bartowski commited on
Commit
9ee892b
1 Parent(s): 91481ad

measurement.json

Browse files
Files changed (2) hide show
  1. README.md +68 -0
  2. measurement.json +0 -0
README.md ADDED
@@ -0,0 +1,68 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ license: agpl-3.0
3
+ language:
4
+ - en
5
+ quantized_by: bartowski
6
+ pipeline_tag: text-generation
7
+ ---
8
+
9
+ ## Exllama v2 Quantizations of Michel-13B
10
+
11
+ Using <a href="https://github.com/turboderp/exllamav2/releases/tag/v0.0.12">turboderp's ExLlamaV2 v0.0.12</a> for quantization.
12
+
13
+ # The "main" branch only contains the measurement.json, download one of the other branches for the model (see below)
14
+
15
+ Each branch contains an individual bits per weight, with the main one containing only the meaurement.json for further conversions.
16
+
17
+ Original model: https://huggingface.co/PotatoOff/Michel-13B
18
+
19
+ No GQA - VRAM requirements will be higher
20
+
21
+ | Branch | Bits | lm_head bits | VRAM (4k) | VRAM (16k) | Description |
22
+ | ----- | ---- | ------- | ------ | ------ | ------------ |
23
+ | [6_5](https://huggingface.co/Bartowski/Michel-13B-exl2/tree/6_5) | 6.5 | 8.0 | 14.4 GB | 24.0 GB | Near unquantized performance at vastly reduced size, **recommended**. |
24
+ | [5_0](https://huggingface.co/Bartowski/Michel-13B-exl2/tree/5_0) | 5.0 | 6.0 | 12.1 GB | 21.7 GB | Slightly lower perplexity vs 6.5, can fit in 12 GB card with even lower context. |
25
+ | [4_25](https://huggingface.co/Bartowski/Michel-13B-exl2/tree/4_25) | 4.25 | 6.0 | 10.9 GB | 20.5 GB | GPTQ equivalent bits per weight. |
26
+ | [3_75](https://huggingface.co/Bartowski/Michel-13B-exl2/tree/3_75) | 3.75 | 6.0 | 10.1 GB | 19.7 GB | Lower quality but still generally usable. |
27
+ | [3_0](https://huggingface.co/Bartowski/Michel-13B-exl2/tree/3_0) | 3.0 | 6.0 | 9.1 GB | 18.7 GB | Very low quality, not recommended unless you have to. |
28
+
29
+ VRAM requirements listed for both 4k context and 16k context since without GQA the differences are massive (9.6 GB)
30
+
31
+ ## Download instructions
32
+
33
+ With git:
34
+
35
+ ```shell
36
+ git clone --single-branch --branch 6_5 https://huggingface.co/bartowski/Michel-13B-exl2 Michel-13B-exl2-6_5
37
+ ```
38
+
39
+ With huggingface hub (credit to TheBloke for instructions):
40
+
41
+ ```shell
42
+ pip3 install huggingface-hub
43
+ ```
44
+
45
+ To download the `main` (only useful if you only care about measurement.json) branch to a folder called `Michel-13B-exl2`:
46
+
47
+ ```shell
48
+ mkdir Michel-13B-exl2
49
+ huggingface-cli download bartowski/Michel-13B-exl2 --local-dir Michel-13B-exl2 --local-dir-use-symlinks False
50
+ ```
51
+
52
+ To download from a different branch, add the `--revision` parameter:
53
+
54
+ Linux:
55
+
56
+ ```shell
57
+ mkdir Michel-13B-exl2-6_5
58
+ huggingface-cli download bartowski/Michel-13B-exl2 --revision 6_5 --local-dir Michel-13B-exl2-6_5 --local-dir-use-symlinks False
59
+ ```
60
+
61
+ Windows (which apparently doesn't like _ in folders sometimes?):
62
+
63
+ ```shell
64
+ mkdir Michel-13B-exl2-6.5
65
+ huggingface-cli download bartowski/Michel-13B-exl2 --revision 6_5 --local-dir Michel-13B-exl2-6.5 --local-dir-use-symlinks False
66
+ ```
67
+
68
+ Want to support my work? Visit my ko-fi page here: https://ko-fi.com/bartowski
measurement.json ADDED
The diff for this file is too large to render. See raw diff