Model is speaking differently in the web ui compared to the API

[gdet]

Hello everyone,
I have loaded a local model and it is working very well in the chat tab but when I am using the API it is not speaking so good.
I tried to pass as many arguments that I found but still there are issues.

 payload = {
    "messages": history,  # The user's input with the history
    "mode": "chat-instruct", 
    "character": "Assistant",
    "greeting": "Hello. I would like to ask you some questions",
    "max_new_tokens": 512,  # Adjust as necessary     # Adjust as necessary
    "stop": ["\n"],         # Define the stop tokens as needed
    "do_sample": True,      # Set to False for deterministic outpu       
    "temperature": 0.85,
    'top_p': 1,
    "top_k":0,
    'typical_p': 1,
    'min_p':0.05,
    'repetition_penalty': 1.01,
    'encoder_repetition_penalty': 1,
    'presence_penalty':0,
    'frequency_penalty':0,
    'repetition_penalty_range':1024,
    'top_k': 50,
    'min_length': 0,
    'no_repeat_ngram_size': 0,
    'num_beams': 1,
    'penalty_alpha': 0,
    'length_penalty': 1,
    'early_stopping': False,
    'seed': -1,
    'add_bos_token': True,
    'truncation_length': 2048,
    'ban_eos_token': False,
    'attn_implementation':'eager',
    'torch_dtype':'bf16',
     "seed": 42
}

I also added the same chat_instruct_command but it does not improve.Also greeting is not working. Sometimes it is ok and sometimes it is not ok at all.
What else can I do? Am I missing something?
In model tab there is eager_attention enabled and bf16 but I do not know how to pass them as parameters.
Thank you