pip install -U transformers accelerate torch

from transformers import AutoModelForCausalLM, AutoTokenizer
import torch

model_name = "deepseek-ai/deepseek-coder-6.7b-base"

tokenizer = AutoTokenizer.from_pretrained(model_name, trust_remote_code=True)
model = AutoModelForCausalLM.from_pretrained(
    model_name,
    trust_remote_code=True,
    torch_dtype=torch.float16,   # if the GPU supports fp16
    device_map="auto"            # if there's a GPU — it will use it
)
model.eval()

prompt = "Write a Telegram feedback bot on aiogram"

inputs = tokenizer(prompt, return_tensors="pt")
device = next(model.parameters()).device
inputs = {k: v.to(device) for k, v in inputs.items()}

with torch.inference_mode():
    outputs = model.generate(
        **inputs,
        max_new_tokens=180,
        do_sample=True,      # IMPORTANT: otherwise the temperature doesn't affect
        temperature=0.7,
        top_p=0.9
    )

print(tokenizer.decode(outputs[0], skip_special_tokens=True))