c13081bac9221145b156ee9fd5f9ac9ef90eada16325157de61a61ce80429718

base_results.json

@@ -0,0 +1,19 @@
+{
+    "current_gpu_type": "Tesla T4",
+    "current_gpu_total_memory": 15095.0625,
+    "perplexity": 3.4586403369903564,
+    "memory_inference_first": 808.0,
+    "memory_inference": 808.0,
+    "token_generation_latency_sync": 39.485802841186526,
+    "token_generation_latency_async": 38.54773212224245,
+    "token_generation_throughput_sync": 0.025325558252469627,
+    "token_generation_throughput_async": 0.025941863371593512,
+    "token_generation_CO2_emissions": 1.9341882246324658e-05,
+    "token_generation_energy_consumption": 0.0018858868968200813,
+    "inference_latency_sync": 122.10734748840332,
+    "inference_latency_async": 48.58226776123047,
+    "inference_throughput_sync": 0.008189515377811079,
+    "inference_throughput_async": 0.02058364185292351,
+    "inference_CO2_emissions": 1.933046505752413e-05,
+    "inference_energy_consumption": 6.813219116923538e-05
+}

smashed_results.json

@@ -0,0 +1,19 @@
+{
+    "current_gpu_type": "Tesla T4",
+    "current_gpu_total_memory": 15095.0625,
+    "perplexity": 3.8198323249816895,
+    "memory_inference_first": 206.0,
+    "memory_inference": 206.0,
+    "token_generation_latency_sync": 57.987845993041994,
+    "token_generation_latency_async": 58.05423278361559,
+    "token_generation_throughput_sync": 0.017244993030435907,
+    "token_generation_throughput_async": 0.01722527285352785,
+    "token_generation_CO2_emissions": 9.527744243728705e-06,
+    "token_generation_energy_consumption": 0.003671008174955222,
+    "inference_latency_sync": 58.546810150146484,
+    "inference_latency_async": 54.94275093078613,
+    "inference_throughput_sync": 0.017080349850580167,
+    "inference_throughput_async": 0.018200763213690286,
+    "inference_CO2_emissions": 9.524661982723253e-06,
+    "inference_energy_consumption": 3.420485454644485e-05
+}