Commit
•
509ed2d
1
Parent(s):
cdeead5
Added fp16 baseline
Browse files
README.md
CHANGED
@@ -3,9 +3,10 @@ Official [AQLM](https://arxiv.org/abs/2401.06118) quantization of [google/gemma-
|
|
3 |
For this quantization, we used 1 codebook of 16 bits.
|
4 |
|
5 |
|
6 |
-
Results:
|
7 |
-
| Model |
|
8 |
|------|------|------|-------|-------|-------|------|------|
|
9 |
-
|
|
|
|
10 |
|
11 |
To learn more about the inference, as well as the information on how to quantize models yourself, please refer to the [official GitHub repo](https://github.com/Vahe1994/AQLM).
|
|
|
3 |
For this quantization, we used 1 codebook of 16 bits.
|
4 |
|
5 |
|
6 |
+
Results (0-shot `acc`):
|
7 |
+
| Model | Quantization | WinoGrande | PiQA | HellaSwag | ArcE | ArcC | Model size, Gb |
|
8 |
|------|------|------|-------|-------|-------|------|------|
|
9 |
+
|gemma-2b| None | 0.6472 | 0.7715 | 0.5279 | 0.7403 | 0.4053| 5.0 |
|
10 |
+
| |1x16| 0.6275 | 0.7318 | 0.4582 | 0.6923 | 0.3259| 1.7 |
|
11 |
|
12 |
To learn more about the inference, as well as the information on how to quantize models yourself, please refer to the [official GitHub repo](https://github.com/Vahe1994/AQLM).
|