neuralmagic/Llama-3.2-11B-Vision-Instruct-quantized.w4a16

Llama-3.2-11B-Vision-Instruct-quantized.w4a16

Model Overview

• Model Architecture: Llama-3.2-11B-Vision-Instruct
  • Input: Vision-Text
  • Output: Text
• Model Optimizations:
  • Weight quantization: INT4
  • Activation quantization: FP16
• Release Date: 1/31/2025
• Version: 1.0
• Model Developers: Neural Magic

Quantized version of meta-llama/Llama-3.2-11B-Vision-Instruct.

Model Optimizations

This model was obtained by quantizing the weights of meta-llama/Llama-3.2-11B-Vision-Instruct to INT4 data type, ready for inference with vLLM >= 0.5.2.

Deployment

Use with vLLM

This model can be deployed efficiently using the vLLM backend, as shown in the example below.

from transformers import AutoProcessor
from vllm.assets.image import ImageAsset
from vllm import LLM, SamplingParams

# prepare model
model_id = "neuralmagic/Llama-3.2-11B-Vision-Instruct-quantized.w4a16"
llm = LLM(
    model=model_id,
    max_model_len=4096,
    max_num_seqs=16,
    limit_mm_per_prompt={"image": 1},
)
processor = AutoProcessor.from_pretrained(model_id)


# prepare inputs
question = "What is the content of this image?"
messages = [
    {
        "role": "user",
        "content": [
            {"type": "image"},
            {"type": "text", "text": f"{question}"},
        ],
    },
]
prompt = processor.apply_chat_template(
    messages, add_generation_prompt=True,tokenize=False
)
image = ImageAsset("cherry_blossom").pil_image.convert("RGB")
inputs = {
    "prompt": prompt,
    "multi_modal_data": {
        "image": image
    },
}

# generate response
print("========== SAMPLE GENERATION ==============")
outputs = llm.generate(inputs, SamplingParams(temperature=0.2, max_tokens=64))
print(f"PROMPT  : {outputs[0].prompt}")
print(f"RESPONSE: {outputs[0].outputs[0].text}")
print("==========================================")

vLLM also supports OpenAI-compatible serving. See the documentation for more details.

Creation

This model was created with llm-compressor by running the code snippet below as part a multimodal announcement blog.

import requests
import torch
from PIL import Image
from transformers import AutoProcessor

from llmcompressor.modifiers.quantization import GPTQModifier
from llmcompressor.transformers import oneshot
from llmcompressor.transformers.tracing import TraceableMllamaForConditionalGeneration

# Load model.
model_id = "meta-llama/Llama-3.2-11B-Vision-Instruct"
model = TraceableMllamaForConditionalGeneration.from_pretrained(
    model_id, device_map="auto", torch_dtype="auto"
)
processor = AutoProcessor.from_pretrained(model_id, trust_remote_code=True)

# Oneshot arguments
DATASET_ID = "flickr30k"
DATASET_SPLIT = {"calibration": "test[:512]"}
NUM_CALIBRATION_SAMPLES = 512
MAX_SEQUENCE_LENGTH = 2048


# 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="W4A16",
        ignore=["re:.*lm_head", "re:multi_modal_projector.*", "re:vision_model.*"],
    ),
]

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

License

License: Use of Llama 3.2 is governed by the Llama 3.2 Community License (a custom, commercial license agreement).