Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

update alpaca 13B to 4.1 model #353

Open
wants to merge 2 commits into
base: main
Choose a base branch
from

Conversation

MichaelCharles
Copy link

This resolves issue #350 by updating the version of 13B that is installed.

Copy link

@Grarie Grarie left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

You might want to change the name of the resulting file, too. For now this is rather "hacky".
In the same snippet, change the resolve(dir, "ggml-model-q4_0.bin") part.

The problem is, that the script often expects the filename to be the same (the webapp finds only 4_0 named files). I'm not sure how to solve that properly without the need for a rewrite.

@MichaelCharles
Copy link
Author

I've made some changes to try to make the references to the model file names in general more generic, but I haven't had a chance to test it yet.

@MichaelCharles
Copy link
Author

So, I've tested the updated code and it successfully downloaded the right file, and it seems to be pointing to the right file, however when I run the web app with debug this is the output I'm seeing

/Users/michaelaubrey/dalai/alpaca/main --seed -1 --threads 8 --n_predict 200 --model /Users/michaelaubrey/dalai/alpaca/models/13B/ggml-model-q4_1.bin --top_k 40 --top_p 0.1 --temp 0.1 --repeat_last_n 64 --repeat_penalty 1.3 -p "Once upon a time there lived a girl named Darla, and she
"
exit

The default interactive shell is now zsh.
To update your account to use zsh, please run `chsh -s /bin/zsh`.
For more details, please visit https://support.apple.com/kb/HT208050.
bash-3.2$ /Users/michaelaubrey/dalai/alpaca/main --seed -1 --threads 8 --n_predict 200 --model /Users/michaelaubrey/dalai/alpaca/models/13B/ggml-model-q4_1.bin --top_k 40 --top_p 0.1 --temp 0.1 --repeat_last_n 64 --repeat_penalty 1.3 -p "Once upon a time there lived a girl named Darla, and she
> "
main: seed = 1680573720
llama_model_load: loading model from '/Users/michaelaubrey/dalai/alpaca/models/13B/ggml-model-q4_1.bin' - please wait ...
llama_model_load: invalid model file '/Users/michaelaubrey/dalai/alpaca/models/13B/ggml-model-q4_1.bin' (bad magic)
main: failed to load model from '/Users/michaelaubrey/dalai/alpaca/models/13B/ggml-model-q4_1.bin'
bash-3.2$ exit
exit

@MichaelCharles
Copy link
Author

I checked the sha on the file to make sure that it properly downloaded the whole thing, and it seems to check out. I got the sha from https://huggingface.co/Pi3141/alpaca-native-13B-ggml/blob/main/ggml-model-q4_1.bin

shasum -a 256 ggml-model-q4_1.bin | grep -q "7518013326eeb22c409c201f448c0c622d11476f56c2cd5724842930104635b2" && echo "The file has a matching SHA-256 hash value." || echo "The file does not have a matching SHA-256 hash value."
The file has a matching SHA-256 hash value.

@karfly
Copy link

karfly commented Apr 6, 2023

@mcaubrey did you solve the problem? I get the same error with your code

@MichaelCharles
Copy link
Author

@mcaubrey did you solve the problem? I get the same error with your code

No, as mentioned above, I did solve the problem of it not downloading the model.

However, for whatever reason, the model hosted on Hugging Face does not seem to be in the correct format to use with Dalai.

If you use my updated code, you can download that model, but in the end it still won't execute. You'll likely get the same "bad magic" error that I did.

@devmrh
Copy link

devmrh commented Apr 8, 2023

any update? i have 13B/ggml-model-q4_1.bin' (bad magic) error too :(

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

Successfully merging this pull request may close these issues.

None yet

4 participants