Update README.md
Browse files
README.md
CHANGED
|
@@ -1,3 +1,27 @@
|
|
| 1 |
-
---
|
| 2 |
-
license: apache-2.0
|
| 3 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
---
|
| 2 |
+
license: apache-2.0
|
| 3 |
+
tags:
|
| 4 |
+
- math
|
| 5 |
+
- mistral
|
| 6 |
+
- llm
|
| 7 |
+
- gguf
|
| 8 |
+
- mathstral
|
| 9 |
+
- java
|
| 10 |
+
- mistral.java
|
| 11 |
+
---
|
| 12 |
+
|
| 13 |
+
# Pure quantizations of `Mathstral-7B-v0.1` for [mistral.java](https://github.com/mukel/mistral.java).
|
| 14 |
+
|
| 15 |
+
In the wild, Q8_0 quantizations are fine, but Q4_0 quantizations are rarely pure e.g. the output.weights tensor is quantized with Q6_K, instead of Q4_0.
|
| 16 |
+
A pure Q4_0 quantization can be generated from a high precision (F32, F16, BFLOAT16) .gguf source with the quantize utility from llama.cpp as follows:
|
| 17 |
+
|
| 18 |
+
```
|
| 19 |
+
./llama-quantize --pure ./Mathstral-7B-v0.1-F32.gguf ./Mathstral-7B-v0.1-Q4_0.gguf Q4_0
|
| 20 |
+
```
|
| 21 |
+
|
| 22 |
+
Original model: [https://huggingface.co/mistralai/mathstral-7B-v0.1](https://huggingface.co/mistralai/mathstral-7B-v0.1)
|
| 23 |
+
|
| 24 |
+
****Note that this model does not support a System prompt.**
|
| 25 |
+
|
| 26 |
+
Mathstral 7B is a model specializing in mathematical and scientific tasks, based on Mistral 7B.
|
| 27 |
+
You can read more in the [official blog post](https://mistral.ai/news/mathstral/).
|