readme : update gpt4all instructions

[prusnak]

fixes https://github.com/ggerganov/llama.cpp/issues/975

[robbintt]

I am running into one issue with the bin:

python convert.py models/gpt4all-lora-quantized.bin
Loading model file models/gpt4all-lora-quantized.bin
Traceback (most recent call last):
  File "convert.py", line 1145, in <module>
    main()
  File "convert.py", line 1133, in main
    vocab = load_vocab(vocab_dir)
  File "convert.py", line 1075, in load_vocab
    raise FileNotFoundError(f"Could not find tokenizer.model in {path} or its parent; if it's in another directory, pass the directory as --vocab-dir")
FileNotFoundError: Could not find tokenizer.model in models or its parent; if it's in another directory, pass the directory as --vocab-dir```

[prusnak]

As the error says, you don't have the models/tokenizer.model file. You need to obtain it from the same place you are obtaining your LLaMa weights.

[MillionthOdin16]

I am running into one issue with the bin:

Also, based off the model and message, you probably should be using the migrate script, not the convert script. That model file is already ggml, it just needs to be migrated to current version.

[prusnak]

migrate script, not the convert script

there is no migrate script anymore, just one complex one-does-it-all convert script (https://github.com/ggerganov/llama.cpp/pull/545)

[MillionthOdin16]

There's not a migrate-ggml-2023-03-30-pr613.py script?

[prusnak]

There's not a migrate-ggml-2023-03-30-pr613.py script?

No.

[MillionthOdin16]

Welp. Good to know. Unfortunate that it has the same exact name as a previous convert.py script that does a subset of tasks lol

[robbintt]

Anyways this PR is just for the readme. I have another error, will share in 20 min.

On Fri, Apr 14, 2023 at 12:32 MillionthOdin16 @.***> wrote:

Welp. Good to know. Unfortunate that it has the same exact name as a previous convert.py script that does a subset of tasks lol

— Reply to this email directly, view it on GitHub https://github.com/ggerganov/llama.cpp/pull/980#issuecomment-1509123846, or unsubscribe https://github.com/notifications/unsubscribe-auth/AAOZDW2DGXIYDFQYDV4WFUDXBGQ4HANCNFSM6AAAAAAW6Z2L5Y . You are receiving this because you commented.Message ID: <ggerganov/llama. @.***>

[prusnak]

Anyways this PR is just for the readme. I have another error, will share in 20 min.

Exactly, this PR is about fixing readme. Any subsequent errors are out of scope of this PR and should be fixed by future PRs.

[robbintt]

Since this section is about Gpt4all, it might be helpful to enumerate the required files

[comex]

Thanks for cleaning up my oversight.

[prusnak]

Converting to draft since it seems that gpt4all conversion is broken, see https://github.com/ggerganov/llama.cpp/issues/984

[prusnak]

Converting to draft since it seems that gpt4all conversion is broken, see #984

Fix for this error is in https://github.com/ggerganov/llama.cpp/pull/1031

GPT4All-J is not supported at the moment, which is reflected in the README

[prusnak]

Fixes https://github.com/ggerganov/llama.cpp/issues/1036

[hughobrien]

suggest adding the SHASUMS so people can validate their conversions