Text Generation
Transformers
PyTorch
Safetensors
Finnish
llama
finnish
text-generation-inference

Hello. Thanks for your model. I tried to make a ggml version but encountered this error:

#1
by NikolayKozloff - opened

Loading model file f3\pytorch_model.bin
vocabtype: spm
Loading vocab file f3\tokenizer.model
params: n_vocab:64256 n_embd:3200 n_mult:4320 n_head:32 n_layer:26
Traceback (most recent call last):
File "C:\kcp\convert.py", line 1325, in
main()
File "C:\kcp\convert.py", line 1320, in main
OutputFile.write_all(outfile, params, output_type, model, vocab)
File "C:\kcp\convert.py", line 1103, in write_all
check_vocab_size(params, vocab)
File "C:\kcp\convert.py", line 1058, in check_vocab_size
raise Exception(msg)
Exception: Vocab size mismatch (model has 64256, but f3\tokenizer.model has 32000)

I'd like to use this too. Too bad the readme-file is empty, so there are no instructions.

Finnish-NLP org
edited Nov 17, 2023

Sorry for the delayed answer. Previously the repository contained old wrong tokenizer with vocab mismatch. Can you try the ggml conversion again?
Readme-file is still empty as these new llama-finnish models are not yet officially released. Official release with proper readme-files is coming in a couple weeks time.

Thanks for the cool model! Can you give a hint of the prompt template before the official documentation is added, can't wait to get test this one.

Finnish-NLP org

Readme has now been added. For prompt template, there is no template because this model is pretrained model without any fine-tuning. Thus, this model is not good at following any instructions like ChatGPT, and should be fine-tuned to get better results.

aapot changed discussion status to closed

Sign up or log in to comment