Commit
•
1ff5b51
1
Parent(s):
d0778e5
Create README.md
Browse files
README.md
ADDED
|
@@ -0,0 +1,18 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
---
|
| 2 |
+
library_name: transformers
|
| 3 |
+
tags:
|
| 4 |
+
- mistral
|
| 5 |
+
- finetuned
|
| 6 |
+
- conversational
|
| 7 |
+
- text-generation-inference
|
| 8 |
+
---
|
| 9 |
+
Official [AQLM](https://arxiv.org/abs/2401.06118) quantization of [mistralai/Mistral-7B-Instruct-v0.2
|
| 10 |
+
](https://huggingface.co/mistralai/Mistral-7B-Instruct-v0.2).
|
| 11 |
+
|
| 12 |
+
For this quantization, we used 2 codebooks of 8 bits.
|
| 13 |
+
|
| 14 |
+
Results:
|
| 15 |
+
| Model | Quantization | MMLU (5-shot) | Model size, Gb |
|
| 16 |
+
|------|------|-------|------|------|
|
| 17 |
+
|CohereForAI/c4ai-command-r-v01| None | 0.5912 | 14.5 |
|
| 18 |
+
| | 2x8 | 0.4384 | 2.3 |
|