How to convert old ALPACA q4_0 model into ggjt format?

[multimediaconverter]

I'm trying to use a python script, but it returns the following error:

d:\ALPACA2>python migrate-ggml-2023-03-30-pr613.py ggml-alpaca-7b-q4.bin ggml-alpaca-7b-q4-ggjt.bin
Traceback (most recent call last):
File "d:\ALPACA2\migrate-ggml-2023-03-30-pr613.py", line 313, in
main()
File "d:\ALPACA2\migrate-ggml-2023-03-30-pr613.py", line 274, in main
tokens = read_tokens(fin, hparams)
File "d:\ALPACA2\migrate-ggml-2023-03-30-pr613.py", line 135, in read_tokens
word = fin.read(length)
ValueError: read length must be non-negative or -1

[thefcraft]

Same error for gpt4all model

[akx]

Convert it with convert-unversioned-ggml-to-ggml.py first.

[multimediaconverter]

I converted alpaca model into this new format successfully.

1. convert-unversioned-ggml-to-ggml.py with tokenizer.model file and migrate-ggml-2023-03-30-pr613.py.

2. VS2017 Express x64 with AVX2 to compile examples/main/main.cpp on Windows.

And what can I say: finally, it works! llama.cpp generates text ~1.5-2 times faster than alpaca.cpp

[sw]

Original issue resolved -> Closing this, duplicates #704