Update README.md
Browse files
README.md
CHANGED
@@ -22,30 +22,26 @@ Using <a href="https://github.com/turboderp/exllamav2/releases/tag/v0.0.11">turb
|
|
22 |
|
23 |
Each branch contains an individual bits per weight, with the main one containing only the meaurement.json for further conversions.
|
24 |
|
25 |
-
Conversion was done using the default calibration dataset.
|
26 |
-
|
27 |
-
Default arguments used except when the bits per weight is above 6.0, at that point the lm_head layer is quantized at 8 bits per weight instead of the default 6.
|
28 |
-
|
29 |
Original model: https://huggingface.co/fblgit/UNA-dolphin-2.6-mistral-7b-dpo-laser
|
30 |
|
|
|
31 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
32 |
|
33 |
-
|
34 |
-
|
35 |
-
<a href="https://huggingface.co/bartowski/UNA-dolphin-2.6-mistral-7b-dpo-laser-exl2/tree/6_5">6.5 bits per weight</a>
|
36 |
-
|
37 |
-
<a href="https://huggingface.co/bartowski/UNA-dolphin-2.6-mistral-7b-dpo-laser-exl2/tree/5_0">5.0 bits per weight</a>
|
38 |
-
|
39 |
-
<a href="https://huggingface.co/bartowski/UNA-dolphin-2.6-mistral-7b-dpo-laser-exl2/tree/4_0">4.0 bits per weight</a>
|
40 |
-
|
41 |
-
<a href="https://huggingface.co/bartowski/UNA-dolphin-2.6-mistral-7b-dpo-laser-exl2/tree/3_5">3.5 bits per weight</a>
|
42 |
|
43 |
## Download instructions
|
44 |
|
45 |
With git:
|
46 |
|
47 |
```shell
|
48 |
-
git clone --single-branch --branch
|
49 |
```
|
50 |
|
51 |
With huggingface hub (credit to TheBloke for instructions):
|
@@ -64,6 +60,6 @@ huggingface-cli download bartowski/UNA-dolphin-2.6-mistral-7b-dpo-laser-exl2 --l
|
|
64 |
To download from a different branch, add the `--revision` parameter:
|
65 |
|
66 |
```shell
|
67 |
-
mkdir UNA-dolphin-2.6-mistral-7b-dpo-laser-exl2
|
68 |
-
huggingface-cli download bartowski/UNA-dolphin-2.6-mistral-7b-dpo-laser-exl2 --revision
|
69 |
```
|
|
|
22 |
|
23 |
Each branch contains an individual bits per weight, with the main one containing only the meaurement.json for further conversions.
|
24 |
|
|
|
|
|
|
|
|
|
25 |
Original model: https://huggingface.co/fblgit/UNA-dolphin-2.6-mistral-7b-dpo-laser
|
26 |
|
27 |
+
Model Size: 7b
|
28 |
|
29 |
+
| Branch | Bits | lm_head bits | Dataset | Size | Description |
|
30 |
+
| ----- | ---- | ------- | ------- | ------ | ------------ |
|
31 |
+
| [8_0](https://huggingface.co/Bartowski/UNA-dolphin-2.6-mistral-7b-dpo-laser-exl2/tree/8_0) | 8.0 | 8.0 | Default | 9.8 GB | Maximum quality that ExLlamaV2 can produce, near unquantized performance. |
|
32 |
+
| [6_5](https://huggingface.co/Bartowski/UNA-dolphin-2.6-mistral-7b-dpo-laser-exl2/tree/6_5) | 6.5 | 8.0 | Default | 8.6 GB | Very similar to 8.0, good tradeoff of size vs performance, **recommended**. |
|
33 |
+
| [5_0](https://huggingface.co/Bartowski/UNA-dolphin-2.6-mistral-7b-dpo-laser-exl2/tree/5_0) | 5.0 | 6.0 | Default | 7.4 GB | Slightly lower perplexity vs 6.5. |
|
34 |
+
| [4_0](https://huggingface.co/Bartowski/UNA-dolphin-2.6-mistral-7b-dpo-laser-exl2/tree/4_0) | 4.0 | 6.0 | Default | 6.5 GB | Just under GPTQ equivalent bits per weight. |
|
35 |
+
| [3_5](https://huggingface.co/Bartowski/UNA-dolphin-2.6-mistral-7b-dpo-laser-exl2/tree/3_5) | 3.5 | 6.0 | Default | 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/UNA-dolphin-2.6-mistral-7b-dpo-laser-exl2
|
45 |
```
|
46 |
|
47 |
With huggingface hub (credit to TheBloke for instructions):
|
|
|
60 |
To download from a different branch, add the `--revision` parameter:
|
61 |
|
62 |
```shell
|
63 |
+
mkdir UNA-dolphin-2.6-mistral-7b-dpo-laser-exl2-6_5
|
64 |
+
huggingface-cli download bartowski/UNA-dolphin-2.6-mistral-7b-dpo-laser-exl2 --revision 6_5 --local-dir UNA-dolphin-2.6-mistral-7b-dpo-laser-exl2-6_5 --local-dir-use-symlinks False
|
65 |
```
|