Invalid model error : too old, regenerate your model files!

[strfic]

Downloaded Alpaca 7B model successfully using the following command as mentioned in README.md:
curl -o ./models/ggml-alpaca-7b-q4.bin -C - https://gateway.estuary.tech/gw/ipfs/QmUp1UGeQFDqJKvtjbSYPBiZZKRjLp8shVP9hT8ZB9Ynv1

When I try to execute the command:
main -m ./models/ggml-alpaca-7b-q4.bin --color -f ./prompts/alpaca.txt -ins

This is the error output:
main: seed = 1679417098
llama_model_load: loading model from './models/ggml-alpaca-7b-q4.bin' - please wait ...
llama_model_load: invalid model file './models/ggml-alpaca-7b-q4.bin' (too old, regenerate your model files!)
main: failed to load model from './models/ggml-alpaca-7b-q4.bin'

How to fix this? Is the downloaded model corrupted and should I download it again? What is the SHA1 hash of the model so that I can verify that the downloaded model is corrupted or not?

[ggerganov]

Use the new links from the README which were updated like an hour or two ago

Edit: nvm - I see you are using them. I guess these are the old Alpaca models. You can convert them to the new format with a helper script. It's somewhere in the repo 😄

[ggerganov]

I think this script should help, but not sure: #324 (comment)

[edwios]

Yes, the link @ggerganov gave above works. Just use the same tokenizer.model that comes with the LLaMA models.

Download the script mentioned in the link above, save it as, for example, convert.py at the same directory as the main, then just run:
python convert.py models/Alpaca/7B models/tokenizer.model (adjust the paths to the model directory and to the tokenizer as needed)

You will find a file called ggml-alpaca-7b-q4.bin.tmp in the same directory as your 7B model, move the original one somewhere and rename this one to ggml-alpaca-7b-q4.bin and you are good to go.

[strfic]

Thanks @ggerganov and @edwios

[ThatCoffeeGuy]

I have the same issue, but in my case the script isn't working either:

python3 convert.py models/alpaca-7B-ggml/ models/tokenizer.model:
[...]
    raise Exception('Invalid file magic. Must be an old style ggml file.')
Exception: Invalid file magic. Must be an old style ggml file.

[jessejohnson]

Same issue as @ThatCoffeeGuy has. My model is the old style. It doesn't work with binaries created after the breaking change.

Edit: This script worked on the 7B alpaca model I downloaded a week ago.

[sachinspanicker]

Does anyone have a converted Alpaca 13B model bin file that you could share so I could download it please?
I am getting the below error no matter what. I tried converting.

Response:

"
main: seed = 1681003839
llama_model_load: loading model from 'models/13B/ggml-model-q4_0.bin' - please wait ...
llama_model_load: invalid model file 'models/13B/ggml-model-q4_0.bin' (bad magic)
main: failed to load model from 'models/13B/ggml-model-q4_0.bin'
bash-3.2$ exit
exit

[sachinspanicker]

can someone share the converted alpaca 13b model bin file ?

[edwios]

Use this script convert-unversioned-ggml-to-ggml.py that comes with the llama.cpp.

For example:

python convert-unversioned-ggml-to-ggml.py models/13B/oldggml/ggml-model-q4_0.bin models/13B/ggml-model-q4_0.bin

[sachinspanicker]

thank you

[sachinspanicker]

Use this script convert-unversioned-ggml-to-ggml.py that comes with the llama.cpp.

For example:

python convert-unversioned-ggml-to-ggml.py models/13B/oldggml/ggml-model-q4_0.bin models/13B/ggml-model-q4_0.bin

And just to confirm, the file that I should be converting is this right ? (https://huggingface.co/Pi3141/alpaca-native-13B-ggml/blob/main/ggml-model-q4_1.bin)

[edwios]

Use this script convert-unversioned-ggml-to-ggml.py that comes with the llama.cpp.
For example:
python convert-unversioned-ggml-to-ggml.py models/13B/oldggml/ggml-model-q4_0.bin models/13B/ggml-model-q4_0.bin

And just to confirm, the file that I should be converting is this right ? (https://huggingface.co/Pi3141/alpaca-native-13B-ggml/blob/main/ggml-model-q4_1.bin)

That I don't know. I have had my models converted from their .pth files in the beginning. But if you've got the message telling you the file magic is wrong, then likely the above script would help.

[sachinspanicker]

Use this script convert-unversioned-ggml-to-ggml.py that comes with the llama.cpp.
For example:
python convert-unversioned-ggml-to-ggml.py models/13B/oldggml/ggml-model-q4_0.bin models/13B/ggml-model-q4_0.bin

And just to confirm, the file that I should be converting is this right ? (https://huggingface.co/Pi3141/alpaca-native-13B-ggml/blob/main/ggml-model-q4_1.bin)

That I don't know. I have had my models converted from their .pth files in the beginning. But if you've got the message telling you the file magic is wrong, then likely the above script would help.

Hi, am just not able to do the convert. Keep getting an error as below

cduser@CDPL17-QA:~/dalai/llama$ python3 convert-pth-to-ggml.py models/13B/ggml-model-q4_1.bin models/13B/ggml-model-q4_0.bin
usage: convert-pth-to-ggml.py [-h] dir_model {0,1} [vocab_only]
convert-pth-to-ggml.py: error: argument ftype: invalid int value: 'models/13B/ggml-model-q4_0.bin'

I even tried what is mentioned in the instructions at https://github.com/ggerganov/llama.cpp

Use this script convert-unversioned-ggml-to-ggml.py that comes with the llama.cpp.
For example:
python convert-unversioned-ggml-to-ggml.py models/13B/oldggml/ggml-model-q4_0.bin models/13B/ggml-model-q4_0.bin

And just to confirm, the file that I should be converting is this right ? (https://huggingface.co/Pi3141/alpaca-native-13B-ggml/blob/main/ggml-model-q4_1.bin)

That I don't know. I have had my models converted from their .pth files in the beginning. But if you've got the message telling you the file magic is wrong, then likely the above script would help.

Hi, I only have this script in llama.cpp after installation - convert-pth-to-ggml.py
Where can I get the convert-unversioned-ggml-to-ggml.py ?

[sachinspanicker]

Use this script convert-unversioned-ggml-to-ggml.py that comes with the llama.cpp.
For example:
python convert-unversioned-ggml-to-ggml.py models/13B/oldggml/ggml-model-q4_0.bin models/13B/ggml-model-q4_0.bin

And just to confirm, the file that I should be converting is this right ? (https://huggingface.co/Pi3141/alpaca-native-13B-ggml/blob/main/ggml-model-q4_1.bin)

That I don't know. I have had my models converted from their .pth files in the beginning. But if you've got the message telling you the file magic is wrong, then likely the above script would help.

I was able to locate the script. Stumbled upon another roadblock. Trying to solve it. Any help appreciated.

:~/llama.cpp$ python3 convert-unversioned-ggml-to-ggml.py ../ggml-model-q4_1.bin ../dalai/alpaca/models/13B/ggml-model-q4_1.bin
terminate called after throwing an instance of 'std::bad_alloc'
what(): std::bad_alloc
Aborted (core dumped)