| { | |
| "base_current_gpu_type": "NVIDIA A100-PCIE-40GB", | |
| "base_current_gpu_total_memory": 40339.3125, | |
| "base_perplexity": 8.891793251037598, | |
| "base_token_generation_latency_sync": 39.96690216064453, | |
| "base_token_generation_latency_async": 39.75822236388922, | |
| "base_token_generation_throughput_sync": 0.025020703280443424, | |
| "base_token_generation_throughput_async": 0.02515202995866987, | |
| "base_token_generation_CO2_emissions": null, | |
| "base_token_generation_energy_consumption": null, | |
| "base_inference_latency_sync": 119.90845489501953, | |
| "base_inference_latency_async": 39.05057907104492, | |
| "base_inference_throughput_sync": 0.008339695485822956, | |
| "base_inference_throughput_async": 0.025607814884913607, | |
| "base_inference_CO2_emissions": null, | |
| "base_inference_energy_consumption": null, | |
| "smashed_current_gpu_type": "NVIDIA A100-PCIE-40GB", | |
| "smashed_current_gpu_total_memory": 40339.3125, | |
| "smashed_perplexity": 9.47966194152832, | |
| "smashed_token_generation_latency_sync": 167.34748077392578, | |
| "smashed_token_generation_latency_async": 168.06612703949213, | |
| "smashed_token_generation_throughput_sync": 0.0059755904025285385, | |
| "smashed_token_generation_throughput_async": 0.005950038937739193, | |
| "smashed_token_generation_CO2_emissions": null, | |
| "smashed_token_generation_energy_consumption": null, | |
| "smashed_inference_latency_sync": 266.22822265625, | |
| "smashed_inference_latency_async": 196.9926118850708, | |
| "smashed_inference_throughput_sync": 0.0037561757728863535, | |
| "smashed_inference_throughput_async": 0.005076332510294441, | |
| "smashed_inference_CO2_emissions": null, | |
| "smashed_inference_energy_consumption": null | |
| } |