bartowski commited on
Commit
d43da27
1 Parent(s): 9f7dc5a

Add bits table

Browse files
Files changed (1) hide show
  1. README.md +12 -5
README.md CHANGED
@@ -10,11 +10,18 @@ Using <a href="https://github.com/turboderp/exllamav2/releases/tag/v0.0.11">turb
10
 
11
  Each branch contains an individual bits per weight, with the main one containing only the meaurement.json for further conversions.
12
 
13
- Conversion was done using the default calibration dataset.
14
 
15
- 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.
16
 
17
- Original model: https://huggingface.co/cognitivecomputations/dolphin-2.6-mistral-7b-dpo
 
 
 
 
 
 
 
18
 
19
  <a href="https://huggingface.co/bartowski/dolphin-2.6-mistral-7b-dpo-exl2/tree/4_0">4.0 bits per weight</a>
20
 
@@ -48,6 +55,6 @@ huggingface-cli download bartowski/dolphin-2.6-mistral-7b-dpo-exl2 --local-dir d
48
  To download from a different branch, add the `--revision` parameter:
49
 
50
  ```shell
51
- mkdir dolphin-2.6-mistral-7b-dpo-exl2
52
- huggingface-cli download bartowski/dolphin-2.6-mistral-7b-dpo-exl2 --revision 4_0 --local-dir dolphin-2.6-mistral-7b-dpo-exl2 --local-dir-use-symlinks False
53
  ```
 
10
 
11
  Each branch contains an individual bits per weight, with the main one containing only the meaurement.json for further conversions.
12
 
13
+ Original model: https://huggingface.co/cognitivecomputations/dolphin-2.6-mistral-7b-dpo
14
 
15
+ Model Size: 7b
16
 
17
+ | Branch | Bits | lm_head bits | Dataset | Size | Description |
18
+ | ----- | ---- | ------- | ------- | ------ | ------------ |
19
+ | [8_0](https://huggingface.co/Bartowski/dolphin-2.6-mistral-7b-dpo-exl2/tree/8_0) | 8.0 | 8.0 | Default | 9.8 GB | Maximum quality that ExLlamaV2 can produce, near unquantized performance. |
20
+ | [6_5](https://huggingface.co/Bartowski/dolphin-2.6-mistral-7b-dpo-exl2/tree/6_5) | 6.5 | 8.0 | Default | 8.6 GB | Very similar to 8.0, good tradeoff of size vs performance, **recommended**. |
21
+ | [5_0](https://huggingface.co/Bartowski/dolphin-2.6-mistral-7b-dpo-exl2/tree/5_0) | 5.0 | 6.0 | Default | 7.4 GB | Slightly lower perplexity vs 6.5. |
22
+ | [4_0](https://huggingface.co/Bartowski/dolphin-2.6-mistral-7b-dpo-exl2/tree/4_0) | 4.0 | 6.0 | Default | 6.5 GB | Just under GPTQ equivalent bits per weight. |
23
+
24
+ All VRAM requirements estimated from 16k context. For 32k context add ~2 GB.
25
 
26
  <a href="https://huggingface.co/bartowski/dolphin-2.6-mistral-7b-dpo-exl2/tree/4_0">4.0 bits per weight</a>
27
 
 
55
  To download from a different branch, add the `--revision` parameter:
56
 
57
  ```shell
58
+ mkdir dolphin-2.6-mistral-7b-dpo-exl2-6_5
59
+ huggingface-cli download bartowski/dolphin-2.6-mistral-7b-dpo-exl2 --revision 6_5 --local-dir dolphin-2.6-mistral-7b-dpo-exl2-6_5 --local-dir-use-symlinks False
60
  ```