Upload benchmark_medusa_real.json with huggingface_hub
Browse files- benchmark_medusa_real.json +20 -0
benchmark_medusa_real.json
ADDED
|
@@ -0,0 +1,20 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"vanilla_tok_s": 9.702234454033151,
|
| 3 |
+
"vanilla_steps": 5120,
|
| 4 |
+
"vanilla_tokens": 5120,
|
| 5 |
+
"vanilla_time_s": 527.7134895324707,
|
| 6 |
+
"medusa_tok_s": 5.806649835769172,
|
| 7 |
+
"medusa_steps": 2473,
|
| 8 |
+
"medusa_tokens": 5136,
|
| 9 |
+
"medusa_time_s": 884.5031378269196,
|
| 10 |
+
"speedup": 0.5984858295560347,
|
| 11 |
+
"tokens_per_step": 2.0768297614233724,
|
| 12 |
+
"head_acceptance": [
|
| 13 |
+
0.6295996765062677,
|
| 14 |
+
0.29033562474727054,
|
| 15 |
+
0.11120097048119693,
|
| 16 |
+
0.04569348968863728
|
| 17 |
+
],
|
| 18 |
+
"n_sequences": 20,
|
| 19 |
+
"max_tokens_per_seq": 256
|
| 20 |
+
}
|