jester6136

/

nanonets-ocr-w4a16-g128

like 0

Sorry for the late response!

Here’s my script for quantizing a VLM. You should run calibration on your own dataset (the one you’ll actually use the model on):

import base64
from io import BytesIO

import torch
from datasets import load_dataset
from qwen_vl_utils import process_vision_info
from transformers import AutoProcessor, Qwen2_5_VLForConditionalGeneration
from llmcompressor.modifiers.quantization import QuantizationModifier
from llmcompressor.modifiers.quantization import GPTQModifier
from llmcompressor import oneshot
from llmcompressor.utils import dispatch_for_generation
from compressed_tensors.quantization.quant_args import (
QuantizationArgs,
QuantizationStrategy,
QuantizationType,
)

# Load model.
model_id = "nanonets/Nanonets-OCR-s"
model = Qwen2_5_VLForConditionalGeneration.from_pretrained(model_id, torch_dtype="auto")
processor = AutoProcessor.from_pretrained(model_id, trust_remote_code=True)

# Oneshot arguments
DATASET_ID = "nielsr/funsd"
DATASET_SPLIT = "train[:512]"
NUM_CALIBRATION_SAMPLES = 512
MAX_SEQUENCE_LENGTH = 10000

# Load dataset and preprocess.
ds = load_dataset(DATASET_ID, split=DATASET_SPLIT)
ds = ds.shuffle(seed=42)


# Apply chat template and tokenize inputs.
def preprocess_and_tokenize(example):
    # preprocess
    buffered = BytesIO()
    example["image"].save(buffered, format="PNG")
    encoded_image = base64.b64encode(buffered.getvalue())
    encoded_image_text = encoded_image.decode("utf-8")
    base64_qwen = f"data:image;base64,{encoded_image_text}"
    messages = [
        {
            "role": "user",
            "content": [
                {"type": "image", "image": base64_qwen},
                {"type": "text", "text": "Extract the text from the above document as if you were reading it naturally. Text extracted as Markdown format. Return the tables as htlm format."},
            ],
        }
    ]
    text = processor.apply_chat_template(
        messages, tokenize=False, add_generation_prompt=True
    )

    image_inputs, video_inputs = process_vision_info(messages)

    # tokenize
    return processor(
        text=[text],
        images=image_inputs,
        videos=video_inputs,
        padding=False,
        max_length=MAX_SEQUENCE_LENGTH,
        truncation=True,
    )


ds = ds.map(preprocess_and_tokenize, remove_columns=ds.column_names)


# Define a oneshot data collator for multimodal inputs.
def data_collator(batch):
    assert len(batch) == 1
    return {key: torch.tensor(value) for key, value in batch[0].items()}


# Recipe
recipe = [
    GPTQModifier(
        targets="Linear",
        scheme="W8A8",
        ignore=["lm_head", "re:visual.*", "re:model.visual.*"],
    ),
]


# Perform oneshot
oneshot(
    model=model,
    tokenizer=model_id,
    dataset=ds,
    recipe=recipe,
    max_seq_length=MAX_SEQUENCE_LENGTH,
    num_calibration_samples=NUM_CALIBRATION_SAMPLES,
    trust_remote_code_model=True,
    data_collator=data_collator,
)

# Confirm generations of the quantized model look sane.
print("========== SAMPLE GENERATION ==============")
dispatch_for_generation(model)
messages = [
    {
        "role": "user",
        "content": [
            {
                "type": "image",
                "image": "https://i1.rgstatic.net/publication/293322356_Are_Your_Digital_Documents_Web_Friendly_Making_Scanned_Documents_Web_Accessible/links/574846a408ae2301b0b97d32/largepreview.png",
            },
            {"type": "text", "text": "Extract the text from the above document as if you were reading it naturally. Text extracted as Markdown format. Return the tables as htlm format."},
        ],
    }
]
prompt = processor.apply_chat_template(messages, add_generation_prompt=True)
image_inputs, video_inputs = process_vision_info(messages)
inputs = processor(
    text=[prompt],
    images=image_inputs,
    videos=video_inputs,
    padding=False,
    max_length=MAX_SEQUENCE_LENGTH,
    truncation=True,
    return_tensors="pt",
).to("cuda")
output = model.generate(**inputs, max_new_tokens=100)
print(processor.decode(output[0], skip_special_tokens=True))
print("==========================================")


# Save to disk compressed.
SAVE_DIR = model_id.rstrip("/").split("/")[-1] + "-w8a8-1"
model.save_pretrained(SAVE_DIR, save_compressed=True)
processor.save_pretrained(SAVE_DIR)

thanks!