bartowski commited on
Commit
38a4408
1 Parent(s): 3135968

measurement.json

Browse files
Files changed (2) hide show
  1. README.md +74 -0
  2. measurement.json +0 -0
README.md ADDED
@@ -0,0 +1,74 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ license: apache-2.0
3
+ base_model: alignment-handbook/zephyr-7b-sft-full
4
+ tags:
5
+ - alignment-handbook
6
+ - generated_from_trainer
7
+ - trl
8
+ - dpo
9
+ - generated_from_trainer
10
+ datasets:
11
+ - HuggingFaceH4/ultrafeedback_binarized
12
+ model-index:
13
+ - name: zephyr-7b-dpo-full
14
+ results: []
15
+ quantized_by: bartowski
16
+ pipeline_tag: text-generation
17
+ ---
18
+
19
+ ## Exllama v2 Quantizations of zephyr-7b-dpo-full
20
+
21
+ Using <a href="https://github.com/turboderp/exllamav2/releases/tag/v0.0.12">turboderp's ExLlamaV2 v0.0.12</a> for quantization.
22
+
23
+ # The "main" branch only contains the measurement.json, download one of the other branches for the model (see below)
24
+
25
+ Each branch contains an individual bits per weight, with the main one containing only the meaurement.json for further conversions.
26
+
27
+ Original model: https://huggingface.co/alignment-handbook/zephyr-7b-dpo-full
28
+
29
+ | Branch | Bits | lm_head bits | Size | Description |
30
+ | ----- | ---- | ------- | ------ | ------------ |
31
+ | [8_0](https://huggingface.co/Bartowski/zephyr-7b-dpo-full-exl2/tree/8_0) | 8.0 | 8.0 | 9.8 GB | Maximum quality that ExLlamaV2 can produce, near unquantized performance. |
32
+ | [6_5](https://huggingface.co/Bartowski/zephyr-7b-dpo-full-exl2/tree/6_5) | 6.5 | 8.0 | 8.6 GB | Very similar to 8.0, good tradeoff of size vs performance, **recommended**. |
33
+ | [5_0](https://huggingface.co/Bartowski/zephyr-7b-dpo-full-exl2/tree/5_0) | 5.0 | 6.0 | 7.4 GB | Slightly lower quality vs 6.5, but usable on 8GB cards. |
34
+ | [4_25](https://huggingface.co/Bartowski/zephyr-7b-dpo-full-exl2/tree/4_25) | 4.25 | 6.0 | 6.7 GB | GPTQ equivalent bits per weight, slightly higher quality. |
35
+ | [3_5](https://huggingface.co/Bartowski/zephyr-7b-dpo-full-exl2/tree/3_5) | 3.5 | 6.0 | 6.1 GB | Lower quality, only use if you have to. |
36
+
37
+ All VRAM requirements estimated from 16k context. For 32k context add ~2 GB.
38
+
39
+ ## Download instructions
40
+
41
+ With git:
42
+
43
+ ```shell
44
+ git clone --single-branch --branch 6_5 https://huggingface.co/bartowski/zephyr-7b-dpo-full-exl2 zephyr-7b-dpo-full-exl2-6_5
45
+ ```
46
+
47
+ With huggingface hub (credit to TheBloke for instructions):
48
+
49
+ ```shell
50
+ pip3 install huggingface-hub
51
+ ```
52
+
53
+ To download the `main` (only useful if you only care about measurement.json) branch to a folder called `zephyr-7b-dpo-full-exl2`:
54
+
55
+ ```shell
56
+ mkdir zephyr-7b-dpo-full-exl2
57
+ huggingface-cli download bartowski/zephyr-7b-dpo-full-exl2 --local-dir zephyr-7b-dpo-full-exl2 --local-dir-use-symlinks False
58
+ ```
59
+
60
+ To download from a different branch, add the `--revision` parameter:
61
+
62
+ Linux:
63
+
64
+ ```shell
65
+ mkdir zephyr-7b-dpo-full-exl2-6_5
66
+ huggingface-cli download bartowski/zephyr-7b-dpo-full-exl2 --revision 6_5 --local-dir zephyr-7b-dpo-full-exl2-6_5 --local-dir-use-symlinks False
67
+ ```
68
+
69
+ Windows (which apparently doesn't like _ in folders sometimes?):
70
+
71
+ ```shell
72
+ mkdir zephyr-7b-dpo-full-exl2-6.5
73
+ huggingface-cli download bartowski/zephyr-7b-dpo-full-exl2 --revision 6_5 --local-dir zephyr-7b-dpo-full-exl2-6.5 --local-dir-use-symlinks False
74
+ ```
measurement.json ADDED
The diff for this file is too large to render. See raw diff