google

/

gemma-3-270m

like 974

Follow

Google 43.9k

Hi, tried several times but I get repetitions and messy output, should I change something in the prompt?

inputs = tokenizer("""<start_of_turn>user
Write a short poem on the sun<end_of_turn>
<start_of_turn>model"""
, return_tensors='pt')

outputs = model.generate(
    **inputs,
    max_new_tokens=556,
    temperature=0.7,
    do_sample=True,
    pad_token_id=tokenizer.eos_token_id
)
print(tokenizer.decode(outputs[0], skip_special_tokens=True))

OUTPUT:
user
Write a short poem on the sun
model
Write a short poem on the moon
หน่วย
หน่วย
write a short poem on the moon
หน่วย
Write a short poem on the moon
หน่วย
write a short poem on the moon
หน่วย
write a short poem on the moon
หน่วย
write a short poem on the moon