Commit ·
cfe6f8e
1
Parent(s): 9e0641d
impl: Benchmark results of Mistral
Browse files
results/mistral-7b/benchmark_results.json
ADDED
|
@@ -0,0 +1,67 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"model": "mistral-7b",
|
| 3 |
+
"avg_bits": 6.95,
|
| 4 |
+
"compression": [
|
| 5 |
+
{
|
| 6 |
+
"context_len": 512,
|
| 7 |
+
"fp16_mb": 67.11,
|
| 8 |
+
"uniform8_mb": 33.55,
|
| 9 |
+
"mixed_precision_mb": 29.17,
|
| 10 |
+
"compression_vs_fp16": 2.3,
|
| 11 |
+
"compression_vs_8bit": 1.15
|
| 12 |
+
},
|
| 13 |
+
{
|
| 14 |
+
"context_len": 1024,
|
| 15 |
+
"fp16_mb": 134.22,
|
| 16 |
+
"uniform8_mb": 67.11,
|
| 17 |
+
"mixed_precision_mb": 58.33,
|
| 18 |
+
"compression_vs_fp16": 2.3,
|
| 19 |
+
"compression_vs_8bit": 1.15
|
| 20 |
+
},
|
| 21 |
+
{
|
| 22 |
+
"context_len": 2048,
|
| 23 |
+
"fp16_mb": 268.44,
|
| 24 |
+
"uniform8_mb": 134.22,
|
| 25 |
+
"mixed_precision_mb": 116.66,
|
| 26 |
+
"compression_vs_fp16": 2.3,
|
| 27 |
+
"compression_vs_8bit": 1.15
|
| 28 |
+
},
|
| 29 |
+
{
|
| 30 |
+
"context_len": 4096,
|
| 31 |
+
"fp16_mb": 536.87,
|
| 32 |
+
"uniform8_mb": 268.44,
|
| 33 |
+
"mixed_precision_mb": 233.31,
|
| 34 |
+
"compression_vs_fp16": 2.3,
|
| 35 |
+
"compression_vs_8bit": 1.15
|
| 36 |
+
},
|
| 37 |
+
{
|
| 38 |
+
"context_len": 8192,
|
| 39 |
+
"fp16_mb": 1073.74,
|
| 40 |
+
"uniform8_mb": 536.87,
|
| 41 |
+
"mixed_precision_mb": 466.62,
|
| 42 |
+
"compression_vs_fp16": 2.3,
|
| 43 |
+
"compression_vs_8bit": 1.15
|
| 44 |
+
}
|
| 45 |
+
],
|
| 46 |
+
"memory": [
|
| 47 |
+
{
|
| 48 |
+
"context": 1024,
|
| 49 |
+
"peak_memory_gb": 14.76
|
| 50 |
+
},
|
| 51 |
+
{
|
| 52 |
+
"context": 4096,
|
| 53 |
+
"peak_memory_gb": 15.53
|
| 54 |
+
},
|
| 55 |
+
{
|
| 56 |
+
"context": 8192,
|
| 57 |
+
"peak_memory_gb": 16.56
|
| 58 |
+
}
|
| 59 |
+
],
|
| 60 |
+
"decode_tokens_per_sec": 37.2,
|
| 61 |
+
"perplexity": 14.23,
|
| 62 |
+
"summary": {
|
| 63 |
+
"fp16_8k_mb": 1073.74,
|
| 64 |
+
"ours_8k_mb": 466.62,
|
| 65 |
+
"compression_8k": 2.3
|
| 66 |
+
}
|
| 67 |
+
}
|