app.py

from transformers import AutoTokenizer, AutoModelForCausalLM
import torch

model_name = "S1mp1eXXX/Nimi-1b-thinking"

tokenizer = AutoTokenizer.from_pretrained(model_name)
model = AutoModelForCausalLM.from_pretrained(
    model_name,
    torch_dtype=torch.float16,
    device_map="auto"
)

def respond(message, history, system_message, max_tokens, temperature, top_p):
    messages = system_message + "\n"

    for h in history:
        messages += f"{h['role']}: {h['content']}\n"

    messages += f"user: {message}\nassistant:"

    inputs = tokenizer(messages, return_tensors="pt").to(model.device)

    output = model.generate(
        **inputs,
        max_new_tokens=max_tokens,
        temperature=temperature,
        top_p=top_p,
    )

    decoded = tokenizer.decode(output[0], skip_special_tokens=True)
    yield decoded