Lewdiculous
commited on
Commit
•
62d6e17
1
Parent(s):
6fd157b
Update README.md
Browse files
README.md
CHANGED
@@ -8,7 +8,30 @@ Quants:
|
|
8 |
"Q6_K", "Q8_0", "IQ3_M", "IQ3_S", "IQ3_XXS"
|
9 |
]
|
10 |
```
|
11 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
12 |
|
13 |
# Original model information:
|
14 |
|
|
|
8 |
"Q6_K", "Q8_0", "IQ3_M", "IQ3_S", "IQ3_XXS"
|
9 |
]
|
10 |
```
|
11 |
+
This repository hosts GGUF-IQ-Imatrix quants for [SanjiWatsuki/Sonya-7B](https://huggingface.co/SanjiWatsuki/Sonya-7B).
|
12 |
+
|
13 |
+
**What does "Imatrix" mean?**
|
14 |
+
|
15 |
+
It stands for **Importance Matrix**, a technique used to improve the quality of quantized models.
|
16 |
+
The **Imatrix** is calculated based on calibration data, and it helps determine the importance of different model activations during the quantization process.
|
17 |
+
The idea is to preserve the most important information during quantization, which can help reduce the loss of model performance, especially when the calibration data is diverse.
|
18 |
+
[[1]](https://github.com/ggerganov/llama.cpp/discussions/5006) [[2]](https://github.com/ggerganov/llama.cpp/discussions/5263#discussioncomment-8395384)
|
19 |
+
|
20 |
+
For imatrix data generation, kalomaze's `groups_merged.txt` with added roleplay chats was used, you can find it [here](https://huggingface.co/Lewdiculous/Datura_7B-GGUF-Imatrix/blob/main/imatrix-with-rp-format-data.txt). This was just to add a bit more diversity to the data.
|
21 |
+
|
22 |
+
**Steps:**
|
23 |
+
|
24 |
+
```
|
25 |
+
Base⇢ GGUF(F16)⇢ Imatrix-Data(F16)⇢ GGUF(Imatrix-Quants)
|
26 |
+
```
|
27 |
+
*Using the latest llama.cpp at the time.*
|
28 |
+
|
29 |
+
```python
|
30 |
+
quantization_options = [
|
31 |
+
"Q4_K_M", "Q4_K_S", "IQ4_XS", "Q5_K_M", "Q5_K_S",
|
32 |
+
"Q6_K", "Q8_0", "IQ3_M", "IQ3_S", "IQ3_XXS"
|
33 |
+
]
|
34 |
+
```
|
35 |
|
36 |
# Original model information:
|
37 |
|