Add gpt-oss to vLLM dashboard (#74)

* Add gpt-oss to vLLM dashboard

Signed-off-by: Huy Do <huydhn@gmail.com>

* Typo

Signed-off-by: Huy Do <huydhn@gmail.com>

* Default to h100

Signed-off-by: Huy Do <huydhn@gmail.com>

* Typo

Signed-off-by: Huy Do <huydhn@gmail.com>

* Set input and output lengths

Signed-off-by: Huy Do <huydhn@gmail.com>

* Another tweak

Signed-off-by: Huy Do <huydhn@gmail.com>

* Another tweak

Signed-off-by: Huy Do <huydhn@gmail.com>

---------

Signed-off-by: Huy Do <huydhn@gmail.com>

Author: huydhn

.github/workflows/vllm-benchmark.yml

@@ -53,7 +53,7 @@ jobs:
         shell: bash
         env:
           MODELS: ${{ inputs.models || '' }}
-          RUNNERS: ${{ inputs.runners || '' }}
+          RUNNERS: ${{ inputs.runners || 'h100' }}
         run: |
           set -eux
 

vllm-benchmarks/benchmarks/cuda/latency-tests.json

@@ -50,5 +50,27 @@
             "num_iters": 15,
             "max_model_len": 8192
         }
+    },
+    {
+        "test_name": "latency_gpt_oss_20b_tp1",
+        "parameters": {
+            "model": "openai/gpt-oss-20b",
+            "tensor_parallel_size": 1,
+            "load_format": "dummy",
+            "num_iters_warmup": 5,
+            "num_iters": 15,
+            "max_model_len": 8192
+        }
+    },
+    {
+        "test_name": "latency_gpt_oss_120b_tp4",
+        "parameters": {
+            "model": "openai/gpt-oss-120b",
+            "tensor_parallel_size": 4,
+            "load_format": "dummy",
+            "num_iters_warmup": 5,
+            "num_iters": 15,
+            "max_model_len": 8192
+        }
     }
 ]

vllm-benchmarks/benchmarks/cuda/serving-tests.json

@@ -411,5 +411,45 @@
             "random_input_len": 30720,
             "random_output_len": 100
         }
+    },
+    {
+        "test_name": "serving_gpt_oss_20b_tp1_random_in5k_out8k",
+        "qps_list": [1, 4, 16, "inf"],
+        "server_parameters": {
+            "model": "openai/gpt-oss-20b",
+            "tensor_parallel_size": 1,
+            "swap_space": 16,
+            "disable_log_stats": "",
+            "disable_log_requests": "",
+            "load_format": "dummy"
+        },
+        "client_parameters": {
+            "model": "openai/gpt-oss-20b",
+            "backend": "vllm",
+            "dataset_name": "random",
+            "num_prompts": 200,
+            "random_input_len": 5250,
+            "random_output_len": 8250
+        }
+    },
+    {
+        "test_name": "serving_gpt_oss_120b_tp4_random_in5k_out8k",
+        "qps_list": [1, 4, 16, "inf"],
+        "server_parameters": {
+            "model": "openai/gpt-oss-120b",
+            "tensor_parallel_size": 4,
+            "swap_space": 16,
+            "disable_log_stats": "",
+            "disable_log_requests": "",
+            "load_format": "dummy"
+        },
+        "client_parameters": {
+            "model": "openai/gpt-oss-120b",
+            "backend": "vllm",
+            "dataset_name": "random",
+            "num_prompts": 200,
+            "random_input_len": 5250,
+            "random_output_len": 8250
+        }
     }
 ]

vllm-benchmarks/benchmarks/cuda/throughput-tests.json

@@ -55,5 +55,29 @@
             "backend": "vllm",
             "max_model_len": 8192
         }
+    },
+    {
+        "test_name": "throughput_gpt_oss_20b_tp1",
+        "parameters": {
+            "model": "openai/gpt-oss-20b",
+            "tensor_parallel_size": 1,
+            "load_format": "dummy",
+            "dataset": "./ShareGPT_V3_unfiltered_cleaned_split.json",
+            "num_prompts": 200,
+            "backend": "vllm",
+            "max_model_len": 8192
+        }
+    },
+    {
+        "test_name": "throughput_gpt_oss_120b_tp4",
+        "parameters": {
+            "model": "openai/gpt-oss-120b",
+            "tensor_parallel_size": 4,
+            "load_format": "dummy",
+            "dataset": "./ShareGPT_V3_unfiltered_cleaned_split.json",
+            "num_prompts": 200,
+            "backend": "vllm",
+            "max_model_len": 8192
+        }
     }
 ]