File size: 1,327 Bytes
0d4c2a2
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
9e7e61d
883600b
3d096d1
0d4c2a2
487b963
 
 
 
 
67ab3fc
487b963
67ab3fc
487b963
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
# Phind-CodeLlama-34B-v2 EXL2
 
Weights of [Phind-CodeLlama-34B-v2](https://huggingface.co/Phind/Phind-CodeLlama-34B-v2) converted
to [EXL2](https://github.com/turboderp/exllamav2#exl2-quantization) format.
 
Each separate quant is in a different branch, like in The Bloke's GPTQ repos.
 
```
export BRANCH=5_0-bpw-h8
git clone --single-branch --branch ${BRANCH} https://huggingface.co/latimar/Phind-Codellama-34B-v2-exl2
```
 
There are the following branches:
 
```
5_0-bpw-h8
4_625-bpw-h6
4_125-bpw-h6
2_75-bpw-h6
2_55-bpw-h6
```

* Calibration dataset used for conversion: [wikitext-v2](https://huggingface.co/datasets/wikitext/blob/refs%2Fconvert%2Fparquet/wikitext-2-v1/test/0000.parquet)
* Evaluation dataset used to calculate perplexity: [wikitext-v2](https://huggingface.co/datasets/wikitext/blob/refs%2Fconvert%2Fparquet/wikitext-2-v1/validation/0000.parquet)
* PPL max seq. length used: 1792 (2048 with 5.0-bpw-h8 causes OOM on RTX 4090 when evaluating ppl, so had to go down a bit)
 

| BPW         | Perplexity  | File Size (Gb) | 
|--------------------------------------------|
|  2.55-h6    |  15.0901    |    10.56       |
|  2.75-h6    |  13.6153    |    11.33       |
|  4.125-h6   |  6.8095     |    16.65       |
|  4.625-h6   |  6.7992     |    18.58       |
|  5.0-h8     |  6.7785     |    20.09       |