Model issues - Ollama
The model doesn't answer well, it completes my last word like "Ahoj" where is the answer "te" and continues with an infinitely long answer in hundreds of sentences. I tryed several modelfiles for import into Ollama but all time it is same. It is same with prepared ollama models from Huggingface... Any idea? Tryed ollama + openwebui
why do you need Modelfile if there are 3 GGUF files, I tried all three, and I have the same behavior, but ollama does not see the any parameters, most likely it generates text based on your prompt (Ahoj\mask\ -> Ahoj+te ... ), it seems to me, you need to use vllm there is at least a temperature value and control of the number of generated tokens, you need to give specific questions in the prompt so the probability will increase that the correct answer will be generated if it was in the dataset (based on the answers it seems to me that they trained on data from a social network), I hope it helped
As it is written in the description, this model is just a baseline/foundation language model. It was pre-trained on continuous text, and it doesn't even recognise EOS token. It is not intended for prompting. If you like to use it for some specific task, you need to fine-tune it on the annotated instruction data.