fix config jsons

neuralmagic · Feb 28, 2024 · 3dc2de6 · 3dc2de6
1 parent 32b08c4
commit 3dc2de6
Show file tree

Hide file tree

Showing 2 changed files with 6 additions and 33 deletions.
diff --git a/neuralmagic/benchmarks/configs/benchmark_serving.json b/neuralmagic/benchmarks/configs/benchmark_serving.json
@@ -8,6 +8,9 @@
 				"mistralai/Mistral-7B-Instruct-v0.2",
 				"NousResearch/Llama-2-7b-chat-hf"
 			],
+			"tensor_parallel_size" : [
+				1
+			],
 			"max_model_lens": [
 				4096
 			],
@@ -27,36 +30,6 @@
 					"sharegpt"
 				]
 			}
-		},
-		{
-			"description": "Benchmark vllm serving",
-			"models": [
-				"facebook/opt-125m",
-				"TinyLlama/TinyLlama-1.1B-Chat-v1.0",
-				"mistralai/Mistral-7B-Instruct-v0.2",
-				"NousResearch/Llama-2-7b-chat-hf"
-			],
-			"max_model_lens": [
-				4096
-			],
-			"sparsity": [],
-			"script_name": "benchmark_serving",
-			"script_args": {
-				"num-prompts_": [
-					50,
-					100
-				],
-				"request-rate_": [
-					0.5,
-					"inf"
-				],
-				"best-of": [
-					1
-				],
-				"dataset": [
-					"sharegpt"
-				]
-			}
 		}
 	]
 }
diff --git a/neuralmagic/benchmarks/configs/benchmark_throughput.json b/neuralmagic/benchmarks/configs/benchmark_throughput.json
@@ -20,7 +20,7 @@
 				"output-len": [
 					128
 				],
-				"tensor-parallel-size": [
+				"tensor-parallel-size_": [
 					1
 				],
 				"n": [
@@ -64,7 +64,7 @@
 				"output-len": [
 					1
 				],
-				"tensor-parallel-size": [
+				"tensor-parallel-size_": [
 					1
 				],
 				"n": [
@@ -101,7 +101,7 @@
 				"output-len": [
 					128
 				],
-				"tensor-parallel-size": [
+				"tensor-parallel-size_": [
 					1
 				],
 				"n": [