Visual-language assistant with nanoLLaVA and OpenVINO#

This Jupyter notebook can be launched on-line, opening an interactive environment in a browser window. You can also make a local installation. Choose one of the following options:

nanoLLaVA is a “small but mighty” 1B vision-language model designed to run efficiently on edge devices. It uses SigLIP-400m as Image Encoder and Qwen1.5-0.5B as LLM. In this tutorial, we consider how to convert and run nanoLLaVA model using OpenVINO. Additionally, we will optimize model using NNCF

Table of contents:

Prerequisites
Select Model
Download PyTorch model
Convert and Optimize model
Prepare model inference pipeline
Run OpenVINO Model Inference
- Select device
Interactive demo

Installation Instructions#

This is a self-contained example that relies solely on its own code.

We recommend running the notebook in a virtual environment. You only need a Jupyter server to start. For details, please refer to Installation Guide.

Prerequisites#

%pip install -q "torch>=2.1" "transformers>=4.40" "accelerate" "pillow" "gradio>=4.26" "tqdm" --extra-index-url https://download.pytorch.org/whl/cpu
%pip install -q "nncf>=2.13"
%pip install -q -U --pre --extra-index-url https://storage.openvinotoolkit.org/simple/wheels/nightly "openvino-tokenizers[transformers]" "openvino>=2024.4.0"
%pip install -q "git+https://github.com/eaidova/optimum-intel.git@ea/minicpmv"

ERROR: pip's dependency resolver does not currently take into account all the packages that are installed. This behaviour is the source of the following dependency conflicts.
mobileclip 0.1.0 requires torchvision==0.14.1, but you have torchvision 0.17.2+cpu which is incompatible.
Note: you may need to restart the kernel to use updated packages.
Note: you may need to restart the kernel to use updated packages.
ERROR: pip's dependency resolver does not currently take into account all the packages that are installed. This behaviour is the source of the following dependency conflicts.
openvino-genai 2024.4.0.0 requires openvino_tokenizers~=2024.4.0.0.dev, but you have openvino-tokenizers 2024.5.0.0.dev20241022 which is incompatible.
Note: you may need to restart the kernel to use updated packages.
ERROR: pip's dependency resolver does not currently take into account all the packages that are installed. This behaviour is the source of the following dependency conflicts.
mobileclip 0.1.0 requires torchvision==0.14.1, but you have torchvision 0.17.2+cpu which is incompatible.
Note: you may need to restart the kernel to use updated packages.

from pathlib import Path
import requests

helper_file = Path("ov_nano_llava_helper.py")

if not helper_file.exists():
    r = requests.get(
        url=f"https://raw.githubusercontent.com/openvinotoolkit/openvino_notebooks/latest/notebooks/nano-llava-multimodal-chatbot/{helper_file.name}"
    )
    helper_file.open("w").write(r.text)

Select Model#

The tutorial supports the following models from Phi-3 model family:

You can select one from the provided options below.

import ipywidgets as widgets

model_ids = ["qnguyen3/nanoLLaVA", "qnguyen3/nanoLLaVA-1.5"]

model_dropdown = widgets.Dropdown(
    options=model_ids,
    value=model_ids[0],
    description="Model:",
    disabled=False,
)

model_dropdown

Dropdown(description='Model:', options=('qnguyen3/nanoLLaVA', 'qnguyen3/nanoLLaVA-1.5'), value='qnguyen3/nanoL…

Download PyTorch model#

from ov_nano_llava_helper import download_original_model, converted_model_exists, copy_model_files

model_id = model_dropdown.value
model_dir = Path(model_id.split("/")[-1])
ov_model_dir = Path("ov_" + model_dir.name) / "FP16"

if not converted_model_exists(ov_model_dir):
    download_original_model(model_id, model_dir)

Fetching 14 files:   0%|          | 0/14 [00:00<?, ?it/s]

README.md:   0%|          | 0.00/3.73k [00:00<?, ?B/s]

example_1.png:   0%|          | 0.00/200k [00:00<?, ?B/s]

.gitattributes:   0%|          | 0.00/1.52k [00:00<?, ?B/s]

Convert and Optimize model#

Our model conversion and optimization consist of following steps: 1. Convert model to OpenVINO format and save it on disk. 2. Compress model weights using NNCF

Let’s consider each step deeply.

Convert model to OpenVINO IR format#

NanoLLaVA implementation is based on HuggingFace Transformers library. For convenience, we will use OpenVINO integration with HuggingFace Optimum. Optimum Intel is the interface between the Transformers and Diffusers libraries and the different tools and libraries provided by Intel to accelerate end-to-end pipelines on Intel architectures.

Among other use cases, Optimum Intel provides a simple interface to optimize your Transformers and Diffusers models, convert them to the OpenVINO Intermediate Representation (IR) format and run inference using OpenVINO Runtime. optimum-cli provides command line interface for model conversion and optimization.

General command format:

optimum-cli export openvino --model <model_id_or_path> --task <task> <output_dir>

where task is task to export the model for, if not specified, the task will be auto-inferred based on the model. You can find a mapping between tasks and model classes in Optimum TaskManager documentation. Additionally, you can specify weights compression using --weight-format argument with one of following options: fp32, fp16, int8 and int4. Fro int8 and int4 nncf will be used for weight compression. More details about model export provided in Optimum Intel documentation.

if not converted_model_exists(ov_model_dir):
    !optimum-cli export openvino --model {model_id} --task image-text-to-text --trust-remote-code --weight-format fp16 {ov_model_dir}

2024-11-05 02:09:06.532025: I tensorflow/core/util/port.cc:110] oneDNN custom operations are on. You may see slightly different numerical results due to floating-point round-off errors from different computation orders. To turn them off, set the environment variable TF_ENABLE_ONEDNN_OPTS=0.
2024-11-05 02:09:06.565023: I tensorflow/core/platform/cpu_feature_guard.cc:182] This TensorFlow binary is optimized to use available CPU instructions in performance-critical operations.
To enable the following instructions: AVX2 AVX512F AVX512_VNNI FMA, in other operations, rebuild TensorFlow with the appropriate compiler flags.
2024-11-05 02:09:07.173118: W tensorflow/compiler/tf2tensorrt/utils/py_utils.cc:38] TF-TRT Warning: Could not find TensorRT
Some weights of the model checkpoint at qnguyen3/nanoLLaVA were not used when initializing LlavaQwen2ForCausalLM: ['model.vision_tower.vision_tower.vision_model.embeddings.patch_embedding.bias', 'model.vision_tower.vision_tower.vision_model.embeddings.patch_embedding.weight', 'model.vision_tower.vision_tower.vision_model.embeddings.position_embedding.weight', 'model.vision_tower.vision_tower.vision_model.encoder.layers.0.layer_norm1.bias', 'model.vision_tower.vision_tower.vision_model.encoder.layers.0.layer_norm1.weight', 'model.vision_tower.vision_tower.vision_model.encoder.layers.0.layer_norm2.bias', 'model.vision_tower.vision_tower.vision_model.encoder.layers.0.layer_norm2.weight', 'model.vision_tower.vision_tower.vision_model.encoder.layers.0.mlp.fc1.bias', 'model.vision_tower.vision_tower.vision_model.encoder.layers.0.mlp.fc1.weight', 'model.vision_tower.vision_tower.vision_model.encoder.layers.0.mlp.fc2.bias', 'model.vision_tower.vision_tower.vision_model.encoder.layers.0.mlp.fc2.weight', 'model.vision_tower.vision_tower.vision_model.encoder.layers.0.self_attn.k_proj.bias', 'model.vision_tower.vision_tower.vision_model.encoder.layers.0.self_attn.k_proj.weight', 'model.vision_tower.vision_tower.vision_model.encoder.layers.0.self_attn.out_proj.bias', 'model.vision_tower.vision_tower.vision_model.encoder.layers.0.self_attn.out_proj.weight', 'model.vision_tower.vision_tower.vision_model.encoder.layers.0.self_attn.q_proj.bias', 'model.vision_tower.vision_tower.vision_model.encoder.layers.0.self_attn.q_proj.weight', 'model.vision_tower.vision_tower.vision_model.encoder.layers.0.self_attn.v_proj.bias', 'model.vision_tower.vision_tower.vision_model.encoder.layers.0.self_attn.v_proj.weight', 'model.vision_tower.vision_tower.vision_model.encoder.layers.1.layer_norm1.bias', 'model.vision_tower.vision_tower.vision_model.encoder.layers.1.layer_norm1.weight', 'model.vision_tower.vision_tower.vision_model.encoder.layers.1.layer_norm2.bias', 'model.vision_tower.vision_tower.vision_model.encoder.layers.1.layer_norm2.weight', 'model.vision_tower.vision_tower.vision_model.encoder.layers.1.mlp.fc1.bias', 'model.vision_tower.vision_tower.vision_model.encoder.layers.1.mlp.fc1.weight', 'model.vision_tower.vision_tower.vision_model.encoder.layers.1.mlp.fc2.bias', 'model.vision_tower.vision_tower.vision_model.encoder.layers.1.mlp.fc2.weight', 'model.vision_tower.vision_tower.vision_model.encoder.layers.1.self_attn.k_proj.bias', 'model.vision_tower.vision_tower.vision_model.encoder.layers.1.self_attn.k_proj.weight', 'model.vision_tower.vision_tower.vision_model.encoder.layers.1.self_attn.out_proj.bias', 'model.vision_tower.vision_tower.vision_model.encoder.layers.1.self_attn.out_proj.weight', 'model.vision_tower.vision_tower.vision_model.encoder.layers.1.self_attn.q_proj.bias', 'model.vision_tower.vision_tower.vision_model.encoder.layers.1.self_attn.q_proj.weight', 'model.vision_tower.vision_tower.vision_model.encoder.layers.1.self_attn.v_proj.bias', 'model.vision_tower.vision_tower.vision_model.encoder.layers.1.self_attn.v_proj.weight', 'model.vision_tower.vision_tower.vision_model.encoder.layers.10.layer_norm1.bias', 'model.vision_tower.vision_tower.vision_model.encoder.layers.10.layer_norm1.weight', 'model.vision_tower.vision_tower.vision_model.encoder.layers.10.layer_norm2.bias', 'model.vision_tower.vision_tower.vision_model.encoder.layers.10.layer_norm2.weight', 'model.vision_tower.vision_tower.vision_model.encoder.layers.10.mlp.fc1.bias', 'model.vision_tower.vision_tower.vision_model.encoder.layers.10.mlp.fc1.weight', 'model.vision_tower.vision_tower.vision_model.encoder.layers.10.mlp.fc2.bias', 'model.vision_tower.vision_tower.vision_model.encoder.layers.10.mlp.fc2.weight', 'model.vision_tower.vision_tower.vision_model.encoder.layers.10.self_attn.k_proj.bias', 'model.vision_tower.vision_tower.vision_model.encoder.layers.10.self_attn.k_proj.weight', 'model.vision_tower.vision_tower.vision_model.encoder.layers.10.self_attn.out_proj.bias', 'model.vision_tower.vision_tower.vision_model.encoder.layers.10.self_attn.out_proj.weight', 'model.vision_tower.vision_tower.vision_model.encoder.layers.10.self_attn.q_proj.bias', 'model.vision_tower.vision_tower.vision_model.encoder.layers.10.self_attn.q_proj.weight', 'model.vision_tower.vision_tower.vision_model.encoder.layers.10.self_attn.v_proj.bias', 'model.vision_tower.vision_tower.vision_model.encoder.layers.10.self_attn.v_proj.weight', 'model.vision_tower.vision_tower.vision_model.encoder.layers.11.layer_norm1.bias', 'model.vision_tower.vision_tower.vision_model.encoder.layers.11.layer_norm1.weight', 'model.vision_tower.vision_tower.vision_model.encoder.layers.11.layer_norm2.bias', 'model.vision_tower.vision_tower.vision_model.encoder.layers.11.layer_norm2.weight', 'model.vision_tower.vision_tower.vision_model.encoder.layers.11.mlp.fc1.bias', 'model.vision_tower.vision_tower.vision_model.encoder.layers.11.mlp.fc1.weight', 'model.vision_tower.vision_tower.vision_model.encoder.layers.11.mlp.fc2.bias', 'model.vision_tower.vision_tower.vision_model.encoder.layers.11.mlp.fc2.weight', 'model.vision_tower.vision_tower.vision_model.encoder.layers.11.self_attn.k_proj.bias', 'model.vision_tower.vision_tower.vision_model.encoder.layers.11.self_attn.k_proj.weight', 'model.vision_tower.vision_tower.vision_model.encoder.layers.11.self_attn.out_proj.bias', 'model.vision_tower.vision_tower.vision_model.encoder.layers.11.self_attn.out_proj.weight', 'model.vision_tower.vision_tower.vision_model.encoder.layers.11.self_attn.q_proj.bias', 'model.vision_tower.vision_tower.vision_model.encoder.layers.11.self_attn.q_proj.weight', 'model.vision_tower.vision_tower.vision_model.encoder.layers.11.self_attn.v_proj.bias', 'model.vision_tower.vision_tower.vision_model.encoder.layers.11.self_attn.v_proj.weight', 'model.vision_tower.vision_tower.vision_model.encoder.layers.12.layer_norm1.bias', 'model.vision_tower.vision_tower.vision_model.encoder.layers.12.layer_norm1.weight', 'model.vision_tower.vision_tower.vision_model.encoder.layers.12.layer_norm2.bias', 'model.vision_tower.vision_tower.vision_model.encoder.layers.12.layer_norm2.weight', 'model.vision_tower.vision_tower.vision_model.encoder.layers.12.mlp.fc1.bias', 'model.vision_tower.vision_tower.vision_model.encoder.layers.12.mlp.fc1.weight', 'model.vision_tower.vision_tower.vision_model.encoder.layers.12.mlp.fc2.bias', 'model.vision_tower.vision_tower.vision_model.encoder.layers.12.mlp.fc2.weight', 'model.vision_tower.vision_tower.vision_model.encoder.layers.12.self_attn.k_proj.bias', 'model.vision_tower.vision_tower.vision_model.encoder.layers.12.self_attn.k_proj.weight', 'model.vision_tower.vision_tower.vision_model.encoder.layers.12.self_attn.out_proj.bias', 'model.vision_tower.vision_tower.vision_model.encoder.layers.12.self_attn.out_proj.weight', 'model.vision_tower.vision_tower.vision_model.encoder.layers.12.self_attn.q_proj.bias', 'model.vision_tower.vision_tower.vision_model.encoder.layers.12.self_attn.q_proj.weight', 'model.vision_tower.vision_tower.vision_model.encoder.layers.12.self_attn.v_proj.bias', 'model.vision_tower.vision_tower.vision_model.encoder.layers.12.self_attn.v_proj.weight', 'model.vision_tower.vision_tower.vision_model.encoder.layers.13.layer_norm1.bias', 'model.vision_tower.vision_tower.vision_model.encoder.layers.13.layer_norm1.weight', 'model.vision_tower.vision_tower.vision_model.encoder.layers.13.layer_norm2.bias', 'model.vision_tower.vision_tower.vision_model.encoder.layers.13.layer_norm2.weight', 'model.vision_tower.vision_tower.vision_model.encoder.layers.13.mlp.fc1.bias', 'model.vision_tower.vision_tower.vision_model.encoder.layers.13.mlp.fc1.weight', 'model.vision_tower.vision_tower.vision_model.encoder.layers.13.mlp.fc2.bias', 'model.vision_tower.vision_tower.vision_model.encoder.layers.13.mlp.fc2.weight', 'model.vision_tower.vision_tower.vision_model.encoder.layers.13.self_attn.k_proj.bias', 'model.vision_tower.vision_tower.vision_model.encoder.layers.13.self_attn.k_proj.weight', 'model.vision_tower.vision_tower.vision_model.encoder.layers.13.self_attn.out_proj.bias', 'model.vision_tower.vision_tower.vision_model.encoder.layers.13.self_attn.out_proj.weight', 'model.vision_tower.vision_tower.vision_model.encoder.layers.13.self_attn.q_proj.bias', 'model.vision_tower.vision_tower.vision_model.encoder.layers.13.self_attn.q_proj.weight', 'model.vision_tower.vision_tower.vision_model.encoder.layers.13.self_attn.v_proj.bias', 'model.vision_tower.vision_tower.vision_model.encoder.layers.13.self_attn.v_proj.weight', 'model.vision_tower.vision_tower.vision_model.encoder.layers.14.layer_norm1.bias', 'model.vision_tower.vision_tower.vision_model.encoder.layers.14.layer_norm1.weight', 'model.vision_tower.vision_tower.vision_model.encoder.layers.14.layer_norm2.bias', 'model.vision_tower.vision_tower.vision_model.encoder.layers.14.layer_norm2.weight', 'model.vision_tower.vision_tower.vision_model.encoder.layers.14.mlp.fc1.bias', 'model.vision_tower.vision_tower.vision_model.encoder.layers.14.mlp.fc1.weight', 'model.vision_tower.vision_tower.vision_model.encoder.layers.14.mlp.fc2.bias', 'model.vision_tower.vision_tower.vision_model.encoder.layers.14.mlp.fc2.weight', 'model.vision_tower.vision_tower.vision_model.encoder.layers.14.self_attn.k_proj.bias', 'model.vision_tower.vision_tower.vision_model.encoder.layers.14.self_attn.k_proj.weight', 'model.vision_tower.vision_tower.vision_model.encoder.layers.14.self_attn.out_proj.bias', 'model.vision_tower.vision_tower.vision_model.encoder.layers.14.self_attn.out_proj.weight', 'model.vision_tower.vision_tower.vision_model.encoder.layers.14.self_attn.q_proj.bias', 'model.vision_tower.vision_tower.vision_model.encoder.layers.14.self_attn.q_proj.weight', 'model.vision_tower.vision_tower.vision_model.encoder.layers.14.self_attn.v_proj.bias', 'model.vision_tower.vision_tower.vision_model.encoder.layers.14.self_attn.v_proj.weight', 'model.vision_tower.vision_tower.vision_model.encoder.layers.15.layer_norm1.bias', 'model.vision_tower.vision_tower.vision_model.encoder.layers.15.layer_norm1.weight', 'model.vision_tower.vision_tower.vision_model.encoder.layers.15.layer_norm2.bias', 'model.vision_tower.vision_tower.vision_model.encoder.layers.15.layer_norm2.weight', 'model.vision_tower.vision_tower.vision_model.encoder.layers.15.mlp.fc1.bias', 'model.vision_tower.vision_tower.vision_model.encoder.layers.15.mlp.fc1.weight', 'model.vision_tower.vision_tower.vision_model.encoder.layers.15.mlp.fc2.bias', 'model.vision_tower.vision_tower.vision_model.encoder.layers.15.mlp.fc2.weight', 'model.vision_tower.vision_tower.vision_model.encoder.layers.15.self_attn.k_proj.bias', 'model.vision_tower.vision_tower.vision_model.encoder.layers.15.self_attn.k_proj.weight', 'model.vision_tower.vision_tower.vision_model.encoder.layers.15.self_attn.out_proj.bias', 'model.vision_tower.vision_tower.vision_model.encoder.layers.15.self_attn.out_proj.weight', 'model.vision_tower.vision_tower.vision_model.encoder.layers.15.self_attn.q_proj.bias', 'model.vision_tower.vision_tower.vision_model.encoder.layers.15.self_attn.q_proj.weight', 'model.vision_tower.vision_tower.vision_model.encoder.layers.15.self_attn.v_proj.bias', 'model.vision_tower.vision_tower.vision_model.encoder.layers.15.self_attn.v_proj.weight', 'model.vision_tower.vision_tower.vision_model.encoder.layers.16.layer_norm1.bias', 'model.vision_tower.vision_tower.vision_model.encoder.layers.16.layer_norm1.weight', 'model.vision_tower.vision_tower.vision_model.encoder.layers.16.layer_norm2.bias', 'model.vision_tower.vision_tower.vision_model.encoder.layers.16.layer_norm2.weight', 'model.vision_tower.vision_tower.vision_model.encoder.layers.16.mlp.fc1.bias', 'model.vision_tower.vision_tower.vision_model.encoder.layers.16.mlp.fc1.weight', 'model.vision_tower.vision_tower.vision_model.encoder.layers.16.mlp.fc2.bias', 'model.vision_tower.vision_tower.vision_model.encoder.layers.16.mlp.fc2.weight', 'model.vision_tower.vision_tower.vision_model.encoder.layers.16.self_attn.k_proj.bias', 'model.vision_tower.vision_tower.vision_model.encoder.layers.16.self_attn.k_proj.weight', 'model.vision_tower.vision_tower.vision_model.encoder.layers.16.self_attn.out_proj.bias', 'model.vision_tower.vision_tower.vision_model.encoder.layers.16.self_attn.out_proj.weight', 'model.vision_tower.vision_tower.vision_model.encoder.layers.16.self_attn.q_proj.bias', 'model.vision_tower.vision_tower.vision_model.encoder.layers.16.self_attn.q_proj.weight', 'model.vision_tower.vision_tower.vision_model.encoder.layers.16.self_attn.v_proj.bias', 'model.vision_tower.vision_tower.vision_model.encoder.layers.16.self_attn.v_proj.weight', 'model.vision_tower.vision_tower.vision_model.encoder.layers.17.layer_norm1.bias', 'model.vision_tower.vision_tower.vision_model.encoder.layers.17.layer_norm1.weight', 'model.vision_tower.vision_tower.vision_model.encoder.layers.17.layer_norm2.bias', 'model.vision_tower.vision_tower.vision_model.encoder.layers.17.layer_norm2.weight', 'model.vision_tower.vision_tower.vision_model.encoder.layers.17.mlp.fc1.bias', 'model.vision_tower.vision_tower.vision_model.encoder.layers.17.mlp.fc1.weight', 'model.vision_tower.vision_tower.vision_model.encoder.layers.17.mlp.fc2.bias', 'model.vision_tower.vision_tower.vision_model.encoder.layers.17.mlp.fc2.weight', 'model.vision_tower.vision_tower.vision_model.encoder.layers.17.self_attn.k_proj.bias', 'model.vision_tower.vision_tower.vision_model.encoder.layers.17.self_attn.k_proj.weight', 'model.vision_tower.vision_tower.vision_model.encoder.layers.17.self_attn.out_proj.bias', 'model.vision_tower.vision_tower.vision_model.encoder.layers.17.self_attn.out_proj.weight', 'model.vision_tower.vision_tower.vision_model.encoder.layers.17.self_attn.q_proj.bias', 'model.vision_tower.vision_tower.vision_model.encoder.layers.17.self_attn.q_proj.weight', 'model.vision_tower.vision_tower.vision_model.encoder.layers.17.self_attn.v_proj.bias', 'model.vision_tower.vision_tower.vision_model.encoder.layers.17.self_attn.v_proj.weight', 'model.vision_tower.vision_tower.vision_model.encoder.layers.18.layer_norm1.bias', 'model.vision_tower.vision_tower.vision_model.encoder.layers.18.layer_norm1.weight', 'model.vision_tower.vision_tower.vision_model.encoder.layers.18.layer_norm2.bias', 'model.vision_tower.vision_tower.vision_model.encoder.layers.18.layer_norm2.weight', 'model.vision_tower.vision_tower.vision_model.encoder.layers.18.mlp.fc1.bias', 'model.vision_tower.vision_tower.vision_model.encoder.layers.18.mlp.fc1.weight', 'model.vision_tower.vision_tower.vision_model.encoder.layers.18.mlp.fc2.bias', 'model.vision_tower.vision_tower.vision_model.encoder.layers.18.mlp.fc2.weight', 'model.vision_tower.vision_tower.vision_model.encoder.layers.18.self_attn.k_proj.bias', 'model.vision_tower.vision_tower.vision_model.encoder.layers.18.self_attn.k_proj.weight', 'model.vision_tower.vision_tower.vision_model.encoder.layers.18.self_attn.out_proj.bias', 'model.vision_tower.vision_tower.vision_model.encoder.layers.18.self_attn.out_proj.weight', 'model.vision_tower.vision_tower.vision_model.encoder.layers.18.self_attn.q_proj.bias', 'model.vision_tower.vision_tower.vision_model.encoder.layers.18.self_attn.q_proj.weight', 'model.vision_tower.vision_tower.vision_model.encoder.layers.18.self_attn.v_proj.bias', 'model.vision_tower.vision_tower.vision_model.encoder.layers.18.self_attn.v_proj.weight', 'model.vision_tower.vision_tower.vision_model.encoder.layers.19.layer_norm1.bias', 'model.vision_tower.vision_tower.vision_model.encoder.layers.19.layer_norm1.weight', 'model.vision_tower.vision_tower.vision_model.encoder.layers.19.layer_norm2.bias', 'model.vision_tower.vision_tower.vision_model.encoder.layers.19.layer_norm2.weight', 'model.vision_tower.vision_tower.vision_model.encoder.layers.19.mlp.fc1.bias', 'model.vision_tower.vision_tower.vision_model.encoder.layers.19.mlp.fc1.weight', 'model.vision_tower.vision_tower.vision_model.encoder.layers.19.mlp.fc2.bias', 'model.vision_tower.vision_tower.vision_model.encoder.layers.19.mlp.fc2.weight', 'model.vision_tower.vision_tower.vision_model.encoder.layers.19.self_attn.k_proj.bias', 'model.vision_tower.vision_tower.vision_model.encoder.layers.19.self_attn.k_proj.weight', 'model.vision_tower.vision_tower.vision_model.encoder.layers.19.self_attn.out_proj.bias', 'model.vision_tower.vision_tower.vision_model.encoder.layers.19.self_attn.out_proj.weight', 'model.vision_tower.vision_tower.vision_model.encoder.layers.19.self_attn.q_proj.bias', 'model.vision_tower.vision_tower.vision_model.encoder.layers.19.self_attn.q_proj.weight', 'model.vision_tower.vision_tower.vision_model.encoder.layers.19.self_attn.v_proj.bias', 'model.vision_tower.vision_tower.vision_model.encoder.layers.19.self_attn.v_proj.weight', 'model.vision_tower.vision_tower.vision_model.encoder.layers.2.layer_norm1.bias', 'model.vision_tower.vision_tower.vision_model.encoder.layers.2.layer_norm1.weight', 'model.vision_tower.vision_tower.vision_model.encoder.layers.2.layer_norm2.bias', 'model.vision_tower.vision_tower.vision_model.encoder.layers.2.layer_norm2.weight', 'model.vision_tower.vision_tower.vision_model.encoder.layers.2.mlp.fc1.bias', 'model.vision_tower.vision_tower.vision_model.encoder.layers.2.mlp.fc1.weight', 'model.vision_tower.vision_tower.vision_model.encoder.layers.2.mlp.fc2.bias', 'model.vision_tower.vision_tower.vision_model.encoder.layers.2.mlp.fc2.weight', 'model.vision_tower.vision_tower.vision_model.encoder.layers.2.self_attn.k_proj.bias', 'model.vision_tower.vision_tower.vision_model.encoder.layers.2.self_attn.k_proj.weight', 'model.vision_tower.vision_tower.vision_model.encoder.layers.2.self_attn.out_proj.bias', 'model.vision_tower.vision_tower.vision_model.encoder.layers.2.self_attn.out_proj.weight', 'model.vision_tower.vision_tower.vision_model.encoder.layers.2.self_attn.q_proj.bias', 'model.vision_tower.vision_tower.vision_model.encoder.layers.2.self_attn.q_proj.weight', 'model.vision_tower.vision_tower.vision_model.encoder.layers.2.self_attn.v_proj.bias', 'model.vision_tower.vision_tower.vision_model.encoder.layers.2.self_attn.v_proj.weight', 'model.vision_tower.vision_tower.vision_model.encoder.layers.20.layer_norm1.bias', 'model.vision_tower.vision_tower.vision_model.encoder.layers.20.layer_norm1.weight', 'model.vision_tower.vision_tower.vision_model.encoder.layers.20.layer_norm2.bias', 'model.vision_tower.vision_tower.vision_model.encoder.layers.20.layer_norm2.weight', 'model.vision_tower.vision_tower.vision_model.encoder.layers.20.mlp.fc1.bias', 'model.vision_tower.vision_tower.vision_model.encoder.layers.20.mlp.fc1.weight', 'model.vision_tower.vision_tower.vision_model.encoder.layers.20.mlp.fc2.bias', 'model.vision_tower.vision_tower.vision_model.encoder.layers.20.mlp.fc2.weight', 'model.vision_tower.vision_tower.vision_model.encoder.layers.20.self_attn.k_proj.bias', 'model.vision_tower.vision_tower.vision_model.encoder.layers.20.self_attn.k_proj.weight', 'model.vision_tower.vision_tower.vision_model.encoder.layers.20.self_attn.out_proj.bias', 'model.vision_tower.vision_tower.vision_model.encoder.layers.20.self_attn.out_proj.weight', 'model.vision_tower.vision_tower.vision_model.encoder.layers.20.self_attn.q_proj.bias', 'model.vision_tower.vision_tower.vision_model.encoder.layers.20.self_attn.q_proj.weight', 'model.vision_tower.vision_tower.vision_model.encoder.layers.20.self_attn.v_proj.bias', 'model.vision_tower.vision_tower.vision_model.encoder.layers.20.self_attn.v_proj.weight', 'model.vision_tower.vision_tower.vision_model.encoder.layers.21.layer_norm1.bias', 'model.vision_tower.vision_tower.vision_model.encoder.layers.21.layer_norm1.weight', 'model.vision_tower.vision_tower.vision_model.encoder.layers.21.layer_norm2.bias', 'model.vision_tower.vision_tower.vision_model.encoder.layers.21.layer_norm2.weight', 'model.vision_tower.vision_tower.vision_model.encoder.layers.21.mlp.fc1.bias', 'model.vision_tower.vision_tower.vision_model.encoder.layers.21.mlp.fc1.weight', 'model.vision_tower.vision_tower.vision_model.encoder.layers.21.mlp.fc2.bias', 'model.vision_tower.vision_tower.vision_model.encoder.layers.21.mlp.fc2.weight', 'model.vision_tower.vision_tower.vision_model.encoder.layers.21.self_attn.k_proj.bias', 'model.vision_tower.vision_tower.vision_model.encoder.layers.21.self_attn.k_proj.weight', 'model.vision_tower.vision_tower.vision_model.encoder.layers.21.self_attn.out_proj.bias', 'model.vision_tower.vision_tower.vision_model.encoder.layers.21.self_attn.out_proj.weight', 'model.vision_tower.vision_tower.vision_model.encoder.layers.21.self_attn.q_proj.bias', 'model.vision_tower.vision_tower.vision_model.encoder.layers.21.self_attn.q_proj.weight', 'model.vision_tower.vision_tower.vision_model.encoder.layers.21.self_attn.v_proj.bias', 'model.vision_tower.vision_tower.vision_model.encoder.layers.21.self_attn.v_proj.weight', 'model.vision_tower.vision_tower.vision_model.encoder.layers.22.layer_norm1.bias', 'model.vision_tower.vision_tower.vision_model.encoder.layers.22.layer_norm1.weight', 'model.vision_tower.vision_tower.vision_model.encoder.layers.22.layer_norm2.bias', 'model.vision_tower.vision_tower.vision_model.encoder.layers.22.layer_norm2.weight', 'model.vision_tower.vision_tower.vision_model.encoder.layers.22.mlp.fc1.bias', 'model.vision_tower.vision_tower.vision_model.encoder.layers.22.mlp.fc1.weight', 'model.vision_tower.vision_tower.vision_model.encoder.layers.22.mlp.fc2.bias', 'model.vision_tower.vision_tower.vision_model.encoder.layers.22.mlp.fc2.weight', 'model.vision_tower.vision_tower.vision_model.encoder.layers.22.self_attn.k_proj.bias', 'model.vision_tower.vision_tower.vision_model.encoder.layers.22.self_attn.k_proj.weight', 'model.vision_tower.vision_tower.vision_model.encoder.layers.22.self_attn.out_proj.bias', 'model.vision_tower.vision_tower.vision_model.encoder.layers.22.self_attn.out_proj.weight', 'model.vision_tower.vision_tower.vision_model.encoder.layers.22.self_attn.q_proj.bias', 'model.vision_tower.vision_tower.vision_model.encoder.layers.22.self_attn.q_proj.weight', 'model.vision_tower.vision_tower.vision_model.encoder.layers.22.self_attn.v_proj.bias', 'model.vision_tower.vision_tower.vision_model.encoder.layers.22.self_attn.v_proj.weight', 'model.vision_tower.vision_tower.vision_model.encoder.layers.23.layer_norm1.bias', 'model.vision_tower.vision_tower.vision_model.encoder.layers.23.layer_norm1.weight', 'model.vision_tower.vision_tower.vision_model.encoder.layers.23.layer_norm2.bias', 'model.vision_tower.vision_tower.vision_model.encoder.layers.23.layer_norm2.weight', 'model.vision_tower.vision_tower.vision_model.encoder.layers.23.mlp.fc1.bias', 'model.vision_tower.vision_tower.vision_model.encoder.layers.23.mlp.fc1.weight', 'model.vision_tower.vision_tower.vision_model.encoder.layers.23.mlp.fc2.bias', 'model.vision_tower.vision_tower.vision_model.encoder.layers.23.mlp.fc2.weight', 'model.vision_tower.vision_tower.vision_model.encoder.layers.23.self_attn.k_proj.bias', 'model.vision_tower.vision_tower.vision_model.encoder.layers.23.self_attn.k_proj.weight', 'model.vision_tower.vision_tower.vision_model.encoder.layers.23.self_attn.out_proj.bias', 'model.vision_tower.vision_tower.vision_model.encoder.layers.23.self_attn.out_proj.weight', 'model.vision_tower.vision_tower.vision_model.encoder.layers.23.self_attn.q_proj.bias', 'model.vision_tower.vision_tower.vision_model.encoder.layers.23.self_attn.q_proj.weight', 'model.vision_tower.vision_tower.vision_model.encoder.layers.23.self_attn.v_proj.bias', 'model.vision_tower.vision_tower.vision_model.encoder.layers.23.self_attn.v_proj.weight', 'model.vision_tower.vision_tower.vision_model.encoder.layers.24.layer_norm1.bias', 'model.vision_tower.vision_tower.vision_model.encoder.layers.24.layer_norm1.weight', 'model.vision_tower.vision_tower.vision_model.encoder.layers.24.layer_norm2.bias', 'model.vision_tower.vision_tower.vision_model.encoder.layers.24.layer_norm2.weight', 'model.vision_tower.vision_tower.vision_model.encoder.layers.24.mlp.fc1.bias', 'model.vision_tower.vision_tower.vision_model.encoder.layers.24.mlp.fc1.weight', 'model.vision_tower.vision_tower.vision_model.encoder.layers.24.mlp.fc2.bias', 'model.vision_tower.vision_tower.vision_model.encoder.layers.24.mlp.fc2.weight', 'model.vision_tower.vision_tower.vision_model.encoder.layers.24.self_attn.k_proj.bias', 'model.vision_tower.vision_tower.vision_model.encoder.layers.24.self_attn.k_proj.weight', 'model.vision_tower.vision_tower.vision_model.encoder.layers.24.self_attn.out_proj.bias', 'model.vision_tower.vision_tower.vision_model.encoder.layers.24.self_attn.out_proj.weight', 'model.vision_tower.vision_tower.vision_model.encoder.layers.24.self_attn.q_proj.bias', 'model.vision_tower.vision_tower.vision_model.encoder.layers.24.self_attn.q_proj.weight', 'model.vision_tower.vision_tower.vision_model.encoder.layers.24.self_attn.v_proj.bias', 'model.vision_tower.vision_tower.vision_model.encoder.layers.24.self_attn.v_proj.weight', 'model.vision_tower.vision_tower.vision_model.encoder.layers.25.layer_norm1.bias', 'model.vision_tower.vision_tower.vision_model.encoder.layers.25.layer_norm1.weight', 'model.vision_tower.vision_tower.vision_model.encoder.layers.25.layer_norm2.bias', 'model.vision_tower.vision_tower.vision_model.encoder.layers.25.layer_norm2.weight', 'model.vision_tower.vision_tower.vision_model.encoder.layers.25.mlp.fc1.bias', 'model.vision_tower.vision_tower.vision_model.encoder.layers.25.mlp.fc1.weight', 'model.vision_tower.vision_tower.vision_model.encoder.layers.25.mlp.fc2.bias', 'model.vision_tower.vision_tower.vision_model.encoder.layers.25.mlp.fc2.weight', 'model.vision_tower.vision_tower.vision_model.encoder.layers.25.self_attn.k_proj.bias', 'model.vision_tower.vision_tower.vision_model.encoder.layers.25.self_attn.k_proj.weight', 'model.vision_tower.vision_tower.vision_model.encoder.layers.25.self_attn.out_proj.bias', 'model.vision_tower.vision_tower.vision_model.encoder.layers.25.self_attn.out_proj.weight', 'model.vision_tower.vision_tower.vision_model.encoder.layers.25.self_attn.q_proj.bias', 'model.vision_tower.vision_tower.vision_model.encoder.layers.25.self_attn.q_proj.weight', 'model.vision_tower.vision_tower.vision_model.encoder.layers.25.self_attn.v_proj.bias', 'model.vision_tower.vision_tower.vision_model.encoder.layers.25.self_attn.v_proj.weight', 'model.vision_tower.vision_tower.vision_model.encoder.layers.26.layer_norm1.bias', 'model.vision_tower.vision_tower.vision_model.encoder.layers.26.layer_norm1.weight', 'model.vision_tower.vision_tower.vision_model.encoder.layers.26.layer_norm2.bias', 'model.vision_tower.vision_tower.vision_model.encoder.layers.26.layer_norm2.weight', 'model.vision_tower.vision_tower.vision_model.encoder.layers.26.mlp.fc1.bias', 'model.vision_tower.vision_tower.vision_model.encoder.layers.26.mlp.fc1.weight', 'model.vision_tower.vision_tower.vision_model.encoder.layers.26.mlp.fc2.bias', 'model.vision_tower.vision_tower.vision_model.encoder.layers.26.mlp.fc2.weight', 'model.vision_tower.vision_tower.vision_model.encoder.layers.26.self_attn.k_proj.bias', 'model.vision_tower.vision_tower.vision_model.encoder.layers.26.self_attn.k_proj.weight', 'model.vision_tower.vision_tower.vision_model.encoder.layers.26.self_attn.out_proj.bias', 'model.vision_tower.vision_tower.vision_model.encoder.layers.26.self_attn.out_proj.weight', 'model.vision_tower.vision_tower.vision_model.encoder.layers.26.self_attn.q_proj.bias', 'model.vision_tower.vision_tower.vision_model.encoder.layers.26.self_attn.q_proj.weight', 'model.vision_tower.vision_tower.vision_model.encoder.layers.26.self_attn.v_proj.bias', 'model.vision_tower.vision_tower.vision_model.encoder.layers.26.self_attn.v_proj.weight', 'model.vision_tower.vision_tower.vision_model.encoder.layers.3.layer_norm1.bias', 'model.vision_tower.vision_tower.vision_model.encoder.layers.3.layer_norm1.weight', 'model.vision_tower.vision_tower.vision_model.encoder.layers.3.layer_norm2.bias', 'model.vision_tower.vision_tower.vision_model.encoder.layers.3.layer_norm2.weight', 'model.vision_tower.vision_tower.vision_model.encoder.layers.3.mlp.fc1.bias', 'model.vision_tower.vision_tower.vision_model.encoder.layers.3.mlp.fc1.weight', 'model.vision_tower.vision_tower.vision_model.encoder.layers.3.mlp.fc2.bias', 'model.vision_tower.vision_tower.vision_model.encoder.layers.3.mlp.fc2.weight', 'model.vision_tower.vision_tower.vision_model.encoder.layers.3.self_attn.k_proj.bias', 'model.vision_tower.vision_tower.vision_model.encoder.layers.3.self_attn.k_proj.weight', 'model.vision_tower.vision_tower.vision_model.encoder.layers.3.self_attn.out_proj.bias', 'model.vision_tower.vision_tower.vision_model.encoder.layers.3.self_attn.out_proj.weight', 'model.vision_tower.vision_tower.vision_model.encoder.layers.3.self_attn.q_proj.bias', 'model.vision_tower.vision_tower.vision_model.encoder.layers.3.self_attn.q_proj.weight', 'model.vision_tower.vision_tower.vision_model.encoder.layers.3.self_attn.v_proj.bias', 'model.vision_tower.vision_tower.vision_model.encoder.layers.3.self_attn.v_proj.weight', 'model.vision_tower.vision_tower.vision_model.encoder.layers.4.layer_norm1.bias', 'model.vision_tower.vision_tower.vision_model.encoder.layers.4.layer_norm1.weight', 'model.vision_tower.vision_tower.vision_model.encoder.layers.4.layer_norm2.bias', 'model.vision_tower.vision_tower.vision_model.encoder.layers.4.layer_norm2.weight', 'model.vision_tower.vision_tower.vision_model.encoder.layers.4.mlp.fc1.bias', 'model.vision_tower.vision_tower.vision_model.encoder.layers.4.mlp.fc1.weight', 'model.vision_tower.vision_tower.vision_model.encoder.layers.4.mlp.fc2.bias', 'model.vision_tower.vision_tower.vision_model.encoder.layers.4.mlp.fc2.weight', 'model.vision_tower.vision_tower.vision_model.encoder.layers.4.self_attn.k_proj.bias', 'model.vision_tower.vision_tower.vision_model.encoder.layers.4.self_attn.k_proj.weight', 'model.vision_tower.vision_tower.vision_model.encoder.layers.4.self_attn.out_proj.bias', 'model.vision_tower.vision_tower.vision_model.encoder.layers.4.self_attn.out_proj.weight', 'model.vision_tower.vision_tower.vision_model.encoder.layers.4.self_attn.q_proj.bias', 'model.vision_tower.vision_tower.vision_model.encoder.layers.4.self_attn.q_proj.weight', 'model.vision_tower.vision_tower.vision_model.encoder.layers.4.self_attn.v_proj.bias', 'model.vision_tower.vision_tower.vision_model.encoder.layers.4.self_attn.v_proj.weight', 'model.vision_tower.vision_tower.vision_model.encoder.layers.5.layer_norm1.bias', 'model.vision_tower.vision_tower.vision_model.encoder.layers.5.layer_norm1.weight', 'model.vision_tower.vision_tower.vision_model.encoder.layers.5.layer_norm2.bias', 'model.vision_tower.vision_tower.vision_model.encoder.layers.5.layer_norm2.weight', 'model.vision_tower.vision_tower.vision_model.encoder.layers.5.mlp.fc1.bias', 'model.vision_tower.vision_tower.vision_model.encoder.layers.5.mlp.fc1.weight', 'model.vision_tower.vision_tower.vision_model.encoder.layers.5.mlp.fc2.bias', 'model.vision_tower.vision_tower.vision_model.encoder.layers.5.mlp.fc2.weight', 'model.vision_tower.vision_tower.vision_model.encoder.layers.5.self_attn.k_proj.bias', 'model.vision_tower.vision_tower.vision_model.encoder.layers.5.self_attn.k_proj.weight', 'model.vision_tower.vision_tower.vision_model.encoder.layers.5.self_attn.out_proj.bias', 'model.vision_tower.vision_tower.vision_model.encoder.layers.5.self_attn.out_proj.weight', 'model.vision_tower.vision_tower.vision_model.encoder.layers.5.self_attn.q_proj.bias', 'model.vision_tower.vision_tower.vision_model.encoder.layers.5.self_attn.q_proj.weight', 'model.vision_tower.vision_tower.vision_model.encoder.layers.5.self_attn.v_proj.bias', 'model.vision_tower.vision_tower.vision_model.encoder.layers.5.self_attn.v_proj.weight', 'model.vision_tower.vision_tower.vision_model.encoder.layers.6.layer_norm1.bias', 'model.vision_tower.vision_tower.vision_model.encoder.layers.6.layer_norm1.weight', 'model.vision_tower.vision_tower.vision_model.encoder.layers.6.layer_norm2.bias', 'model.vision_tower.vision_tower.vision_model.encoder.layers.6.layer_norm2.weight', 'model.vision_tower.vision_tower.vision_model.encoder.layers.6.mlp.fc1.bias', 'model.vision_tower.vision_tower.vision_model.encoder.layers.6.mlp.fc1.weight', 'model.vision_tower.vision_tower.vision_model.encoder.layers.6.mlp.fc2.bias', 'model.vision_tower.vision_tower.vision_model.encoder.layers.6.mlp.fc2.weight', 'model.vision_tower.vision_tower.vision_model.encoder.layers.6.self_attn.k_proj.bias', 'model.vision_tower.vision_tower.vision_model.encoder.layers.6.self_attn.k_proj.weight', 'model.vision_tower.vision_tower.vision_model.encoder.layers.6.self_attn.out_proj.bias', 'model.vision_tower.vision_tower.vision_model.encoder.layers.6.self_attn.out_proj.weight', 'model.vision_tower.vision_tower.vision_model.encoder.layers.6.self_attn.q_proj.bias', 'model.vision_tower.vision_tower.vision_model.encoder.layers.6.self_attn.q_proj.weight', 'model.vision_tower.vision_tower.vision_model.encoder.layers.6.self_attn.v_proj.bias', 'model.vision_tower.vision_tower.vision_model.encoder.layers.6.self_attn.v_proj.weight', 'model.vision_tower.vision_tower.vision_model.encoder.layers.7.layer_norm1.bias', 'model.vision_tower.vision_tower.vision_model.encoder.layers.7.layer_norm1.weight', 'model.vision_tower.vision_tower.vision_model.encoder.layers.7.layer_norm2.bias', 'model.vision_tower.vision_tower.vision_model.encoder.layers.7.layer_norm2.weight', 'model.vision_tower.vision_tower.vision_model.encoder.layers.7.mlp.fc1.bias', 'model.vision_tower.vision_tower.vision_model.encoder.layers.7.mlp.fc1.weight', 'model.vision_tower.vision_tower.vision_model.encoder.layers.7.mlp.fc2.bias', 'model.vision_tower.vision_tower.vision_model.encoder.layers.7.mlp.fc2.weight', 'model.vision_tower.vision_tower.vision_model.encoder.layers.7.self_attn.k_proj.bias', 'model.vision_tower.vision_tower.vision_model.encoder.layers.7.self_attn.k_proj.weight', 'model.vision_tower.vision_tower.vision_model.encoder.layers.7.self_attn.out_proj.bias', 'model.vision_tower.vision_tower.vision_model.encoder.layers.7.self_attn.out_proj.weight', 'model.vision_tower.vision_tower.vision_model.encoder.layers.7.self_attn.q_proj.bias', 'model.vision_tower.vision_tower.vision_model.encoder.layers.7.self_attn.q_proj.weight', 'model.vision_tower.vision_tower.vision_model.encoder.layers.7.self_attn.v_proj.bias', 'model.vision_tower.vision_tower.vision_model.encoder.layers.7.self_attn.v_proj.weight', 'model.vision_tower.vision_tower.vision_model.encoder.layers.8.layer_norm1.bias', 'model.vision_tower.vision_tower.vision_model.encoder.layers.8.layer_norm1.weight', 'model.vision_tower.vision_tower.vision_model.encoder.layers.8.layer_norm2.bias', 'model.vision_tower.vision_tower.vision_model.encoder.layers.8.layer_norm2.weight', 'model.vision_tower.vision_tower.vision_model.encoder.layers.8.mlp.fc1.bias', 'model.vision_tower.vision_tower.vision_model.encoder.layers.8.mlp.fc1.weight', 'model.vision_tower.vision_tower.vision_model.encoder.layers.8.mlp.fc2.bias', 'model.vision_tower.vision_tower.vision_model.encoder.layers.8.mlp.fc2.weight', 'model.vision_tower.vision_tower.vision_model.encoder.layers.8.self_attn.k_proj.bias', 'model.vision_tower.vision_tower.vision_model.encoder.layers.8.self_attn.k_proj.weight', 'model.vision_tower.vision_tower.vision_model.encoder.layers.8.self_attn.out_proj.bias', 'model.vision_tower.vision_tower.vision_model.encoder.layers.8.self_attn.out_proj.weight', 'model.vision_tower.vision_tower.vision_model.encoder.layers.8.self_attn.q_proj.bias', 'model.vision_tower.vision_tower.vision_model.encoder.layers.8.self_attn.q_proj.weight', 'model.vision_tower.vision_tower.vision_model.encoder.layers.8.self_attn.v_proj.bias', 'model.vision_tower.vision_tower.vision_model.encoder.layers.8.self_attn.v_proj.weight', 'model.vision_tower.vision_tower.vision_model.encoder.layers.9.layer_norm1.bias', 'model.vision_tower.vision_tower.vision_model.encoder.layers.9.layer_norm1.weight', 'model.vision_tower.vision_tower.vision_model.encoder.layers.9.layer_norm2.bias', 'model.vision_tower.vision_tower.vision_model.encoder.layers.9.layer_norm2.weight', 'model.vision_tower.vision_tower.vision_model.encoder.layers.9.mlp.fc1.bias', 'model.vision_tower.vision_tower.vision_model.encoder.layers.9.mlp.fc1.weight', 'model.vision_tower.vision_tower.vision_model.encoder.layers.9.mlp.fc2.bias', 'model.vision_tower.vision_tower.vision_model.encoder.layers.9.mlp.fc2.weight', 'model.vision_tower.vision_tower.vision_model.encoder.layers.9.self_attn.k_proj.bias', 'model.vision_tower.vision_tower.vision_model.encoder.layers.9.self_attn.k_proj.weight', 'model.vision_tower.vision_tower.vision_model.encoder.layers.9.self_attn.out_proj.bias', 'model.vision_tower.vision_tower.vision_model.encoder.layers.9.self_attn.out_proj.weight', 'model.vision_tower.vision_tower.vision_model.encoder.layers.9.self_attn.q_proj.bias', 'model.vision_tower.vision_tower.vision_model.encoder.layers.9.self_attn.q_proj.weight', 'model.vision_tower.vision_tower.vision_model.encoder.layers.9.self_attn.v_proj.bias', 'model.vision_tower.vision_tower.vision_model.encoder.layers.9.self_attn.v_proj.weight', 'model.vision_tower.vision_tower.vision_model.head.attention.in_proj_bias', 'model.vision_tower.vision_tower.vision_model.head.attention.in_proj_weight', 'model.vision_tower.vision_tower.vision_model.head.attention.out_proj.bias', 'model.vision_tower.vision_tower.vision_model.head.attention.out_proj.weight', 'model.vision_tower.vision_tower.vision_model.head.layernorm.bias', 'model.vision_tower.vision_tower.vision_model.head.layernorm.weight', 'model.vision_tower.vision_tower.vision_model.head.mlp.fc1.bias', 'model.vision_tower.vision_tower.vision_model.head.mlp.fc1.weight', 'model.vision_tower.vision_tower.vision_model.head.mlp.fc2.bias', 'model.vision_tower.vision_tower.vision_model.head.mlp.fc2.weight', 'model.vision_tower.vision_tower.vision_model.head.probe', 'model.vision_tower.vision_tower.vision_model.post_layernorm.bias', 'model.vision_tower.vision_tower.vision_model.post_layernorm.weight']
- This IS expected if you are initializing LlavaQwen2ForCausalLM from the checkpoint of a model trained on another task or with another architecture (e.g. initializing a BertForSequenceClassification model from a BertForPreTraining model).
- This IS NOT expected if you are initializing LlavaQwen2ForCausalLM from the checkpoint of a model that you expect to be exactly identical (initializing a BertForSequenceClassification model from a BertForSequenceClassification model).
/opt/home/k8sworker/.cache/huggingface/modules/transformers_modules/qnguyen3/nanoLLaVA/13d60cec183a86755afed64da495fcc2c382ea80/modeling_llava_qwen2.py:169: TracerWarning: Converting a tensor to a Python boolean might cause the trace to be incorrect. We can't record the data flow of Python values, so this value will be treated as a constant in the future. This means that the trace might not generalize to other inputs!
  if attn_weights.size() != (batch_size, self.num_heads, q_len, k_v_seq_len):
/opt/home/k8sworker/.cache/huggingface/modules/transformers_modules/qnguyen3/nanoLLaVA/13d60cec183a86755afed64da495fcc2c382ea80/modeling_llava_qwen2.py:187: TracerWarning: Converting a tensor to a Python boolean might cause the trace to be incorrect. We can't record the data flow of Python values, so this value will be treated as a constant in the future. This means that the trace might not generalize to other inputs!
  if attn_output.size() != (batch_size, self.num_heads, q_len, self.head_dim):
/opt/home/k8sworker/ci-ai/cibuilds/jobs/ov-notebook/jobs/OVNotebookOps/builds/810/archive/.workspace/scm/ov-notebook/.venv/lib/python3.8/site-packages/transformers/cache_utils.py:447: TracerWarning: Using len to get tensor shape might cause the trace to be incorrect. Recommended usage would be tensor.shape[0]. Passing a tensor of different shape might lead to errors or silently give incorrect results.
  or len(self.key_cache[layer_idx]) == 0  # the layer has no cache
/opt/home/k8sworker/ci-ai/cibuilds/jobs/ov-notebook/jobs/OVNotebookOps/builds/810/archive/.workspace/scm/ov-notebook/.venv/lib/python3.8/site-packages/transformers/modeling_attn_mask_utils.py:116: TracerWarning: Converting a tensor to a Python boolean might cause the trace to be incorrect. We can't record the data flow of Python values, so this value will be treated as a constant in the future. This means that the trace might not generalize to other inputs!
  if (input_shape[-1] > 1 or self.sliding_window is not None) and self.is_causal:
/opt/home/k8sworker/ci-ai/cibuilds/jobs/ov-notebook/jobs/OVNotebookOps/builds/810/archive/.workspace/scm/ov-notebook/.venv/lib/python3.8/site-packages/optimum/exporters/onnx/model_patcher.py:306: TracerWarning: Converting a tensor to a Python boolean might cause the trace to be incorrect. We can't record the data flow of Python values, so this value will be treated as a constant in the future. This means that the trace might not generalize to other inputs!
  if past_key_values_length > 0:
/opt/home/k8sworker/.cache/huggingface/modules/transformers_modules/qnguyen3/nanoLLaVA/13d60cec183a86755afed64da495fcc2c382ea80/modeling_llava_qwen2.py:939: TracerWarning: Converting a tensor to a Python boolean might cause the trace to be incorrect. We can't record the data flow of Python values, so this value will be treated as a constant in the future. This means that the trace might not generalize to other inputs!
  if seq_len > self.max_seq_len_cached:
/opt/home/k8sworker/ci-ai/cibuilds/jobs/ov-notebook/jobs/OVNotebookOps/builds/810/archive/.workspace/scm/ov-notebook/.venv/lib/python3.8/site-packages/transformers/cache_utils.py:432: TracerWarning: Using len to get tensor shape might cause the trace to be incorrect. Recommended usage would be tensor.shape[0]. Passing a tensor of different shape might lead to errors or silently give incorrect results.
  elif len(self.key_cache[layer_idx]) == 0:  # fills previously skipped layers; checking for tensor causes errors
/opt/home/k8sworker/.cache/huggingface/modules/transformers_modules/qnguyen3/nanoLLaVA/13d60cec183a86755afed64da495fcc2c382ea80/modeling_llava_qwen2.py:1499: TracerWarning: Converting a tensor to a Python boolean might cause the trace to be incorrect. We can't record the data flow of Python values, so this value will be treated as a constant in the future. This means that the trace might not generalize to other inputs!
  if attention_mask.size() != (bsz, 1, q_len, kv_seq_len):
OpenVINO and OpenVINO Tokenizers versions are not binary compatible.
OpenVINO version:            2024.5.0-16993
OpenVINO Tokenizers version: 2024.5.0.0
First 3 numbers should be the same. Update OpenVINO Tokenizers to compatible version. It is recommended to use the same day builds for pre-release version. To install both OpenVINO and OpenVINO Tokenizers release version perform:
pip install --force-reinstall openvino openvino-tokenizers
To update both OpenVINO and OpenVINO Tokenizers to the latest pre-release version perform:
pip install --pre -U openvino openvino-tokenizers --extra-index-url https://storage.openvinotoolkit.org/simple/wheels/nightly
Tokenizer won't be converted.
Traceback (most recent call last):
  File "/opt/home/k8sworker/ci-ai/cibuilds/jobs/ov-notebook/jobs/OVNotebookOps/builds/810/archive/.workspace/scm/ov-notebook/.venv/bin/optimum-cli", line 10, in <module>
    sys.exit(main())
  File "/opt/home/k8sworker/ci-ai/cibuilds/jobs/ov-notebook/jobs/OVNotebookOps/builds/810/archive/.workspace/scm/ov-notebook/.venv/lib/python3.8/site-packages/optimum/commands/optimum_cli.py", line 208, in main
    service.run()
  File "/opt/home/k8sworker/ci-ai/cibuilds/jobs/ov-notebook/jobs/OVNotebookOps/builds/810/archive/.workspace/scm/ov-notebook/.venv/lib/python3.8/site-packages/optimum/commands/export/openvino.py", line 349, in run
    main_export(
  File "/opt/home/k8sworker/ci-ai/cibuilds/jobs/ov-notebook/jobs/OVNotebookOps/builds/810/archive/.workspace/scm/ov-notebook/.venv/lib/python3.8/site-packages/optimum/exporters/openvino/__main__.py", line 416, in main_export
    core = Core()
  File "/opt/home/k8sworker/ci-ai/cibuilds/jobs/ov-notebook/jobs/OVNotebookOps/builds/810/archive/.workspace/scm/ov-notebook/.venv/lib/python3.8/site-packages/openvino_tokenizers/__init__.py", line 53, in new_core_init
    self.add_extension(str(_ext_path))  # Core.add_extension doesn't support Path object
RuntimeError: Exception from src/inference/src/cpp/core.cpp:158:
Cannot add extension. Cannot find entry point to the extension library. This error happened: Cannot load library '/opt/home/k8sworker/ci-ai/cibuilds/jobs/ov-notebook/jobs/OVNotebookOps/builds/810/archive/.workspace/scm/ov-notebook/.venv/lib/python3.8/site-packages/openvino_tokenizers/lib/libopenvino_tokenizers.so': /opt/home/k8sworker/ci-ai/cibuilds/jobs/ov-notebook/jobs/OVNotebookOps/builds/810/archive/.workspace/scm/ov-notebook/.venv/lib/python3.8/site-packages/openvino_tokenizers/lib/libopenvino_tokenizers.so: undefined symbol: _ZNK2ov4Node17can_constant_foldERKSt6vectorINS_6OutputIS0_EESaIS3_EE

Compress Model weights to 4 and 8 bits using NNCF#

For reducing memory consumption, weights compression optimization can be applied using NNCF. Weight compression aims to reduce the memory footprint of a model. It can also lead to significant performance improvement for large memory-bound models, such as Large Language Models (LLMs). LLMs and other models, which require extensive memory to store the weights during inference, can benefit from weight compression in the following ways:

enabling the inference of exceptionally large models that cannot be accommodated in the memory of the device;
improving the inference performance of the models by reducing the latency of the memory access when computing the operations with weights, for example, Linear layers.

Neural Network Compression Framework (NNCF) provides 4-bit / 8-bit mixed weight quantization as a compression method primarily designed to optimize LLMs. The main difference between weights compression and full model quantization (post-training quantization) is that activations remain floating-point in the case of weights compression which leads to a better accuracy. Weight compression for LLMs provides a solid inference performance improvement which is on par with the performance of the full model quantization. In addition, weight compression is data-free and does not require a calibration dataset, making it easy to use.

nncf.compress_weights function can be used for performing weights compression. The function accepts an OpenVINO model and other compression parameters. Compared to INT8 compression, INT4 compression improves performance even more, but introduces a minor drop in prediction quality.

More details about weights compression, can be found in OpenVINO documentation.

Please select below whether you would like to run INT4 weight compression instead of INT8 weight compression.

import ipywidgets as widgets

compression_mode = widgets.Dropdown(
    options=["INT4", "INT8"],
    value="INT4",
    description="Compression mode:",
    disabled=False,
)

compression_mode

Dropdown(description='Compression mode:', options=('INT4', 'INT8'), value='INT4')

import nncf
import openvino as ov

core = ov.Core()

if compression_mode.value == "INT4":
    ov_compressed_model_dir = ov_model_dir.parent / "INT4"
    llava_wc_parameters = dict(mode=nncf.CompressWeightsMode.INT4_ASYM, group_size=128, ratio=0.8)
else:
    ov_compressed_model_dir = ov_model_dir.parent / "INT8"
    llava_wc_parameters = dict(mode=nncf.CompressWeightsMode.INT8)

image_encoder_wc_parameters = dict(mode=nncf.CompressWeightsMode.INT8)

INFO:nncf:NNCF initialized successfully. Supported frameworks detected: torch, tensorflow, onnx, openvino

Image Encoder#

Image Encoder is represented in nanoLLaVA by pretrained SigLIP model. Image encoder is responsible for encoding input images into embedding space. Code bellow demonstrates how to apply weights compression for image encoder model.

import gc

compressed_vision_encoder_path = ov_compressed_model_dir / "openvino_vision_embeddings_model.xml"
vision_encoder_path = ov_model_dir / "openvino_vision_embeddings_model.xml"
if not compressed_vision_encoder_path.exists():
    ov_vision_encoder = core.read_model(vision_encoder_path)
    ov_compressed_vision_encoder = nncf.compress_weights(ov_vision_encoder, **image_encoder_wc_parameters)
    ov.save_model(ov_compressed_vision_encoder, compressed_vision_encoder_path)
    del ov_compressed_vision_encoder
    del ov_vision_encoder
    gc.collect();

/opt/home/k8sworker/ci-ai/cibuilds/jobs/ov-notebook/jobs/OVNotebookOps/builds/810/archive/.workspace/scm/ov-notebook/.venv/lib/python3.8/site-packages/nncf/quantization/quantize_model.py:432: FutureWarning: CompressWeightsMode.INT8 is deprecated. Please, use CompressWeightsMode.INT8_ASYM as value instead.
  warning_deprecated(
2024-11-05 02:09:38.791476: I tensorflow/core/util/port.cc:110] oneDNN custom operations are on. You may see slightly different numerical results due to floating-point round-off errors from different computation orders. To turn them off, set the environment variable TF_ENABLE_ONEDNN_OPTS=0.
2024-11-05 02:09:38.825207: I tensorflow/core/platform/cpu_feature_guard.cc:182] This TensorFlow binary is optimized to use available CPU instructions in performance-critical operations.
To enable the following instructions: AVX2 AVX512F AVX512_VNNI FMA, in other operations, rebuild TensorFlow with the appropriate compiler flags.
2024-11-05 02:09:39.427301: W tensorflow/compiler/tf2tensorrt/utils/py_utils.cc:38] TF-TRT Warning: Could not find TensorRT

INFO:nncf:Statistics of the bitwidth distribution:
┍━━━━━━━━━━━━━━━━┯━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┯━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┑
│   Num bits (N) │ % all parameters (layers)   │ % ratio-defining parameters (layers)   │
┝━━━━━━━━━━━━━━━━┿━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┿━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┥
│              8 │ 100% (159 / 159)            │ 100% (159 / 159)                       │
┕━━━━━━━━━━━━━━━━┷━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┷━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┙

Output()

Language Model#

Language Model is responsible for generation answer in LLaVA. This part is very similar to standard LLM for text generation. Our model uses Qwen/Qwen1.5-0.5B as base LLM.

compressed_llm_path = ov_compressed_model_dir / "openvino_language_model.xml"
llm_path = ov_model_dir / "openvino_language_model.xml"

if not compressed_llm_path.exists():
    ov_llm = core.read_model(llm_path)
    ov_compressed_llm = nncf.compress_weights(ov_llm, **llava_wc_parameters)
    ov.save_model(ov_compressed_llm, compressed_llm_path)
    del ov_compressed_llm
    del ov_llm
    gc.collect()

copy_model_files(ov_model_dir, ov_compressed_model_dir)

Output()

INFO:nncf:Statistics of the bitwidth distribution:
┍━━━━━━━━━━━━━━━━┯━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┯━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┑
│   Num bits (N) │ % all parameters (layers)   │ % ratio-defining parameters (layers)   │
┝━━━━━━━━━━━━━━━━┿━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┿━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┥
│              8 │ 47% (48 / 169)              │ 20% (47 / 168)                         │
├────────────────┼─────────────────────────────┼────────────────────────────────────────┤
│              4 │ 53% (121 / 169)             │ 80% (121 / 168)                        │
┕━━━━━━━━━━━━━━━━┷━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┷━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┙

Output()

Prepare model inference pipeline#

OpenVINO integration with Optimum Intel provides ready-to-use API for model inference that can be used for smooth integration with transformers-based solutions. For loading pixtral model, we will use OVModelForVisualCausalLM class that have compatible interface with Transformers Pixtral implementation. For loading a model, from_pretrained method should be used. It accepts path to the model directory or model_id from HuggingFace hub (if model is not converted to OpenVINO format, conversion will be triggered automatically). Additionally, we can provide an inference device, quantization config (if model has not been quantized yet) and device-specific OpenVINO Runtime configuration. More details about model inference with Optimum Intel can be found in documentation.

Run OpenVINO Model Inference#

Select device#

import requests

r = requests.get(
    url="https://raw.githubusercontent.com/openvinotoolkit/openvino_notebooks/latest/utils/notebook_utils.py",
)
open("notebook_utils.py", "w").write(r.text)

from notebook_utils import device_widget

device = device_widget("CPU", exclude=["NPU"])

device

Dropdown(description='Device:', options=('CPU', 'AUTO'), value='CPU')

Optimum Intel provides Transformers-like interface for inference OpenVINO models that allows smooth integration into user application, where you need just replace model class, other parts of pipeline - preprocessing and postprocessing code remains the same. It means that we can use the same tokenizer and image processor that provided with model.

from optimum.intel.openvino import OVModelForVisualCausalLM
from transformers import AutoConfig, AutoTokenizer, AutoProcessor, TextStreamer

# prepare tokenizer
tokenizer = AutoTokenizer.from_pretrained(ov_compressed_model_dir, trust_remote_code=True)

# prepare image processor
config = AutoConfig.from_pretrained(ov_compressed_model_dir, trust_remote_code=True)
processor = AutoProcessor.from_pretrained(config.mm_vision_tower)

# initialize OpenVINO model inference class
ov_model = OVModelForVisualCausalLM.from_pretrained(ov_compressed_model_dir, device=device.value, trust_remote_code=True)

from ov_nano_llava_helper import process_images, process_text_input
from PIL import Image

prompt = "Describe this image in detail"

messages = [{"role": "user", "content": f"<image>\n{prompt}"}]
text = tokenizer.apply_chat_template(messages, tokenize=False, add_generation_prompt=True)
url = "https://github.com/openvinotoolkit/openvino_notebooks/assets/29454499/8bf7d9f2-018a-4498-bec4-55f17c273ecc"
image = Image.open(requests.get(url, stream=True).raw)
image_tensor = process_images(image, None, processor)
input_ids, attention_mask = process_text_input(text, tokenizer)

streamer = TextStreamer(tokenizer, skip_prompt=True, skip_special_tokens=True)

display(image)
print(f"Question:\n{prompt}")
print("Answer:")

output_ids = ov_model.generate(input_ids, attention_mask=attention_mask, images=image_tensor, max_new_tokens=128, use_cache=True, streamer=streamer)

../_images/nano-llava-multimodal-chatbot-with-output_22_0.png

Setting pad_token_id to eos_token_id:None for open-end generation.

Question:
Describe this image in detail
Answer:
The image features a small, adorable white lamb standing amidst a fire. The lamb's fur is fluffy and white, and it is adorned with tiny black eyes that are bright and lively. The lamb's face is cute, with a small black nose and a small mouth. It seems like the lamb is looking straight at the camera, making it appear even more adorable.
The lamb's right ear is visible, and it is white and pink. The lamb's right eye is also black and pink. The lamb's face is quite detailed, with the nose and mouth visible. There are also details like the lamb's right foot, which is white

Interactive demo#

from transformers import TextIteratorStreamer, StoppingCriteria
from threading import Thread
import torch


class KeywordsStoppingCriteria(StoppingCriteria):
    def __init__(self, keywords, tokenizer, input_ids):
        self.keywords = keywords
        self.keyword_ids = []
        self.max_keyword_len = 0
        for keyword in keywords:
            cur_keyword_ids = tokenizer(keyword).input_ids
            if len(cur_keyword_ids) > 1 and cur_keyword_ids[0] == tokenizer.bos_token_id:
                cur_keyword_ids = cur_keyword_ids[1:]
            if len(cur_keyword_ids) > self.max_keyword_len:
                self.max_keyword_len = len(cur_keyword_ids)
            self.keyword_ids.append(torch.tensor(cur_keyword_ids))
        self.tokenizer = tokenizer
        self.start_len = input_ids.shape[1]

    def call_for_batch(self, output_ids: torch.LongTensor, scores: torch.FloatTensor, **kwargs) -> bool:
        offset = min(output_ids.shape[1] - self.start_len, self.max_keyword_len)
        self.keyword_ids = [keyword_id.to(output_ids.device) for keyword_id in self.keyword_ids]
        for keyword_id in self.keyword_ids:
            truncated_output_ids = output_ids[0, -keyword_id.shape[0] :]
            if torch.equal(truncated_output_ids, keyword_id):
                return True
        outputs = self.tokenizer.batch_decode(output_ids[:, -offset:], skip_special_tokens=True)[0]
        for keyword in self.keywords:
            if keyword in outputs:
                return True
        return False

    def __call__(self, output_ids: torch.LongTensor, scores: torch.FloatTensor, **kwargs) -> bool:
        outputs = []
        for i in range(output_ids.shape[0]):
            outputs.append(self.call_for_batch(output_ids[i].unsqueeze(0), scores))
        return all(outputs)


def bot_streaming(message, history):
    messages = []
    if message["files"]:
        image = message["files"][-1]["path"] if isinstance(message["files"][-1], dict) else message["files"][-1]
    else:
        for _, hist in enumerate(history):
            if isinstance(hist[0], tuple):
                image = hist[0][0]

    if len(history) > 0 and image is not None:
        messages.append({"role": "user", "content": f"<image>\n{history[1][0]}"})
        messages.append({"role": "assistant", "content": history[1][1]})
        for human, assistant in history[2:]:
            if assistant is None:
                continue
            messages.append({"role": "user", "content": human})
            messages.append({"role": "assistant", "content": assistant})
        messages.append({"role": "user", "content": message["text"]})
    elif len(history) > 0 and image is None:
        for human, assistant in history:
            if assistant is None:
                continue
            messages.append({"role": "user", "content": human})
            messages.append({"role": "assistant", "content": assistant})
        messages.append({"role": "user", "content": message["text"]})
    elif len(history) == 0 and image is not None:
        messages.append({"role": "user", "content": f"<image>\n{message['text']}"})
    elif len(history) == 0 and image is None:
        messages.append({"role": "user", "content": message["text"]})

    print(messages)
    image = Image.open(image).convert("RGB")
    text = tokenizer.apply_chat_template(messages, tokenize=False, add_generation_prompt=True)
    image_tensor = process_images(image, None, processor)
    input_ids, attention_mask = process_text_input(text, tokenizer)
    stop_str = "<|im_end|>"
    keywords = [stop_str]
    stopping_criteria = KeywordsStoppingCriteria(keywords, tokenizer, input_ids)
    streamer = TextIteratorStreamer(tokenizer, skip_prompt=True, skip_special_tokens=True)
    generation_kwargs = dict(
        input_ids=input_ids,
        attention_mask=attention_mask,
        images=image_tensor,
        streamer=streamer,
        max_new_tokens=128,
        stopping_criteria=[stopping_criteria],
        temperature=0.01,
    )
    thread = Thread(target=ov_model.generate, kwargs=generation_kwargs)
    thread.start()

    buffer = ""
    for new_text in streamer:
        buffer += new_text
        generated_text_without_prompt = buffer[:]
        yield generated_text_without_prompt

if not Path("gradio_helper.py").exists():
    r = requests.get(url="https://raw.githubusercontent.com/openvinotoolkit/openvino_notebooks/latest/notebooks/nano-llava-multimodal-chatbot/gradio_helper.py")
    open("gradio_helper.py", "w").write(r.text)

from gradio_helper import make_demo

demo = make_demo(fn=bot_streaming)

try:
    demo.launch(debug=False)
except Exception:
    demo.launch(share=True, debug=False)
# if you are launching remotely, specify server_name and server_port
# demo.launch(server_name='your server name', server_port='server port in int')
# Read more in the docs: https://gradio.app/docs/

Running on local URL:  http://127.0.0.1:7860

To create a public link, set share=True in launch().