File size: 1,050 Bytes
4ef20dd
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
The model is for: https://github.com/ggerganov/llama.cpp

Date: 2023-04-01
ggml model file magic: 0x67676a74 (ggjt in hex)
ggml model file version: 1

Torrent contents:
The fine tune described at https://huggingface.co/chavinlo/gpt4-x-alpaca converted to ggml format from https://huggingface.co/anon8231489123/gpt4-x-alpaca-13b-native-4bit-128g/blob/f267949dcd5a5e6451933cec3d0b5661f4f9c889/gpt-x-alpaca-13b-native-4bit-128g-cuda.pt
Details about the GPTQ quantization process: https://huggingface.co/anon8231489123/gpt4-x-alpaca-13b-native-4bit-128g/blob/f267949dcd5a5e6451933cec3d0b5661f4f9c889/README.md

Tools used:
[1] Conversion to ggml: https://github.com/ggerganov/llama.cpp/blob/3265b102beb7674d010644ca2a1bd30a58f9f6b5/convert.py and [2]
[2] Added extra tokens: https://huggingface.co/chavinlo/alpaca-13b/blob/464a0bd1ec16f3a7d5295a0035aff87f307e62f1/added_tokens.json
[3] Migration to the latest llama.cpp model format: https://github.com/ggerganov/llama.cpp/blob/3525899277d2e2bdc8ec3f0e6e40c47251608700/migrate-ggml-2023-03-30-pr613.py