configs.json

{
    "general": {
        "device": "cuda"
    },
    "prompts": {
        "vlm": {
            "gemma": "Describe in detail the object in the image in a few words. \n Maintain a concise, factual tone. \n Ignore the background. \n Here is an example of your output if you are given an image of a red apple on a table: red apple",
            "llama_vision": "Describe in detail the object in the image in a few words. \n Maintain a concise, factual tone. \n Ignore the background. \n Here is an example of your output if you are given an image of a red apple on a table: red apple",
            "llava": "Describe in detail the object in the image in a few words. \n Maintain a concise, factual tone. \n Ignore the background. \n Here is an example of your output if you are given an image of a red apple on a table: red apple",
            "qwen": "Describe in detail the object in the image in a few words. \n Maintain a concise, factual tone. \n Ignore the background. \n Here is an example of your output if you are given an image of a red apple on a table: red apple",
            "gemma3n": "Describe in detail the object in the image in a few words. \n Maintain a concise, factual tone. \n Ignore the background. \n Here is an example of your output if you are given an image of a red apple on a table: red apple",
            "mistral": "Describe in detail the object in the image in a few words. \n Maintain a concise, factual tone. \n Ignore the background. \n Here is an example of your output if you are given an image of a red apple on a table: red apple",
            "smolvlm2": "Describe in detail the object in the image in a few words. \n Maintain a concise, factual tone. \n Ignore the background. \n Here is an example of your output if you are given an image of a red apple on a table: red apple"
        }
    }
}