How to use llama3-8b-instruct model besides Ollma

[fantastic-413]

I want to use the llama3-8b-instruct model, but when I use ollama, compared with the gpt api, a considerable part of the return format cannot be directly parsed by the following method, resulting in a considerable part (about one tenth) of the extracted_triples after openie being empty, while GPT may be less than one percent.

[fantastic-413]

I checked that a considerable part of the json format output by ollama's llama3:instruct contains errors, so parsing fails, resulting in an empty extracted_triples. But the llama3-8b-instruct reported in the paper can achieve the effect second only to gpt-3.5-turbo, so I am very confused, is llama3:instruct of ollama not the real llama3-8b-instruct?