Delete results/strict_headline.md with huggingface_hub
Browse files- results/strict_headline.md +0 -20
results/strict_headline.md
DELETED
|
@@ -1,20 +0,0 @@
|
|
| 1 |
-
# strict 8192 MiB / no-swap — headline
|
| 2 |
-
|
| 3 |
-
- 模型:`/root/models/minicpm-v4.7-35b-a3b-q4_k_m.gguf`
|
| 4 |
-
- ctx=4096 ubatch=128 threads=8 n_predict=48 temp=0
|
| 5 |
-
- 上限:8192 MiB(anon),每次執行都由 `tools/strict8g.py` 管治;VmSwap 必須全程為 0,匿名記憶體全程 mlock
|
| 6 |
-
|
| 7 |
-
| 指標 | base |
|
| 8 |
-
| --- | ---: |
|
| 9 |
-
| Decode tok/s (median) | 5.83 |
|
| 10 |
-
| expert cache hit % | 68.68 |
|
| 11 |
-
| SSD MB / decode token | 214.1 |
|
| 12 |
-
| slots | 3714 |
|
| 13 |
-
| arena GiB | 6.35 |
|
| 14 |
-
| peak anon RSS MiB | 7799 |
|
| 15 |
-
| peak VmSwap MiB | 0.000 |
|
| 16 |
-
| draft acceptance % (min-max) | — |
|
| 17 |
-
|
| 18 |
-
每個 arm 9 輪;第一個 arm 的單輪 decode tok/s 全距 5.19–6.63,抖動幅度 25%。 請以中位數比較,不要用單輪數字。
|
| 19 |
-
|
| 20 |
-
完整逐輪資料:`strict_headline.json`
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|