RichardErkhov/

piotr-ai_-_polanka-3b-pretrain-full-v0.4-gguf

from transformers import AutoTokenizer, AutoModelForCausalLM
import torch

device = "cuda"
device_map = {
    "lm_head": device,
    "model": device
}

tokenizer = AutoTokenizer.from_pretrained("piotr-ai/polanka-3b-pretrain-full-v0.4")
model = AutoModelForCausalLM.from_pretrained("piotr-ai/polanka-3b-pretrain-full-v0.4", torch_dtype=torch.bfloat16, device_map=device_map)

prompt = "Psychologia to"

model_input = tokenizer(prompt, return_tensors="pt").to(device)
generated = model.generate(**model_input, do_sample=True, temperature=0.6, max_new_tokens=100)[0]
decoded = tokenizer.decode(generated, skip_special_tokens=False)
print(decoded)