Full parameters pretrain checkpoint with Polish content from base model: openlm-research/open_llama_3b_v2
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)
<s> Psychologia to fascynuj膮ca dziedzina wiedzy, kt贸ra wci膮偶 kryje przed nami wiele tajemnic. Mechanizmy, kt贸re rz膮dz膮 naszymi zachowaniami, decyzjami i wyborami, to nadal przedmiot bada艅 wielu naukowc贸w. Niekt贸re z nich ok
- Downloads last month
- 8