leon-se / Aria-sequential_mlp-bnb_nf4

Aria-sequential_mlp-bnb_nf4

BitsAndBytes NF4 quantization from Aria-sequential_mlp , requires about 15.5 GB of VRAM and runs on a RTX 3090 and (not really practical, only without device_map=auto ) on a RTX 4060 Ti 16 GB. Currently the model is not 5 GB sharded, as this seems to cause problems when loading serialized BNB models. This might make it impossible to load the model in free-tier Colab.

Installation

pip install transformers==4.45.0 accelerate==0.34.1 sentencepiece==0.2.0 torchvision requests torch Pillow bitsandbytes
pip install flash-attn --no-build-isolation

Inference

Run this model with:

import requests
import torch
from PIL import Image
from transformers import AutoModelForCausalLM, AutoProcessor, BitsAndBytesConfig
torch.cuda.set_device(0)

model_id_or_path = "thwin27/Aria-sequential_mlp-bnb_nf4"

model = AutoModelForCausalLM.from_pretrained(model_id_or_path, device_map="auto", torch_dtype=torch.bfloat16, trust_remote_code=True)
processor = AutoProcessor.from_pretrained(model_id_or_path, trust_remote_code=True)

image_path = "https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/diffusers/cat.png"

image = Image.open(requests.get(image_path, stream=True).raw)

messages = [
    {
        "role": "user",
        "content": [
            {"text": None, "type": "image"},
            {"text": "what is the image?", "type": "text"},
        ],
    }
]

text = processor.apply_chat_template(messages, add_generation_prompt=True)
inputs = processor(text=text, images=image, return_tensors="pt")
inputs["pixel_values"] = inputs["pixel_values"].to(model.dtype)
inputs = {k: v.to(model.device) for k, v in inputs.items()}

with torch.inference_mode(), torch.amp.autocast("cuda", dtype=torch.bfloat16):
    output = model.generate(
        **inputs,
        max_new_tokens=500,
        stop_strings=["<|im_end|>"],
        tokenizer=processor.tokenizer,
        do_sample=True,
        temperature=0.9,
    )
    output_ids = output[0][inputs["input_ids"].shape[1]:]
    result = processor.decode(output_ids, skip_special_tokens=True)

print(result)
print(f'Max allocated memory: {torch.cuda.max_memory_allocated(device="cuda") / 1024 ** 3:.3f}GiB')

Quantization

Quantization created with:

from transformers import AutoModelForCausalLM, AutoTokenizer, BitsAndBytesConfig

model_id = "rhymes-ai/Aria-sequential_mlp"

nf4_config = BitsAndBytesConfig(
    load_in_4bit=True,
    bnb_4bit_quant_type="nf4",
    bnb_4bit_compute_dtype=torch.bfloat16,
    bnb_4bit_use_double_quant=True,
    llm_int8_enable_fp32_cpu_offload=True,
    llm_int8_skip_modules=["language_model.lm_head", "multi_modal_projector", "vision_tower"],
    )

model_nf4 = AutoModelForCausalLM.from_pretrained(model_id, quantization_config=nf4_config)