-
Notifications
You must be signed in to change notification settings - Fork 83
llama: introduce good TinyLlama-15M model #174
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
base: master
Are you sure you want to change the base?
Conversation
sorry for the mess.
when I tried it wasn't ouputing correctly either. |
I had replied to you on slack what's going on. I'll repeat here: If you run it, it works:
Note that the prompt used above is The prompt you used, the one that was also in the README, was
You can fix it by appending a comma at the end, like so:
All in all, the model does not seem to be the best Stories-TinyLlama, judged by how quickly it gets into a repetitive state. I also tested with a different prompt that does not end in a comma:
|
I updated the documentation so that it does not suggest that very prompt that confuses the model. Also changed the name back to |
I think your model is still corrupted. This is what I get with official weights from Karpathy (current zml/master):
|
This change re-introduces TinyLlama-15M, but with HF tokenizer and safetensors weights.