File size: 820 Bytes
3c9170e
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
b89fd23
3c9170e
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
## This is a 4bit quant of https://huggingface.co/Aeala/GPT4-x-AlpacaDente2-30b



# My secret sauce:
  * Using comit <a href="https://github.com/0cc4m/GPTQ-for-LLaMa/tree/3c16fd9c7946ebe85df8d951cb742adbc1966ec7">3c16fd9</a> of 0cc4m's GPTQ fork
  * Using PTB as the calibration dataset    
  * Act-order, True-sequential, perfdamp 0.1
     (<i>the default perfdamp is 0.01</i>)
  * No groupsize
  * Will run with CUDA, does not need triton.
  * Quant completed on a 'Premium GPU' and 'High Memory' Google Colab.

## Benchmark results

|<b>Model<b>|<b>C4<b>|<b>WikiText2<b>|<b>PTB<b>|
|:---:|---|---|---|
|This Quant|7.326207160949707|4.957101345062256|24.941526412963867|
|Aela's Quant <a href="https://huggingface.co/Aeala/GPT4-x-AlpacaDente2-30b/resolve/main/4bit.safetensors">here</a>|x.xxxxxx|x.xxxxxx|x.xxxxxx|