bartowski commited on
Commit
07a0625
1 Parent(s): 982cc74

measurement.json

Browse files
Files changed (2) hide show
  1. README.md +69 -0
  2. measurement.json +0 -0
README.md ADDED
@@ -0,0 +1,69 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ license: apache-2.0
3
+ datasets:
4
+ - teknium/OpenHermes-2.5
5
+ - jondurbin/truthy-dpo-v0.1
6
+ - jondurbin/gutenberg-dpo-v0.1
7
+ - argilla/dpo-mix-7k
8
+ language:
9
+ - en
10
+ quantized_by: bartowski
11
+ pipeline_tag: text-generation
12
+ ---
13
+
14
+ ## Exllama v2 Quantizations of sparsetral-16x7B-v2-SPIN_iter1
15
+
16
+ Using <a href="https://github.com/turboderp/exllamav2/releases/tag/v0.0.13">turboderp's ExLlamaV2 v0.0.13</a> for quantization.
17
+
18
+ <b>The "main" branch only contains the measurement.json, download one of the other branches for the model (see below)</b>
19
+
20
+ Each branch contains an individual bits per weight, with the main one containing only the meaurement.json for further conversions.
21
+
22
+ Original model: https://huggingface.co/serpdotai/sparsetral-16x7B-v2-SPIN_iter1
23
+
24
+ | Branch | Bits | lm_head bits | VRAM (4k) | VRAM (16k) | VRAM (32k) | Description |
25
+ | ----- | ---- | ------- | ------ | ------ | ------ | ------------ |
26
+ | [8_0](https://huggingface.co/bartowski/sparsetral-16x7B-v2-SPIN_iter1-exl2/tree/8_0) | 8.0 | 8.0 | 8.3 GB | 9.7 GB | 11.8 GB | Maximum quality that ExLlamaV2 can produce, near unquantized performance. |
27
+ | [6_5](https://huggingface.co/bartowski/sparsetral-16x7B-v2-SPIN_iter1-exl2/tree/6_5) | 6.5 | 8.0 | 7.1 GB | 8.5 GB | 10.6 GB | Very similar to 8.0, good tradeoff of size vs performance, **recommended**. |
28
+ | [5_0](https://huggingface.co/bartowski/sparsetral-16x7B-v2-SPIN_iter1-exl2/tree/5_0) | 5.0 | 6.0 | 5.7 GB | 7.1 GB | 9.2 GB | Slightly lower quality vs 6.5, but usable on 8GB cards. |
29
+ | [4_25](https://huggingface.co/bartowski/sparsetral-16x7B-v2-SPIN_iter1-exl2/tree/4_25) | 4.25 | 6.0 | 5.1 GB | 6.5 GB | 8.6 GB | GPTQ equivalent bits per weight, slightly higher quality. |
30
+ | [3_5](https://huggingface.co/bartowski/sparsetral-16x7B-v2-SPIN_iter1-exl2/tree/3_5) | 3.5 | 6.0 | 4.4 GB | 5.8 GB | 7.9 GB | Lower quality, only use if you have to. |
31
+
32
+ ## Download instructions
33
+
34
+ With git:
35
+
36
+ ```shell
37
+ git clone --single-branch --branch 6_5 https://huggingface.co/bartowski/sparsetral-16x7B-v2-SPIN_iter1-exl2 sparsetral-16x7B-v2-SPIN_iter1-exl2-6_5
38
+ ```
39
+
40
+ With huggingface hub (credit to TheBloke for instructions):
41
+
42
+ ```shell
43
+ pip3 install huggingface-hub
44
+ ```
45
+
46
+ To download the `main` (only useful if you only care about measurement.json) branch to a folder called `sparsetral-16x7B-v2-SPIN_iter1-exl2`:
47
+
48
+ ```shell
49
+ mkdir sparsetral-16x7B-v2-SPIN_iter1-exl2
50
+ huggingface-cli download bartowski/sparsetral-16x7B-v2-SPIN_iter1-exl2 --local-dir sparsetral-16x7B-v2-SPIN_iter1-exl2 --local-dir-use-symlinks False
51
+ ```
52
+
53
+ To download from a different branch, add the `--revision` parameter:
54
+
55
+ Linux:
56
+
57
+ ```shell
58
+ mkdir sparsetral-16x7B-v2-SPIN_iter1-exl2-6_5
59
+ huggingface-cli download bartowski/sparsetral-16x7B-v2-SPIN_iter1-exl2 --revision 6_5 --local-dir sparsetral-16x7B-v2-SPIN_iter1-exl2-6_5 --local-dir-use-symlinks False
60
+ ```
61
+
62
+ Windows (which apparently doesn't like _ in folders sometimes?):
63
+
64
+ ```shell
65
+ mkdir sparsetral-16x7B-v2-SPIN_iter1-exl2-6.5
66
+ huggingface-cli download bartowski/sparsetral-16x7B-v2-SPIN_iter1-exl2 --revision 6_5 --local-dir sparsetral-16x7B-v2-SPIN_iter1-exl2-6.5 --local-dir-use-symlinks False
67
+ ```
68
+
69
+ 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