Visual-language assistant with MiniCPM-V2 and OpenVINO

This Jupyter notebook can be launched after a local installation only.

MiniCPM-V 2 is a strong multimodal large language model for efficient end-side deployment. MiniCPM-V 2.6 is the latest and most capable model in the MiniCPM-V series. The model is built on SigLip-400M and Qwen2-7B with a total of 8B parameters. It exhibits a significant performance improvement over previous versions, and introduces new features for multi-image and video understanding.

More details about model can be found in model card and original repo.

In this tutorial we consider how to convert and optimize MiniCPM-V2 model for creating multimodal chatbot. Additionally, we demonstrate how to apply stateful transformation on LLM part and model optimization techniques like weights compression using NNCF

Table of contents:

• Prerequisites

• Convert model to OpenVINO Intermediate Representation

  • Compress Language Model Weights to 4 bits

• Prepare model inference pipeline

• Run OpenVINO model inference

  • Select device

  • Select language model variant

• 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" "torchvision" "timm>=0.9.2" "transformers>=4.40" "Pillow" "gradio>=4.19" "tqdm" "sentencepiece" "peft" "huggingface-hub>=0.24.0" --extra-index-url https://download.pytorch.org/whl/cpu
%pip install -q "openvino>=2024.3.0" "nncf>=2.12.0"

WARNING: Error parsing dependencies of torchsde: .* suffix can only be used with == or != operators
    numpy (>=1.19.*) ; python_version >= "3.7"
           ~~~~~~~^
Note: you may need to restart the kernel to use updated packages.
WARNING: Error parsing dependencies of torchsde: .* suffix can only be used with == or != operators
    numpy (>=1.19.*) ; python_version >= "3.7"
           ~~~~~~~^
Note: you may need to restart the kernel to use updated packages.

import requests
from pathlib import Path

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


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

if not Path("notebook_utils.py").exists():
    r = requests.get(url="https://raw.githubusercontent.com/openvinotoolkit/openvino_notebooks/latest/utils/notebook_utils.py")
    open("notebook_utils.py", "w").write(r.text)

Convert model to OpenVINO Intermediate Representation

OpenVINO supports PyTorch models via conversion to OpenVINO Intermediate Representation (IR). OpenVINO model conversion API should be used for these purposes. ov.convert_model function accepts original PyTorch model instance and example input for tracing and returns ov.Model representing this model in OpenVINO framework. Converted model can be used for saving on disk using ov.save_model function or directly loading on device using core.complie_model.

minicpm_helper.py script contains helper function for model conversion, please check its content if you interested in conversion details.

Click here for more detailed explanation of conversion steps

MiniCPM-V2.6 is autoregressive transformer generative model, it means that each next model step depends from model output from previous step. The generation approach is based on the assumption that the probability distribution of a word sequence can be decomposed into the product of conditional next word distributions. In other words, model predicts the next token in the loop guided by previously generated tokens until the stop-condition will be not reached (generated sequence of maximum length or end of string token obtained). The way the next token will be selected over predicted probabilities is driven by the selected decoding methodology. You can find more information about the most popular decoding methods in this blog. The entry point for the generation process for models from the Hugging Face Transformers library is the generate method. You can find more information about its parameters and configuration in the documentation. To preserve flexibility in the selection decoding methodology, we will convert only model inference for one step.

The inference flow has difference on first step and for the next. On the first step, model accept preprocessed input instruction and image, that transformed to the unified embedding space using input_embedding and image encoder models, after that language model, LLM-based part of model, runs on input embeddings to predict probability of next generated tokens. On the next step, language_model accepts only next token id selected based on sampling strategy and processed by input_embedding model and cached attention key and values. Since the output side is auto-regressive, an output token hidden state remains the same once computed for every further generation step. Therefore, recomputing it every time you want to generate a new token seems wasteful. With the cache, the model saves the hidden state once it has been computed. The model only computes the one for the most recently generated output token at each time step, re-using the saved ones for hidden tokens. This reduces the generation complexity from \(O(n^3)\) to \(O(n^2)\) for a transformer model. More details about how it works can be found in this article.

With increasing model size like in modern LLMs, we also can note an increase in the number of attention blocks and size past key values tensors respectively. The strategy for handling cache state as model inputs and outputs in the inference cycle may become a bottleneck for memory-bounded systems, especially with processing long input sequences, for example in a chatbot scenario. OpenVINO suggests a transformation that removes inputs and corresponding outputs with cache tensors from the model keeping cache handling logic inside the model. Such models are also called stateful. A stateful model is a model that implicitly preserves data between two consecutive inference calls. The tensors saved from one run are kept in an internal memory buffer called a state or a variable and may be passed to the next run, while never being exposed as model output. Hiding the cache enables storing and updating the cache values in a more device-friendly representation. It helps to reduce memory consumption and additionally optimize model performance. More details about stateful models and working with state can be found in OpenVINO documentation.

In LLMs, input_embedding is a part of language model, but for multimodal case, the first step hidden state produced by this model part should be integrated with image embeddings into common embedding space. For ability to reuse this model part and avoid introduction of llm model instance, we will use it separately.

image_encoder is represented in MiniCPM-V by pretrained SigLIP model. Additionally, MiniCPM uses perceiver resampler that compresses the image representations. To preserve model ability to process images of different size with respect aspect ratio combined in batch, we will use image_encoder and resampler as separated models.

To sum up above, model consists of 4 parts:

• Image Encoder for encoding input images into embedding space. It includes SigLIP model.

• Resampler for compression image representation.

• Input Embedding for conversion input text tokens into embedding space.

• Language Model for generation answer based on input embeddings provided by Image Encoder and Input Embedding models.

Let’s convert each model part.

from minicpm_helper import convert_minicpmv26

# uncomment the line to see model conversion code
# ??convert_minicpmv26

2024-10-07 09:57:53.402018: 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.
2024-10-07 09:57:53.403877: I tensorflow/tsl/cuda/cudart_stub.cc:28] Could not find cuda drivers on your machine, GPU will not be used.
2024-10-07 09:57:53.440490: 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.
2024-10-07 09:57:54.270302: W tensorflow/compiler/tf2tensorrt/utils/py_utils.cc:38] TF-TRT Warning: Could not find TensorRT

model_id = "openbmb/MiniCPM-V-2_6"

model_dir = convert_minicpmv26(model_id)

⌛ openbmb/MiniCPM-V-2_6 conversion started. Be patient, it may takes some time.
⌛ Load Original model

Loading checkpoint shards:   0%|          | 0/4 [00:00<?, ?it/s]

✅ Original model successfully loaded

/home/ea/work/my_optimum_intel/optimum_env/lib/python3.8/site-packages/transformers/models/auto/image_processing_auto.py:513: FutureWarning: The image_processor_class argument is deprecated and will be removed in v4.42. Please use slow_image_processor_class, or fast_image_processor_class instead
  warnings.warn(

⌛ Convert Image embedding model
WARNING:tensorflow:Please fix your imports. Module tensorflow.python.training.tracking.base has been moved to tensorflow.python.trackable.base. The old module will be deleted in version 2.11.

[ WARNING ]  Please fix your imports. Module %s has been moved to %s. The old module will be deleted in version %s.
/home/ea/work/my_optimum_intel/optimum_env/lib/python3.8/site-packages/transformers/modeling_utils.py:4713: FutureWarning: _is_quantized_training_enabled is going to be deprecated in transformers 4.39.0. Please use model.hf_quantizer.is_trainable instead
  warnings.warn(

✅ Image embedding model successfully converted
✅ openbmb/MiniCPM-V-2_6 model sucessfully converted. You can find results in MiniCPM-V-2_6

Compress Language Model Weights to 4 bits

For reducing memory consumption, weights compression optimization can be applied using NNCF.

Click here for more details about weight compression

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.

Note: weights compression process may require additional time and memory for performing. You can disable it using widget below:

from minicpm_helper import compression_widget

to_compress_weights = compression_widget()

to_compress_weights

Checkbox(value=True, description='Weights Compression')

import nncf
import gc
import openvino as ov

from minicpm_helper import llm_path, copy_llm_files


compression_configuration = {"mode": nncf.CompressWeightsMode.INT4_SYM, "group_size": 64, "ratio": 1.0, "all_layers": True}


core = ov.Core()
llm_int4_path = Path("language_model_int4") / llm_path.name
if to_compress_weights.value and not (model_dir / llm_int4_path).exists():
    ov_model = core.read_model(model_dir / llm_path)
    ov_compressed_model = nncf.compress_weights(ov_model, **compression_configuration)
    ov.save_model(ov_compressed_model, model_dir / llm_int4_path)
    del ov_compressed_model
    del ov_model
    gc.collect()
    copy_llm_files(model_dir, llm_int4_path.parent)

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

Prepare model inference pipeline

As discussed, the model comprises Image Encoder and LLM (with separated text embedding part) that generates answer. In minicpm_helper.py we defined LLM inference class OvModelForCausalLMWithEmb that will represent generation cycle, It is based on HuggingFace Transformers GenerationMixin and looks similar to Optimum Intel OVModelForCausalLMthat is used for LLM inference with only difference that it can accept input embedding. In own turn, general multimodal model class OvMiniCPMVModel handles chatbot functionality including image processing and answer generation using LLM.

from minicpm_helper import OvModelForCausalLMWithEmb, OvMiniCPMV, init_model  # noqa: F401

# uncomment the line to see model inference class
# ??OVMiniCPMV

# uncomment the line to see language model inference class
# ??OvModelForCausalLMWithEmb

Run OpenVINO model inference

Select device

from notebook_utils import device_widget

device = device_widget(default="AUTO", exclude=["NPU"])

device

Dropdown(description='Device:', index=1, options=('CPU', 'AUTO'), value='AUTO')

Select language model variant

from minicpm_helper import lm_variant_selector


use_int4_lang_model = lm_variant_selector(model_dir / llm_int4_path)

use_int4_lang_model

Checkbox(value=True, description='INT4 language model')

ov_model = init_model(model_dir, llm_path.parent if not use_int4_lang_model.value else llm_int4_path.parent, device.value)

applied slice for lm head

import requests
from PIL import Image

url = "https://github.com/openvinotoolkit/openvino_notebooks/assets/29454499/d5fbbd1a-d484-415c-88cb-9986625b7b11"
image = Image.open(requests.get(url, stream=True).raw)
question = "What is unusual on this image?"

print(f"Question:\n{question}")
image

Question:
What is unusual on this image?

tokenizer = ov_model.processor.tokenizer

msgs = [{"role": "user", "content": question}]


print("Answer:")
res = ov_model.chat(image=image, msgs=msgs, context=None, tokenizer=tokenizer, sampling=False, stream=True, max_new_tokens=50)

generated_text = ""
for new_text in res:
    generated_text += new_text
    print(new_text, flush=True, end="")

Answer:
The unusual aspect of this image is the cat's relaxed and vulnerable position. Typically, cats avoid exposing their bellies to potential threats or dangers because it leaves them open for attack by predators in nature; however here we see a domesticated pet comfortably lying

Interactive demo

from gradio_helper import make_demo

demo = make_demo(ov_model)

try:
    demo.launch(debug=True, height=600)
except Exception:
    demo.launch(debug=True, share=True, height=600)
# 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/