from transformers import AutoTokenizer, AutoModelForCausalLM, BitsAndBytesConfig
import torch

TORCH_DTYPE = 'bfloat16'
nf4_config = BitsAndBytesConfig(
    load_in_4bit=True,
    bnb_4bit_quant_type='nf4',
    bnb_4bit_use_double_quant=True,
    bnb_4bit_compute_dtype=getattr(torch, TORCH_DTYPE)
)

tokenizer = AutoTokenizer.from_pretrained('mesolitica/mallam-3B-4096')
model = AutoModelForCausalLM.from_pretrained(
    'mesolitica/mallam-3B-4096',
    use_flash_attention_2 = True,
    quantization_config = nf4_config
)
prompt = '<s>nama saya'
inputs = tokenizer([prompt], return_tensors='pt', add_special_tokens=False).to('cuda')

generate_kwargs = dict(
    inputs,
    max_new_tokens=512,
    top_p=0.95,
    top_k=50,
    temperature=0.9,
    do_sample=True,
    num_beams=1,
    repetition_penalty=1.05,
)
r = model.generate(**generate_kwargs)

RichardErkhov/mesolitica_-_mallam-3B-4096-4bits

Original model description:

MaLLaM 🌙 3B (Malaysia Large Language Model), Pretrain 3B 4096 context length on Malaysian text

how-to

RichardErkhov/
mesolitica_-_mallam-3B-4096-4bits