Update README.md
Browse files
README.md
CHANGED
@@ -23,23 +23,21 @@ The files in this repo were then quantized to 4bit and 5bit for use with [llama.
|
|
23 |
| ---- | ---- | ---- | ---- | ---- | ----- |
|
24 |
`gpt4-alpaca-lora-30B.GGML.q4_0.bin` | q4_0 | 4bit | 20.3GB | 23GB | Maximum compatibility |
|
25 |
`gpt4-alpaca-lora-30B.GGML.q4_2.bin` | q4_2 | 4bit | 20.3GB | 23GB | Best compromise between resources, speed and quality |
|
26 |
-
`gpt4-alpaca-lora-30B.GGML.q4_3.bin` | q4_3 | 4bit | 24.4GB | 27GB | Maximum quality 4bit, higher RAM requirements and slower inference |
|
27 |
`gpt4-alpaca-lora-30B.GGML.q5_0.bin` | q5_0 | 5bit | 22.4GB | 25GB | Brand new 5bit method. Potentially higher quality than 4bit, at cost of slightly higher resources. |
|
28 |
`gpt4-alpaca-lora-30B.GGML.q5_1.bin` | q5_1 | 5bit | 24.4GB | 27GB | Brand new 5bit method. Slightly higher resource usage than q5_0.|
|
29 |
|
30 |
-
* The q4_0 file provides lower quality, but maximal compatibility. It will work with past and future versions of llama.cpp
|
31 |
* The q4_2 file offers the best combination of performance and quality. This format is still subject to change and there may be compatibility issues, see below.
|
32 |
-
* The q4_3 file offers the highest quality, at the cost of increased RAM usage and slower inference speed. This format is still subject to change and there may be compatibility issues, see below.
|
33 |
* The q5_0 file is using brand new 5bit method released 26th April. This is the 5bit equivalent of q4_0.
|
34 |
* The q5_1 file is using brand new 5bit method released 26th April. This is the 5bit equivalent of q4_1.
|
35 |
|
36 |
-
## q4_2
|
37 |
|
38 |
-
q4_2
|
39 |
|
40 |
In order to use these files you will need to use recent llama.cpp code. And it's possible that future updates to llama.cpp could require that these files are re-generated.
|
41 |
|
42 |
-
If and when the q4_2
|
43 |
|
44 |
If you want to ensure guaranteed compatibility with a wide range of llama.cpp versions, use the q4_0 file.
|
45 |
|
@@ -47,7 +45,7 @@ If you want to ensure guaranteed compatibility with a wide range of llama.cpp ve
|
|
47 |
|
48 |
These new methods were released to llama.cpp on 26th April. You will need to pull the latest llama.cpp code and rebuild to be able to use them.
|
49 |
|
50 |
-
|
51 |
|
52 |
## How to run in `llama.cpp`
|
53 |
|
|
|
23 |
| ---- | ---- | ---- | ---- | ---- | ----- |
|
24 |
`gpt4-alpaca-lora-30B.GGML.q4_0.bin` | q4_0 | 4bit | 20.3GB | 23GB | Maximum compatibility |
|
25 |
`gpt4-alpaca-lora-30B.GGML.q4_2.bin` | q4_2 | 4bit | 20.3GB | 23GB | Best compromise between resources, speed and quality |
|
|
|
26 |
`gpt4-alpaca-lora-30B.GGML.q5_0.bin` | q5_0 | 5bit | 22.4GB | 25GB | Brand new 5bit method. Potentially higher quality than 4bit, at cost of slightly higher resources. |
|
27 |
`gpt4-alpaca-lora-30B.GGML.q5_1.bin` | q5_1 | 5bit | 24.4GB | 27GB | Brand new 5bit method. Slightly higher resource usage than q5_0.|
|
28 |
|
29 |
+
* The q4_0 file provides lower quality, but maximal compatibility. It will work with past and future versions of llama.cpp* The q4_0 file provides lower quality, but maximal compatibility. It will work with past and future versions of llama.cpp
|
30 |
* The q4_2 file offers the best combination of performance and quality. This format is still subject to change and there may be compatibility issues, see below.
|
|
|
31 |
* The q5_0 file is using brand new 5bit method released 26th April. This is the 5bit equivalent of q4_0.
|
32 |
* The q5_1 file is using brand new 5bit method released 26th April. This is the 5bit equivalent of q4_1.
|
33 |
|
34 |
+
## q4_2 compatibility
|
35 |
|
36 |
+
q4_2 is a relatively new 4bit quantisation method offering improved quality. However they are still under development and their formats are subject to change.
|
37 |
|
38 |
In order to use these files you will need to use recent llama.cpp code. And it's possible that future updates to llama.cpp could require that these files are re-generated.
|
39 |
|
40 |
+
If and when the q4_2 file no longer works with recent versions of llama.cpp I will endeavour to update it.
|
41 |
|
42 |
If you want to ensure guaranteed compatibility with a wide range of llama.cpp versions, use the q4_0 file.
|
43 |
|
|
|
45 |
|
46 |
These new methods were released to llama.cpp on 26th April. You will need to pull the latest llama.cpp code and rebuild to be able to use them.
|
47 |
|
48 |
+
Third-party UIs/tools may not support this yet.
|
49 |
|
50 |
## How to run in `llama.cpp`
|
51 |
|