not getting expected output

#2
by sakshicss - opened

!./build/bin/llama-cli -m {model_path} -p "what is cloud computing?"
when i run this i get output
what is cloud computing
what is cloud computing
what is cloud computing
what is cloud computing
like this

ggml.ai org

@sakshicss Can you try this with one of the instruction tuned (it) models, like https://huggingface.co/ggml-org/gemma-3-270m-it-GGUF, or https://huggingface.co/ggml-org/gemma-3-270m-it-qat-GGUF?
It looks like you might be using a pretrained/base model.

Thank you @danbev for your response! Yes, I have tried the instruction-tuned (it) model, and it is working fine. Could you please explain why the base model behaves this way and how exactly the base model works?

ggml.ai org

Could you please explain why the base model behaves this way and how exactly the base model works?

The pretrained/base model is only trained to predict the next token given the previous tokens. If you prompt it with "What is the capital of France?", it might continue with "What is the capital of Germany? What is the capital of Italy...".
The instruction tuned model starts with a pretrained base model, then gets additional training on examples of instructions/questions paired with appropriate responses.

The base model is like a very sophisticated autocomplete, while the instruction-tuned model is like a knowledgeable assistant who understands that when you ask a question, you want an answer, not just plausible next text. Hope that clarifies the difference between the two.

Sign up or log in to comment