Howdy! I am an AI Engineering Manager at Intel. I enjoy building cutting-edge generative AI and Large Language Model (LLM) solutions across multiple industries. I have experience in AI code development, cybersecurity, and the energy industry. My aim is to empower AI developers and professionals with the software, code, and tools they need to succeed. You can find samples of my work here through the links I provide to videos, code, and articles.
- YouTube and Recorded Content - Here are some examples of my recorded technical talks, covering topics like distributed fine-tuning of LLMs in the cloud, using neural networks and PyTorch for computer vision tasks, and Kubeflow pipelines on Microsoft Azure.
- Medium Articles - Here are some of my published articles on Medium, covering topics like fine-tuning LLMs, automatic speech recognition (ASR), stable diffusion, quantization, computer vision, and PyTorch.
- Publications - A list of my formal research publications, primarily composed of my work in deep learning and subsurface imaging.
- Hugging Face Contributions - Some of the model cards and other materials I have contributed to Hugging Face.
- Kaggle Contributions - Here are a few of my Python notebooks I have published on Kaggle, including one detailing the hardware available on the platform.
- GitHub Activity - A sample of my direct contributions to the GitHub open-source community.
- Podcasts - Podcasts where I was a guest speaker.
- Work Experience - My journey in work has mostly been focused around AI and geophysics.
- Education - Starting from a strong foundation of mathematics, I moved into teaching and then a thesis-based geophysics degree.
- Contact - How to reach me.
Here are some examples of my recorded technical talks, covering topics like distributed fine-tuning of LLMs in the cloud, using neural networks and PyTorch for computer vision tasks, Kubernetes, and Kubeflow.
| Video Link | Description |
|---|---|
How to Set Up Cloud-Based Distributed Training to Fine-Tune an LLM  |
Learn how to fine-tune nanoGPT (124M parameter) model on a cluster of CPUs on Google Cloud Platform. The model is trained on the OpenWebText dataset in a distributed setting, using 4th Gen. Intel® Xeon® Scalable CPUs. The project builds upon the initial codebase of nanoGPT, by Andrej Karpathy. The objective is to understand how to set up distributed training so that you can fine-tune to your specific objective. The end result of training here will result in a base LLM that can generate words, or tokens, but it will only be suitable for your use-case when you fine-tune it on your specific task and dataset. |
Seismic Data to Subsurface Models with OpenFWI: Training an AI Model with PyTorch  |
Obtaining an accurate “picture” of the subsurface is not as simple as snapping a picture on a smartphone. Seismic exploration is a key component in creating images of the subsurface and finding essential minerals in the subsurface. Building images of the subsurface is akin to ultrasound technology used to image the human body. Learn how to train a neural network with PyTorch on a CPU, going directly from seismic data to a subsurface model. |
Find the Humor in Text Data: NLP with Intel & Habana*  |
Learn how to train a binary classification natural language processing (NLP) model on a humor dataset, where each statement is labeled as humorous or not humorous. The training is performed on a powerful Intel Gaudi GPU. Also learn how to quantize a model to speed up inference by 1.8x, taking it from FP32 format to INT8 format without significant accuracy loss. |
CPU accelerated fine-tuning for image segmentation using PyTorch  |
Fine-tuning neural networks has historically been quite slow and cumbersome on CPUs. However, with mixed precision BF16 training, the Intel Extension for PyTorch has made fine-tuning training feasible on a CPU and perhaps even preferred where cost and availability are key factors. In this tutorial, I will walk you through a real-world example of training an AI image segmentation model using PyTorch 1.13.1 (with ResNet34 architecture); the model will learn to identify roads and speed limits only from satellite images. |
Underground Salt Domes! Fun With Deep Learning  |
Interpreting salt bodies in the subsurface is a challenging manual task that can take weeks to complete. Obtaining accurate picks of salt is very important, because errors in the placement of salt can result in severe degradation of the seismic image. The U-Net architecture proved robust with the placement of salt at 98% accuracy. Beyond accuracy, U-Net also proved to be the fastest, requiring only 3.5 hours to predict salt on the 3D seismic volume. The results presented here along with other recent studies of deep learning for salt interpretation represent a clear shift in the seismic interpretation workflow. |
Physics and Deep Learning for First Breaks  |
Microseismic monitoring is a crucial element to understanding hydraulic fracturing operations prior to oil and gas production. One of the more tedious quality control (QC) measures that must often be performed following a microseismic processing workflow is a visual inspection of seismic data to determine whether the data contain microseismic events or only noise. We propose using a supervised deep learning algorithm, a convolutional neural network (CNN), to automatically classify microseismic events from noise. Using our deep learning approach, we show that the time for QC can be reduced from weeks to hours with high accuracy. |
XGBoost* Kubeflow* Pipeline Intel® Cloud Optimization Modules for Microsoft Intel Software  |
Learn how to build secure, scalable, and accelerated XGBoost pipelines on an Azure Kubernetes service cluster, leveraging Intel SGX. This tutorial walks you through the process from setting up the container to building the full Kubeflow Pipeline using an example application. Access the full source code on GitHub. |
Here are some of my published articles on Medium, covering topics like fine-tuning LLMs, automatic speech recognition (ASR), stable diffusion, quantization, computer vision, and PyTorch.
A list of my formal research publications, primarily composed of my work in deep learning and subsurface imaging.
| Citation | |
|---|---|
| - | Jin, P., Y. Feng, S. Feng, H. Wang, Y. Chen, B. Consolvo, Z. Liu, Y. Lin, 2024, Does Full Waveform Inversion Benefit from Big Data? Manuscript submitted for publication. |
| - | Consolvo, B. P., 2024, 3D land full-waveform inversion in the Permian Basin: A case study at Quail Ridge East: Fourth International Meeting for Applied Geoscience & Energy, Society of Exploration Geophysicists and American Association of Petroleum Geologists. |
| 🌄[PDF] | Consolvo, B. P., B. DeMoss, M. Duiker, 2021, Combining physics and deep learning to automatically pick first breaks in the Permian Basin: First International Meeting for Applied Geoscience & Energy, Society of Exploration Geophysicists. doi: https://doi.org/10.1190/segam2021-3579730.1. |
| 🧂[PDF] | Zabihi Naeini, E., B. P. Consolvo, P. Docherty, and J. Uwaifo, 2020, Deep learning for salt body detection: A practical approach: 83rd Annual International Conference and Exhibition, EAGE, Extended Abstracts. doi: https://doi.org/10.3997/2214-4609.202010270. |
| 🧂[PDF] | Consolvo, B. P., E. Zabihi Naeini, and P. Docherty, 2020, Deep learning for salt body detection applied to 3D Gulf of Mexico data: 90th Annual International Meeting, SEG, Expanded Abstracts. doi: https://doi.org/10.1190/segam2020-3417484.1. |
| 〰️[PDF] | Consolvo, B. P., and M. Thornton, 2020, Microseismic event or noise: Automatic classification with convolutional neural networks: 90th Annual International Meeting, SEG, Expanded Abstracts. doi: https://doi.org/10.1190/segam2020-3414896.1. |
| 🌎[PDF] | Consolvo, B. P., 2018, Full-Waveform Inversion with Scaled-Sobolev Preconditioning Applied to Vibroseis Field Data: Western University Electronic Thesis and Dissertation Repository. doi: https://ir.lib.uwo.ca/etd/5199. |
| 🌎[PDF] | Consolvo, B. P., M. A. H. Zuberi, R. G. Pratt, and P. W. Cary, 2017, FWI with Scaled-Sobolev Preconditioning Applied to Short-offset Vibroseis Field Data: 79th Annual International Conference and Exhibition, EAGE, Extended Abstracts. doi: https://doi.org/10.3997/2214-4609.201701164. |
Some of the model cards I have contributed to Hugging Face.
| Title | Contribution Type |
|---|---|
| DPT-Large | Model Card |
| DPT-Hybrid | Model Card |
| 90% Sparse BERT-Base (uncased) Prune Once For All | Model Card |
| 90% Sparse DistilBERT-Base (uncased) Prune Once for All | Model Card |
| 90% Sparse BERT-Large (uncased) Prune Once for All | Model Card |
| 85% Sparse DistilBERT-Base (uncased) Prune Once for All | Model Card |
| 85% Sparse BERT-Base (uncased) Prune Once For All | Model Card |
| 80% 1x4 Block Sparse BERT-Base (uncased) Fine Tuned on SQuADv1.1 | Model Card |
| Question & Answer with Sparse BERT using the SQuAD dataset | Space |
| Dynamic-TinyBERT | Model Card |
| INT8 DistilBERT base uncased finetuned SST-2 | Model Card |
Here are a few of my Python notebooks I have published on Kaggle, including one detailing the hardware available on the platform.
| Title |
|---|
| Training humor detection with DistilBERT |
| Hardware Available on Kaggle |
| U-Net Convolutional Neural Network - Salt or Not |
A sample of my direct contributions to the GitHub open-source community.
| Activity Title & Link | Contribution Type | Description |
|---|---|---|
| Intel® Optimized Cloud Modules for GCP: nanoGPT Distributed Training | Repo | A guided tour on fine-tuning nanoGPT (124M parameter) model on a cluster of CPUs on Google Cloud Platform. |
| CPU Accelerated Fine-Tuning for Image Segmentation using PyTorch | Python Notebook | Comprehensive tutorial on a deep learning (pixel segmentation) task on the official Intel Extension for PyTorch repository. An accompanying blog was posted on the official PyTorch website here. |
| Remove metadata.clusterName entirely from cluster.yaml | Pull Request | In deploying Kubeflow on GCP, I noticed problems with the cluster.yaml file and contributed to the formal GCP implementation |
| Natural Language Processing: Detecting Humor with DistilBERT on Habana Gaudi | Repo | Led an AI hackathon for a NLP task of detecting humor, using deep learning and Habana Gaudi HPU accelerators. |
| Podcast | Episode Title | Apple | Spotify | Published Date | |
|---|---|---|---|---|---|
| Code Together | How Copilot, ChatGPT, Stable Diffusion and Generative AI Will Change How We Develop, Work and Live | Apple | Spotify | Dec. 8, 2022 |
My journey in work has mostly been focused around AI and geophysics. For more detail, please visit my LinkedIn profile.
| Company | Role | Location | Dates |
|---|---|---|---|
| Intel | AI Engineering Manager | Conroe, TX | 04/2022 - Present |
| Zvelo | Senior AI Engineer (Computer Vision) | Spring, TX | 06/2020 - 04/2022 |
| Fairfield Geotechnologies | Research Data Scientist and Geophysicist (FWI) | Houston, TX | 05/2019 - 05/2020 |
| MicroSeismic | Python Developer, Geophysicist, Field Geophysicist | Houston, TX | 02/2018 - 04/2019 |
| ExxonMobil | Geophysics Intern (FWI Research) | Spring, TX | 01/2017 - 06/2017 |
Starting from a strong foundation of mathematics, I moved into teaching and then a thesis-based geophysics degree.
| School | Degree | Location | Dates |
|---|---|---|---|
| University of Western Ontario | Master of Science in Geophysics | London, Ontario | 09/2015 - 01/2018 |
| Crandall University | Bachelor of Education in K-12 Teaching | Moncton, New Brunswick | 09/2010 - 05/2012 |
| Queen's University | Bachelor of Science in Mathematics | Kingston, Ontario | 09/2006 - 05/2010 |
How to reach me.
If you would like to connect with me to collaborate or to ask me questions, feel free to reach out to me over LinkedIn.
React
Typescript
Django
Laravel
Facebook
Microsoft
Google
Alibaba
D3
Tencent