Update README.md
Browse files
README.md
CHANGED
@@ -2,8 +2,29 @@
|
|
2 |
license: apache-2.0
|
3 |
---
|
4 |
|
5 |
-
Experimental quants of 4
|
6 |
|
7 |
-
Goal is to have the best performing MoE <
|
8 |
|
9 |
-
They still need training/finetuning
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2 |
license: apache-2.0
|
3 |
---
|
4 |
|
5 |
+
Experimental quants of 4 expert MoE mixtrals in various GGUF formats.
|
6 |
|
7 |
+
Goal is to have the best performing MoE < 10gb .
|
8 |
|
9 |
+
They still need training/finetuning.
|
10 |
+
|
11 |
+
No sparsity tricks yet.
|
12 |
+
|
13 |
+
8.4gb custom 2bit quant works ok up until 512 token length then starts looping.
|
14 |
+
|
15 |
+
Install llama.cpp from github
|
16 |
+
|
17 |
+
|
18 |
+
```
|
19 |
+
git clone https://github.com/ggerganov/llama.cpp
|
20 |
+
|
21 |
+
cd llama.cpp
|
22 |
+
|
23 |
+
make -j
|
24 |
+
|
25 |
+
wget https://huggingface.co/nisten/quad-mixtrals-gguf/resolve/main/4mixq2.gguf
|
26 |
+
|
27 |
+
./server -m 4mixq2.gguf --host "ec2-3-99-206-122.ca-central-1.compute.amazonaws.com" -c 512```
|
28 |
+
|
29 |
+
|
30 |
+
limit output to 500 tokens
|