Commit
β’
bf493b7
1
Parent(s):
529d931
Update README.md (#15)
Browse files- Update README.md (c7502579da9cfe878db415c5ee62503564b20666)
README.md
CHANGED
@@ -20,9 +20,21 @@ tags:
|
|
20 |
- moe
|
21 |
---
|
22 |
|
|
|
|
|
23 |
# Mixtral-8x22B-v0.1-GGUF
|
24 |
|
25 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
26 |
|
27 |
## Load sharded model
|
28 |
|
@@ -81,7 +93,9 @@ Since this appears to be a base model, it will keep on generating.
|
|
81 |
|
82 |
## Credit
|
83 |
|
84 |
-
|
|
|
|
|
85 |
|
86 |
βββββ
|
87 |
ββββββββββββββββββ
|
|
|
20 |
- moe
|
21 |
---
|
22 |
|
23 |
+
<img src="./mixtral-8x22b.jpeg" width="600" />
|
24 |
+
|
25 |
# Mixtral-8x22B-v0.1-GGUF
|
26 |
|
27 |
+
On April 10th, @MistralAI released a model named "Mixtral 8x22B," an 176B MoE via magnet link (torrent):
|
28 |
+
|
29 |
+
- 176B MoE with ~40B active
|
30 |
+
- Context length of 65k tokens
|
31 |
+
- The base model can be fine-tuned
|
32 |
+
- Requires ~260GB VRAM in fp16, 73GB in int4
|
33 |
+
- Licensed under Apache 2.0, according to their Discord
|
34 |
+
- Available on @huggingface (community)
|
35 |
+
- Utilizes a tokenizer similar to previous models
|
36 |
+
|
37 |
+
The GGUF and quantized models here are based on [v2ray/Mixtral-8x22B-v0.1](https://huggingface.co/v2ray/Mixtral-8x22B-v0.1) model
|
38 |
|
39 |
## Load sharded model
|
40 |
|
|
|
93 |
|
94 |
## Credit
|
95 |
|
96 |
+
- [MistralAI](https://huggingface.co/mistralai) for opening the weights
|
97 |
+
- [v2ray](https://huggingface.co/v2ray/) for downloading, converting, and sharing it with the community [Mixtral-8x22B-v0.1](https://huggingface.co/v2ray/Mixtral-8x22B-v0.1)
|
98 |
+
- [philschmid]([https://huggingface.co/philschmid) for the photo he shared on his Twitter
|
99 |
|
100 |
βββββ
|
101 |
ββββββββββββββββββ
|