metadata

library_name: transformers
language:
  - en
pipeline_tag: text-generation

Meta-Llama-3-8B-Instruct-4bit

BitsAndBytes 4bit Quantized Model

Quantization Configuration

load_in_4bit: True
llm_int8_threshold: 6.0
bnb_4bit_quant_type: nf4
bnb_4bit_use_double_quant: True
bnb_4bit_compute_dtype: bfloat16

How to use

Load Required Libraries

!pip install transformers 
!pip install peft
!pip install -U bitsandbytes

Load model directly

from transformers import AutoTokenizer, AutoModelForCausalLM
tokenizer = AutoTokenizer.from_pretrained("meta-llama/Meta-Llama-3-8B-Instruct")
model = AutoModelForCausalLM.from_pretrained("SwastikM/Meta-Llama-3-8B-Instruct_bitsandbytes_4bit")

messages = [
    {"role": "system", "content": "You are a Coder."},
    {"role": "user", "content": "How to ctrate a list in Python?"}
]

input_ids = tokenizer.apply_chat_template(
    messages,
    add_generation_prompt=True,
    return_tensors="pt"
).to(model.device)

terminators = [
    tokenizer.eos_token_id,
    tokenizer.convert_tokens_to_ids("<|eot_id|>")
]

outputs = model.generate(
    input_ids,
    max_new_tokens=256,
    eos_token_id=terminators,
    do_sample=False,
    temperature=0.0
)

response = outputs[0][input_ids.shape[-1]:]
print(tokenizer.decode(response, skip_special_tokens=True))

Output

In Python, you can create a list in several ways:

1. Using the `list()` function:

my_list = list()

This creates an empty list.

2. Using square brackets `[]`:

my_list = []

This also creates an empty list.

3. Using the `list()` function with an iterable (such as a string or a tuple):

my_list = list("hello")
print(my_list)  # Output: ['h', 'e', 'l', 'l', 'o']

4. Using the `list()` function with a range of numbers:

my_list = list(range(1, 6))
print(my_list)  # Output: [1, 2, 3, 4, 5]

5. Using the `list()` function with a dictionary:

my_dict = {"a": 1, "b": 2, "c": 3}
my_list = list(my_dict.keys())
print(my_list)  # Output: ['a', 'b', 'c']

Note that in Python, lists are mutable, meaning you can add, remove, or modify elements after creating the list.

Size Comparison

The table shows comparison VRAM requirements for loading and training of FP16 Base Model and 4bit GPTQ quantized model with PEFT. The value for base model referenced from Model Memory Calculator from HuggingFace

Model	Total Size
Base Model	28 GB
4bitQuantized	5.21 GB

Acknowledgment

Thanks to @AMerve Noyan for precise intro.
Thanks to @HuggungFace Team for the Blog.
Thanks to @Meta for the Open Source Model.

Model Card Authors

Swastik Maiti