try higher temperature

generate.py

@@ -37,6 +37,7 @@ model = models.transformers(model_id, device=device)
 tokenizer = AutoTokenizer.from_pretrained(model_id)
 sampler = PenalizedMultinomialSampler()
 low_temperature_sampler = PenalizedMultinomialSampler(temperature=0.3)
+high_temperature_sampler = PenalizedMultinomialSampler(temperature=1.1)
 empty_tokens = [token_id for token_id in range(tokenizer.vocab_size) if not tokenizer.decode([token_id], skip_special_tokens=True).strip()]
 sampler.set_max_repeats(empty_tokens, 1)
 disallowed_patterns = [regex.compile(r"\p{Han}")]  # focus on english for now
@@ -60,7 +61,7 @@ class Dataset(BaseModel):
     data: conlist(Sample, min_length=2, max_length=3)  # type: ignore
 
 
-samples_generator_template = generate.json(model, Dataset, sampler=sampler)
+samples_generator_template = generate.json(model, Dataset, sampler=high_temperature_sampler)
 
 class Columns(BaseModel):
     columns: conset(Annotated[str, StringConstraints(pattern=r'[a-z0-9_]+')], min_length=2, max_length=len(Sample.model_fields) - 1)  # type: ignore