docs: add base vs tuned bench comparison
Browse files
README.md
CHANGED
|
@@ -109,3 +109,20 @@ LoRA weights: **apache-2.0** — see License chain table above for derivation ra
|
|
| 109 |
## Related
|
| 110 |
|
| 111 |
See the full [Ailiance-fr LoRA collection](https://huggingface.co/Ailiance-fr).
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 109 |
## Related
|
| 110 |
|
| 111 |
See the full [Ailiance-fr LoRA collection](https://huggingface.co/Ailiance-fr).
|
| 112 |
+
|
| 113 |
+
|
| 114 |
+
## Bench comparison (2026-05-11)
|
| 115 |
+
|
| 116 |
+
### Base model (Apertus-70B-Instruct-2509) capability
|
| 117 |
+
|
| 118 |
+
| Task | Score | Notes |
|
| 119 |
+
|---|---:|---|
|
| 120 |
+
| ARC-Easy acc / acc_norm | **0.81 / 0.77** | W3 lm-eval-harness BF16 |
|
| 121 |
+
| GSM8K-CoT | TIMEOUT (1800s budget) | base 70B BF16 too slow for CoT |
|
| 122 |
+
| MMLU-Pro Computer Science | TIMEOUT | |
|
| 123 |
+
|
| 124 |
+
### This LoRA (tuned) — bench PENDING
|
| 125 |
+
|
| 126 |
+
Production usage: served via gateway alias `ailiance-apertus-<domain>` on
|
| 127 |
+
<https://www.ailiance.fr> through the Apertus multi-LoRA hot-swap server
|
| 128 |
+
(Studio :9322, 1 base + 10 LoRA dynamic swap, ~40GB VRAM).
|