Bug on the tokenizer, using the code that you provided for the inference.

#2
by Ptrnk - opened

I force downloaded the tokenizer many times in case of failure (or file corruption) but it doesnt seem to be the case.

Firstly i had to change the config file (An error occured reuqired only 2 arguments type and factor) :

"rope_scaling": {
"factor": 8.0,
"high_freq_factor": 4.0,
"low_freq_factor": 1.0,
"original_max_position_embeddings": 8192,
"rope_type": "llama3"
},



Second:

Even when i change it there seems to be another error with the tokenizer file :

File ~/.local/lib/python3.10/site-packages/transformers/tokenization_utils_fast.py:111, in PreTrainedTokenizerFast.init(self, *args, **kwargs)
108 fast_tokenizer = copy.deepcopy(tokenizer_object)
109 elif fast_tokenizer_file is not None and not from_slow:
110 # We have a serialization from tokenizers which let us directly build the backend
--> 111 fast_tokenizer = TokenizerFast.from_file(fast_tokenizer_file)
112 elif slow_tokenizer is not None:
113 # We need to convert a slow tokenizer to build the backend
114 fast_tokenizer = convert_slow_tokenizer(slow_tokenizer)

Exception: data did not match any variant of untagged enum ModelWrapper at line 1356219 column 3



Could you please assist me? I would like to test this LLM.

Institute for Language and Speech Processing org
edited 13 days ago

What transformers version are you using? after transformers == 4.45.0 the way the tokenizers are handled, changed.

I'm guessing you're trying to load the model with a transformers version < 4.45.0 which would explain both the RoPE config error and the tokenizer loading error

Institute for Language and Speech Processing org

You can also try out the latest version of Krikri Instruct here.

File upload does not work, it would be nice to set it up to make some rag testing

Thank you very much for the quick response you were correct @LVouk . It is accurate in its responses i love it. Will there be uploaded a gguf version for faster inference at ollama?

You can create one by yourself @Ptrnk
gguf from hg
https://github.com/ggerganov/llama.cpp/discussions/2948

gguf to ollama
https://github.com/ollama/ollama/blob/main/docs/import.md#importing-a-gguf-based-model-or-adapter

Although I believe you dont need ollama at all :P

soksof changed discussion status to closed
Institute for Language and Speech Processing org
edited 13 days ago

File upload does not work, it would be nice to set it up to make some rag testing

We are working to integrate file upload and tool calling to our UI, it's not our top priority at the moment but we'll get to it and finalize it soon! Thank you for your interest in the model!

Sign up or log in to comment