - Dense\\nmodel - meta-llama/Meta-Llama-3-70B-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\": \"0.5.3.1.dev16+g77db673c\",\n \"python_version\": \"3.10.12 (main, Sep 30 2024, 21:31:31) [GCC 11.4.0]\",\n \"torch_version\": \"2.4.0+cu121\",\n \"torch_cuda_version\": \"12.1\",\n \"cuda_devices\": \"[_CudaDeviceProperties(name='NVIDIA A100-SXM4-80GB', major=8, minor=0, total_memory=81049MB, multi_processor_count=108), _CudaDeviceProperties(name='NVIDIA A100-SXM4-80GB', major=8, minor=0, total_memory=81049MB, multi_processor_count=108), _CudaDeviceProperties(name='NVIDIA A100-SXM4-80GB', major=8, minor=0, total_memory=81049MB, multi_processor_count=108), _CudaDeviceProperties(name='NVIDIA A100-SXM4-80GB', major=8, minor=0, total_memory=81049MB, multi_processor_count=108)]\",\n \"cuda_device_names\": [\n \"NVIDIA A100-SXM4-80GB\",\n \"NVIDIA A100-SXM4-80GB\",\n \"NVIDIA A100-SXM4-80GB\",\n \"NVIDIA A100-SXM4-80GB\"\n ]\n },\n \"gpu_description\": \"NVIDIA A100-SXM4-80GB x 4\",\n \"script_name\": \"benchmark_serving.py\",\n \"script_args\": {\n \"description\": \"VLLM Serving - Dense\\nmodel - meta-llama/Meta-Llama-3-70B-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-3-70B-Instruct\",\n \"tokenizer\": \"meta-llama/Meta-Llama-3-70B-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\": 4,\n \"server_args\": \"{'model': 'meta-llama/Meta-Llama-3-70B-Instruct', 'tokenizer': 'meta-llama/Meta-Llama-3-70B-Instruct', 'max-model-len': 4096, 'host': '127.0.0.1', 'port': 9000, 'tensor-parallel-size': 4, 'disable-log-requests': ''}\"\n },\n \"date\": \"2024-10-08 10:01:48 UTC\",\n \"model\": \"meta-llama/Meta-Llama-3-70B-Instruct\",\n \"dataset\": \"sharegpt\"\n}" + }, + { + "name": "{\"name\": \"mean_tpot_ms\", \"description\": \"VLLM Serving - Dense\\nmodel - meta-llama/Meta-Llama-3-70B-Instruct\\nmax-model-len - 4096\\nsparsity - None\\nbenchmark_serving {\\n \\\"nr-qps-pair_\\\": \\\"300,1\\\",\\n \\\"dataset\\\": \\\"sharegpt\\\"\\n}\", \"gpu_description\": \"NVIDIA A100-SXM4-80GB x 4\", \"vllm_version\": \"0.5.3.1.dev16+g77db673c\", \"python_version\": \"3.10.12 (main, Sep 30 2024, 21:31:31) [GCC 11.4.0]\", \"torch_version\": \"2.4.0+cu121\"}", + "value": 48.85804111083955, + "unit": "ms", + "extra": "{\n \"description\": \"VLLM Serving - Dense\\nmodel - meta-llama/Meta-Llama-3-70B-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\": \"0.5.3.1.dev16+g77db673c\",\n \"python_version\": \"3.10.12 (main, Sep 30 2024, 21:31:31) [GCC 11.4.0]\",\n \"torch_version\": \"2.4.0+cu121\",\n \"torch_cuda_version\": \"12.1\",\n \"cuda_devices\": \"[_CudaDeviceProperties(name='NVIDIA A100-SXM4-80GB', major=8, minor=0, total_memory=81049MB, multi_processor_count=108), _CudaDeviceProperties(name='NVIDIA A100-SXM4-80GB', major=8, minor=0, total_memory=81049MB, multi_processor_count=108), _CudaDeviceProperties(name='NVIDIA A100-SXM4-80GB', major=8, minor=0, total_memory=81049MB, multi_processor_count=108), _CudaDeviceProperties(name='NVIDIA A100-SXM4-80GB', major=8, minor=0, total_memory=81049MB, multi_processor_count=108)]\",\n \"cuda_device_names\": [\n \"NVIDIA A100-SXM4-80GB\",\n \"NVIDIA A100-SXM4-80GB\",\n \"NVIDIA A100-SXM4-80GB\",\n \"NVIDIA A100-SXM4-80GB\"\n ]\n },\n \"gpu_description\": \"NVIDIA A100-SXM4-80GB x 4\",\n \"script_name\": \"benchmark_serving.py\",\n \"script_args\": {\n \"description\": \"VLLM Serving - Dense\\nmodel - meta-llama/Meta-Llama-3-70B-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-3-70B-Instruct\",\n \"tokenizer\": \"meta-llama/Meta-Llama-3-70B-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\": 4,\n \"server_args\": \"{'model': 'meta-llama/Meta-Llama-3-70B-Instruct', 'tokenizer': 'meta-llama/Meta-Llama-3-70B-Instruct', 'max-model-len': 4096, 'host': '127.0.0.1', 'port': 9000, 'tensor-parallel-size': 4, 'disable-log-requests': ''}\"\n },\n \"date\": \"2024-10-08 10:01:48 UTC\",\n \"model\": \"meta-llama/Meta-Llama-3-70B-Instruct\",\n \"dataset\": \"sharegpt\"\n}" + } + ] } ] }