vasqu

from transformers import Ernie4_5TokenizerFast


hf_fast_tok = Ernie4_5TokenizerFast.from_pretrained("baidu/ERNIE-4.5-0.3B-Base-PT", from_slow=True)
hf_fast_tok.model_max_length = 131072
hf_fast_tok.init_kwargs.pop("auto_map", None)
hf_fast_tok.init_kwargs.pop("use_default_system_prompt", None)
hf_fast_tok.init_kwargs.pop("legacy", None)
hf_fast_tok.init_kwargs.pop("sp_model_kwargs", None)

vasqu

"""Using https://github.com/vasqu/dia/tree/hf-next"""

import dac
import soundfile as sf
from datasets import Audio, load_dataset

from dia.model import Dia


# prepare dac

vasqu

import dac
import torch
from datasets import Audio, load_dataset

from transformers import AutoProcessor, DacModel


model = DacModel.from_pretrained("descript/dac_44khz")

dac_model_path = dac.utils.download()

vasqu

import torch

from transformers import AutoModelForCausalLM, AutoTokenizer, DataCollatorWithFlattening


model_id = "meta-llama/Llama-3.2-1B"
tokenizer = AutoTokenizer.from_pretrained(model_id)
tokenizer.pad_token = tokenizer.eos_token
collator = DataCollatorWithFlattening(return_flash_attn_kwargs=True)