This doesn't work for Llama3 models

#44
by birgermoell - opened

Tried out trying to gguf a llama-3 model that I merged. But it didn't work.
https://huggingface.co/birgermoell/Llama-3-dare_ties

Error: Error converting to fp16: b'Traceback (most recent call last):\n File "/home/user/app/llama.cpp/convert.py", line 1548, in \n main()\n File "/home/user/app/llama.cpp/convert.py", line 1515, in main\n vocab, special_vocab = vocab_factory.load_vocab(vocab_types, model_parent_path)\n File "/home/user/app/llama.cpp/convert.py", line 1417, in load_vocab\n vocab = self._create_vocab_by_path(vocab_types)\n File "/home/user/app/llama.cpp/convert.py", line 1407, in _create_vocab_by_path\n raise FileNotFoundError(f"Could not find a tokenizer matching any of {vocab_types}")\nFileNotFoundError: Could not find a tokenizer matching any of ['spm', 'hfft']\n'

I get the same error, it would be cool if the developers fix it.

ggml.ai org

I still don't understand, is this temporary or not supported at all?

https://github.com/ggerganov/llama.cpp/pull/6745

this PR has been merged , so Llama3 is now supported.

please update this app .

This comment has been hidden

Just rebuild it. This will work.

ggml.ai org

Just restarted the app - to pull the latest llama.cpp - running some quick tests for it.

ggml.ai org

ALright made a small patch for llama models to go through the hf-convert script and it works now: https://huggingface.co/reach-vb/llama-3-8b-Q8_0-GGUF ๐Ÿค—

https://github.com/ggerganov/llama.cpp/blob/master/docs/HOWTO-add-model.md

Convert the model to GGUF
This step is done in python with a convert script using the gguf library. Depending on the model architecture, you can use either convert.py or convert-hf-to-gguf.py.

Looks like convert-hf-to-gguf.py can convert any model

cool, so i think this can now be closed!

please confirm @birgermoell

It worked as intended for me.

ggml.ai org

(closing since it is fixed)

reach-vb changed discussion status to closed

Sign up or log in to comment