Fix the size of int and bool type when computing module size

[notsyncing]

What does this PR do?

Hello, I'm trying using accelerate to offload a large model (https://huggingface.co/TheBloke/WizardCoder-33B-V1.1-GPTQ) to CPU, with following code (requires #2383 if using Intel GPU, and huggingface/transformers#28755):

import datetime

import torch
import intel_extension_for_pytorch

import accelerate
from transformers import AutoModelForCausalLM, AutoTokenizer, pipeline

model_id = "/mnt/external2/LLMs/WizardCoder-33B-V1.1-GPTQ"
tokenizer = AutoTokenizer.from_pretrained(model_id)

pipe = pipeline("text-generation", model=model_id, model_kwargs={"torch_dtype": torch.bfloat16, "device_map": "auto", "max_memory": {0: "16GB", "cpu": "128GB"}, "offload_buffers": True})
print(str(pipe.model.hf_device_map))

print(str(datetime.datetime.now()) + " Generating...")
results = pipe("public void helloWorld() {")

print(str(datetime.datetime.now()) + " Output:")
print(results)

I have one Intel Arc A770 16GB GPU in my machine, but the code above always OOM on the GPU.

After some digging, I found that in the utils/modeling.py:

def compute_module_sizes(
    # ... (other code) ...
            size = tensor.numel() * min(dtype_size, dtype_byte_size(tensor.dtype))  # here, dtype_size = 2, dtype_byte_size = 4

This will get wrong size 2 for that model TheBloke/WizardCoder-33B-V1.1-GPTQ, because it contains int32 weights, which size should be 4, so the layer size has been under-estimated, causing an OOM on real load.

The int32 weight won't be converted because of these lines in utils/modeling.py:

def set_module_tensor_to_device(
  # ... (other code) ...
        elif not str(value.dtype).startswith(("torch.uint", "torch.int", "torch.bool")):  # This line will ignore int32 values
            value = value.to(dtype)

So I added a check in compute_module_sizes, matching the set_module_tensor_to_device.

With this PR, the layer containing int32 values can get correct size, and my code can finish correctly without OOM.

Before submitting

• This PR fixes a typo or improves the docs (you can dismiss the other checks if that's the case).
• Did you read the contributor guideline,
  Pull Request section?
• Was this discussed/approved via a Github issue or the forum? Please add a link
  to it if that's the case.
• Did you make sure to update the documentation with your changes? Here are the
  documentation guidelines, and
  here are tips on formatting docstrings.
• Did you write any new necessary tests?

Who can review?

Anyone in the community is free to review the PR once the tests have passed. Feel free to tag
members/contributors who may be interested in your PR.

[SunMarc]

Great catch ! Thanks for fixing this !

[HuggingFaceDocBuilderDev]

The docs for this PR live here. All of your documentation changes will be reflected on that endpoint. The docs are available until 30 days after the last update.