Latxa-Qwen3-VL-4B-Instruct-w4a16_nvfp4

AWQ Quantization using https://github.com/vllm-project/llm-compressor. Compatible with vLLM and huggingface transformers. See https://github.com/checkeapp-group/Fact-Checking-API/blob/main/docs/model-quantization.md for more details.

import os

from datasets import load_dataset
from llmcompressor import oneshot
from llmcompressor.modifiers.quantization import QuantizationModifier
from llmcompressor.utils import dispatch_for_generation
from transformers import AutoProcessor, AutoModelForVision2Seq

os.environ["HF_HUB_ENABLE_HF_TRANSFER"] = "1"
MODEL_ID = "HiTZ/Latxa-Qwen3-VL-4B-Instruct"

# Load model.
model = AutoModelForVision2Seq.from_pretrained(MODEL_ID, torch_dtype="auto")
processor = AutoProcessor.from_pretrained(MODEL_ID)


DATASET_ID = "Iker/calibration-dataset"
DATASET_SPLIT = "train"

# Select number of samples. 512 samples is a good place to start.
# Increasing the number of samples can improve accuracy.
NUM_CALIBRATION_SAMPLES = 2048
MAX_SEQUENCE_LENGTH = 8192

# Load dataset and preprocess.
ds = load_dataset(DATASET_ID, split="train")
ds = ds.shuffle(seed=42)
ds = ds.select(range(NUM_CALIBRATION_SAMPLES))


def preprocess(example):
    return {
        "text": processor.apply_chat_template(
            example["messages"],
            tokenize=False,
        )
    }


ds = ds.map(preprocess)


# Tokenize inputs.
def tokenize(sample):
    return processor.tokenizer(
        sample["text"],
        padding=False,
        max_length=MAX_SEQUENCE_LENGTH,
        truncation=True,
        add_special_tokens=False,
    )


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

# Configure the quantization algorithm and scheme.
# In this case, we:
#   * quantize the weights to fp4 with per group 16 via ptq
#   * calibrate a global_scale for activations, which will be used to
#       quantize activations to fp4 on the fly
recipe = QuantizationModifier(targets="Linear", scheme="W4A16_ASYM", ignore=["lm_head"])

# Apply quantization.
oneshot(
    model=model,
    dataset=ds,
    recipe=recipe,
    max_seq_length=MAX_SEQUENCE_LENGTH,
    num_calibration_samples=NUM_CALIBRATION_SAMPLES,
)


print("\n\n")
print("========== SAMPLE GENERATION ==============")
dispatch_for_generation(model)
messages = [
    {
        "role": "user",
        "content": [
            {
                "type": "image",
                "url": "https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/p-blog/candy.JPG",
            },
            {"type": "text", "text": "What animal is on the candy?"},
        ],
    },
]
inputs = processor.apply_chat_template(
    messages,
    add_generation_prompt=True,
    tokenize=True,
    return_dict=True,
    return_tensors="pt",
).to(model.device)

# output = model.generate(**inputs, max_new_tokens=40)
# print(processor.decode(output[0][inputs["input_ids"].shape[-1] :]))
print("==========================================\n\n")


# Save to disk in compressed-tensors format.
SAVE_DIR = MODEL_ID.rstrip("/").split("/")[-1] + "-w4a16_nvfp4"
model.save_pretrained(SAVE_DIR, save_compressed=True)
processor.save_pretrained(SAVE_DIR)


model.push_to_hub("Iker/" + SAVE_DIR)
processor.push_to_hub("Iker/" + SAVE_DIR)