Upload folder using huggingface_hub
#2
by
nifleisch
- opened
- base_results.json +18 -0
- plots.png +0 -0
- smashed_results.json +18 -0
base_results.json
ADDED
@@ -0,0 +1,18 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
{
|
2 |
+
"current_gpu_type": "NVIDIA L40S",
|
3 |
+
"current_gpu_total_memory": 45372.6875,
|
4 |
+
"memory_inference_first": 25590.0,
|
5 |
+
"memory_inference": 25590.0,
|
6 |
+
"token_generation_latency_sync": 30.128338623046876,
|
7 |
+
"token_generation_latency_async": 30.288023874163628,
|
8 |
+
"token_generation_throughput_sync": 0.033191342294428515,
|
9 |
+
"token_generation_throughput_async": 0.03301635009780294,
|
10 |
+
"token_generation_CO2_emissions": 3.236854331824288e-06,
|
11 |
+
"token_generation_energy_consumption": 0.001333430427257508,
|
12 |
+
"inference_latency_sync": 23.984115219116212,
|
13 |
+
"inference_latency_async": 22.998762130737305,
|
14 |
+
"inference_throughput_sync": 0.04169426267611338,
|
15 |
+
"inference_throughput_async": 0.0434806010130225,
|
16 |
+
"inference_CO2_emissions": 2.2244397934888317e-06,
|
17 |
+
"inference_energy_consumption": 7.886087285242966e-06
|
18 |
+
}
|
plots.png
ADDED
smashed_results.json
ADDED
@@ -0,0 +1,18 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
{
|
2 |
+
"current_gpu_type": "NVIDIA L40S",
|
3 |
+
"current_gpu_total_memory": 45372.6875,
|
4 |
+
"memory_inference_first": 4312.0,
|
5 |
+
"memory_inference": 4208.0,
|
6 |
+
"token_generation_latency_sync": 46.756124877929686,
|
7 |
+
"token_generation_latency_async": 46.56997695565224,
|
8 |
+
"token_generation_throughput_sync": 0.02138757227231272,
|
9 |
+
"token_generation_throughput_async": 0.02147306194616077,
|
10 |
+
"token_generation_CO2_emissions": 3.367901542186602e-06,
|
11 |
+
"token_generation_energy_consumption": 0.002321453711341601,
|
12 |
+
"inference_latency_sync": 39.73529586791992,
|
13 |
+
"inference_latency_async": 38.7270450592041,
|
14 |
+
"inference_throughput_sync": 0.02516654219271448,
|
15 |
+
"inference_throughput_async": 0.025821748043808834,
|
16 |
+
"inference_CO2_emissions": 3.481497167551675e-06,
|
17 |
+
"inference_energy_consumption": 1.445772659465061e-05
|
18 |
+
}
|