Comprehensive List of Small LLMs, the Mini-Giants of the LLM World

January 24, 2024


Large Language Models (LLMs) mark a significant evolution in the field of AI and NLP. Initially, language models were simple and small, but as the demand for more sophisticated and accurate natural language understanding (NLU) and natural language generation grew, so did the size and complexity of these models. This led to the emergence of models like ChatGPT, Llama, and BERT, which showcased an unprecedented ability to understand context, generate text, and even perform specific language tasks without extensive task-specific training. These models, characterized by their large number of parameters, have set new standards for what's possible in AI-driven communication, creativity, and comprehension.

While the large number of their parameters makes them performant, this aspect also places significant demands on computational resources. Training and running LLMs require lots of energy, time, and money, making them less accessible to the average researcher, developer, or small organization. Furthermore, as demand for LLMs grows across industries, so does the need for smaller domain-specific models. 

Due to this, a new wave of small LLMs are arising, collectively known as mini-giants. These models are a response to the growing need for more accessible, efficient, and ethical AI tools. Mini-giants embody most of the key capabilities of their larger counterparts but are designed to be significantly smaller in size, requiring less computational power and resources to run. Despite their smaller size, these models are making impressive strides in performance, often rivaling or even surpassing larger models in specific tasks. This represents a paradigm shift in AI, prioritizing not just capability but also accessibility and efficiency. This can improve the reach of LLMs, making it a more integrated and practical part of everyday solutions across industries.

Why Build Smaller-Sized LLMs?

Small LLMs are rapidly gaining popularity within the open-source community, reflecting a broader trend towards more accessible and efficient AI tools. They address several critical challenges associated with large LLMs:

  • Costs: Training and maintaining large-scale LLMs involve substantial investment, making them too resource-intensive for use in myriad use-cases where smaller models could suffice. 
  • Resources: The computational power required to train and run large models is immense. For complex and sophisticated AI applications, organizations commit to GPU setups like HGX 8xH100 or A100s. However, for simpler tasks, often less expensive GPUs such as L40S, A30, A40 suffice.
  • Accessibility: Large models are less flexible and harder to adapt to specific needs or constraints. Smaller models, on the other hand, offer greater adaptability and are easier to fine-tune for specialized tasks or to fit within the constraints of specific hardware, making them suitable for a wide range of applications, from mobile devices to embedded systems.

The rise of mini-giants is a response to these challenges, representing a shift towards wider applications of AI. By reducing the barriers to entry and making powerful language models more widely available, mini-giants can become a catalyst for a more inclusive and responsible approach to AI.

Strategies for Downsizing LLMs

There are several strategies that have been developed to downsize LLM, while maintaining performance. These techniques vary in approach and impact.

  • Model Pruning involves removing the least important weights or neurons from a model, reducing its size without significantly impacting performance. This technique improves the model, making it faster and less resource-intensive, although it requires careful implementation to ensure vital features aren't lost in the pruning process.
  • Knowledge Distillation is a method where a smaller model learns to replicate the behavior of a larger model. By training the student model to mimic the outputs of the teacher, knowledge distillation captures the essence of the larger model's capabilities, resulting in a compact model with performance that can approach that of the original, depending on the intricacies of the relationship and training methods.
  • Quantization decreases the precision of the model's parameters, converting them from floating-point representations to lower-bit formats. This reduction also accelerates inference times. The trade-off typically involves a minimal loss in accuracy, which can often be mitigated or calibrated for specific tasks.
  • Parameter Sharing reduces the model's footprint by using the same parameters across multiple model components, reducing the total number of unique parameters without drastically sacrificing the model's expressive power. While it improves the model, it may also limit the model's flexibility in handling diverse or complex tasks.

Enhancing fine-tuning efficiency is also crucial. It makes the model more adaptable and easier to customize with less computational resources. 

  • Transfer Learning involves adapting a pre-trained model to new tasks by fine-tuning it on a smaller, task-specific dataset. This approach is resource-efficient and uses the model's pre-learned features, although its effectiveness depends on the relevance of the pre-trained model to the new task.
  • Instruction Tuning involves training a small set of parameters or using prompts to guide the model's responses to specific tasks. This method allows rapid adaptation with minimal resource requirements but may vary in effectiveness based on how different the new tasks are from the original training.
  • Adapters are small, task-specific modules inserted into a pre-trained model, allowing targeted adjustments without retraining the entire model. Adapters offer a balance between customization and efficiency, enabling more precise model tuning with less computational overhead.

Each of these strategies reflects a commitment to making language models more usable and effective across a broader range of scenarios. As the field continues to evolve, these techniques are refined and combined, driving towards an era of more sustainable, adaptable, and accessible AI.

Comparative Study of Small Language Models

When choosing small language models, it is crucial to consider various factors like parameter size, performance, innovation, and real-world applications. Let us look at some of the notable small LLMs that have made significant impacts in the field.

  • DistilBERT from Hugging Face combines the essence of BERT's language understanding capabilities in a package that's 40% smaller. Despite its reduced size, DistilBERT retains 97% of BERT's performance, making it an efficient alternative for a variety of NLP tasks. Developed by Hugging Face, this model is a testament to the power of model distillation techniques and is widely used in scenarios where efficiency and speed are critical without substantially compromising the quality of results.
  • GPT-Neo from EleutherAI is an open-source initiative aimed at democratizing access to powerful language models. Available in different sizes, with a popular variant around 2.7 billion parameters, GPT-Neo seeks to provide performance comparable to GPT-3. It's particularly notable for its commitment to open-source principles, allowing for broad adaptation and innovation across various applications from text generation to more sophisticated conversational agents.
  • TinyBERT from Huawei is specifically designed for environments where efficiency is crucial, such as mobile or edge computing. Despite its significantly smaller size compared to the original BERT, TinyBERT uses competitive performance on language understanding benchmarks. The model focuses on transformer distillation techniques, offering a solution that balances the need for compact, efficient models with the desire for robust performance in language tasks.

MobileBERT is optimized for speed and size, specifically targeting mobile platforms. This model is a variant of BERT tailored for on-device performance, providing lower latency and enhanced privacy for users. MobileBERT's architecture can be viewed as making powerful NLP tools more accessible and practical for real-world mobile applications, enabling robust language understanding and generation tasks directly on users' devices.

In summary, these small language models represent a significant shift towards making NLP more accessible and efficient. Each brings unique innovations and strengths to the table, from DistilBERT's efficient distillation approach to GPT-Neo's open-source accessibility, TinyBERT's focus on transformer techniques, and MobileBERT's mobile optimization. As the field evolves, these mini-giants continue to push the boundaries of what's possible, offering powerful capabilities for a wide range of applications while addressing the constraints of resource efficiency and accessibility.

Real-World Applications and Advantages

Small LLMs offer a lot of advantages that make them particularly suitable for a wide range of real-world applications. Two of the most significant advantages are their ability to protect user privacy and their computational efficiency.

  • Privacy Protection: In a time period where data privacy is important, small LLMs offer a solution. They can be deployed locally on a user's device or within an organization's private infrastructure, significantly reducing the risk of sensitive data being exposed to third-party servers. This local deployment capability is especially crucial for industries handling sensitive information, such as healthcare, finance, and legal services. By keeping data on-premises, organizations can maintain control over their information and adhere to strict privacy regulations and standards.
  • Computation Efficiency: Mini-giants are designed to be lean and efficient, requiring fewer computational resources than their larger counterparts. This efficiency leads to faster training and inference times, lower energy consumption, and reduced operational costs. The ability to run on less powerful hardware, including mobile devices and embedded systems, opens up new possibilities for integrating advanced language processing capabilities into a broader array of products and services. 

Case Study of a Small LLM

A compelling case study of small LLMs is seen in the application of a hypothetical therapeutic chatbot. This chatbot integrates principles from Cognitive Behavioral Therapy (CBT) to provide timely mental health support and interventions. It's designed to engage users in meaningful conversation, offering guidance and support for mental wellness.

  • Increased Privacy: Operating directly on a user's device, the chatbot ensures privacy and confidentiality, a must-have for users sharing sensitive personal information. This on-device deployment makes mental health support accessible anytime, overcoming barriers to traditional therapy such as availability, stigma, and cost.
  • Efficiency in Interactions: The chatbot would respond promptly and contextually to user inputs. This immediate interaction in spite for light weight keeps users engaged and ensures that the therapeutic advice is consistent, personalized, and timely. The chatbot's ability to learn and adapt to individual user needs while maintaining a conversational flow is a direct benefit of the underlying small LLM's capabilities.

The prospective applications and advantages of mini-giants are vast and impactful. By prioritizing privacy and efficiency, these models are well-positioned to revolutionize how industries operate and how services are delivered, particularly in fields where privacy is critical. The hypothetical case study of a therapeutic chatbot is just one example of how small LLMs can be applied to address complex challenges in daily life. As small LLMs continue to evolve, their role in driving innovation and improving services is expected to expand, marking a significant shift towards more accessible, efficient, and responsible AI.

Technical Considerations

Integrating small LLMs into various domains, particularly sensitive ones like healthcare, presents a unique set of technical challenges. These challenges need to be carefully navigated to ensure the responsible and beneficial use of AI technologies.

  • Integration with Existing Systems: In many industries, integrating new technologies with existing systems and workflows can be complex and resource-intensive. Small LLMs must be adaptable and compatible with different types of infrastructure and data formats, requiring robust and flexible design.
  • Maintaining Accuracy: While smaller models offer efficiency, ensuring they maintain a high level of accuracy and reliability, especially in critical domains like healthcare, is essential. Continuous monitoring and validation are required to ensure that the models perform consistently and as expected.
  • Handling Bias: Small LLMs must be capable of understanding and processing a wide variety of data types and languages. Additionally, they need to be trained and regularly updated to avoid biases and inaccuracies that could lead to harmful decisions, especially in sensitive applications

Ethical Considerations

Similarly, small LLMs also pose ethical challenges, which needs to be addressed. These considerations can be considered not only for small LLMs, but for LLMs in general.

  • Data Protection: Ensuring user data privacy and protection is necessary, especially when dealing with sensitive personal information. Small LLMs need to be designed with privacy-preserving features, and their deployment must comply with data protection regulations such as GDPR, HIPAA, or others relevant to the specific industry and region.
  • Transparency: There is a growing demand for AI systems to be transparent, especially when they're used in critical decision-making. Users should understand how the model makes decisions, what data it uses, and what limitations it might have. This is particularly important in healthcare, where understanding the rationale behind a diagnosis or treatment recommendation is crucial.
  • Ethical Use: As small LLMs become more widely used, ensuring that they are deployed and used ethically is essential. This involves considering the potential impacts on academic cheating,  employment, mental health, and societal norms. It also means avoiding the use of these models in applications that could lead to harm or discrimination.
  • Regulatory Compliance: Each industry, especially healthcare, is governed by a set of regulations and standards that ensure safety and efficacy. Small LLMs must be developed and deployed in compliance with these standards, which might involve rigorous testing, certification, and ongoing oversight. Navigating the regulatory landscape can be complex, requiring a clear understanding of the requirements and processes involved. 

Future Trends and Conclusion

The open-source community has been a significant beneficiary of a diverse ecosystem. Platforms like Hugging Face and initiatives like EleutherAI have created communities around sharing, improving, and deploying small LLM models. This collaborative environment accelerates the development of small LLMs and ensures that the knowledge and benefits are shared. The landscape of small LLMs is evolving rapidly, driven by a combination of technological advancements, growing demand, and a vibrant community of developers and users. As these models become more sophisticated and widespread, they will play an increasingly central role in shaping the future of AI, making it more accessible, efficient, and impactful. The journey of small LLMs is just beginning, and the potential for positive change and innovation is vast, promising a future where AI is not only powerful but also widely available and beneficial for everyone.

These models are set to transform industries by providing powerful tools that were once the domain of only the most resource-rich organizations. Now, they are within reach of a broader audience, from independent developers and startups to educational institutions and non-profits. The significance of mini-giants lies in their potential to democratize AI. By reducing the resource requirements traditionally associated with sophisticated AI applications, they open up opportunities for innovation and development to a much wider community. 

Small LLMs are also transforming industries by enabling more efficient and effective solutions. In industries like healthcare, education, and customer service,  it is possible to provide high-quality, AI-powered services and insights that are faster, more accurate, and more accessible. They can also be deployed in a variety of contexts, from mobile apps and embedded systems to large-scale enterprise solutions.

The journey of mini-giants is far from over, and ongoing innovation is crucial to realizing their full potential. As the technology continues to evolve, it is important for the developers, researchers, users, and policymakers to stay engaged, share knowledge, and collaborate on developing best practices and standards. This engagement includes not only technical development but also a commitment to ethical considerations. As the use of mini-giants expands, it is imperative to ensure that they are developed and deployed responsibly, with attention to issues like privacy, bias, and transparency. The community must work together to address these challenges, ensuring that AI serves the greater good.

Small LLMs are a beacon of innovation and democratization in the AI landscape. They embody the promise of AI that is accessible, powerful, and beneficial, marking a significant step towards a future where the advantages of AI are shared widely and responsibly. The continued evolution of these models will bring new challenges, but with ongoing innovation, community engagement, and a commitment to responsible development, the potential for positive impact is immense. 

Latest Blogs
This is a decorative image for: A Complete Guide To Customer Acquisition For Startups
October 18, 2022

A Complete Guide To Customer Acquisition For Startups

Any business is enlivened by its customers. Therefore, a strategy to constantly bring in new clients is an ongoing requirement. In this regard, having a proper customer acquisition strategy can be of great importance.

So, if you are just starting your business, or planning to expand it, read on to learn more about this concept.

The problem with customer acquisition

As an organization, when working in a diverse and competitive market like India, you need to have a well-defined customer acquisition strategy to attain success. However, this is where most startups struggle. Now, you may have a great product or service, but if you are not in the right place targeting the right demographic, you are not likely to get the results you want.

To resolve this, typically, companies invest, but if that is not channelized properly, it will be futile.

So, the best way out of this dilemma is to have a clear customer acquisition strategy in place.

How can you create the ideal customer acquisition strategy for your business?

  • Define what your goals are

You need to define your goals so that you can meet the revenue expectations you have for the current fiscal year. You need to find a value for the metrics –

  • MRR – Monthly recurring revenue, which tells you all the income that can be generated from all your income channels.
  • CLV – Customer lifetime value tells you how much a customer is willing to spend on your business during your mutual relationship duration.  
  • CAC – Customer acquisition costs, which tells how much your organization needs to spend to acquire customers constantly.
  • Churn rate – It tells you the rate at which customers stop doing business.

All these metrics tell you how well you will be able to grow your business and revenue.

  • Identify your ideal customers

You need to understand who your current customers are and who your target customers are. Once you are aware of your customer base, you can focus your energies in that direction and get the maximum sale of your products or services. You can also understand what your customers require through various analytics and markers and address them to leverage your products/services towards them.

  • Choose your channels for customer acquisition

How will you acquire customers who will eventually tell at what scale and at what rate you need to expand your business? You could market and sell your products on social media channels like Instagram, Facebook and YouTube, or invest in paid marketing like Google Ads. You need to develop a unique strategy for each of these channels. 

  • Communicate with your customers

If you know exactly what your customers have in mind, then you will be able to develop your customer strategy with a clear perspective in mind. You can do it through surveys or customer opinion forms, email contact forms, blog posts and social media posts. After that, you just need to measure the analytics, clearly understand the insights, and improve your strategy accordingly.

Combining these strategies with your long-term business plan will bring results. However, there will be challenges on the way, where you need to adapt as per the requirements to make the most of it. At the same time, introducing new technologies like AI and ML can also solve such issues easily. To learn more about the use of AI and ML and how they are transforming businesses, keep referring to the blog section of E2E Networks.

Reference Links

This is a decorative image for: Constructing 3D objects through Deep Learning
October 18, 2022

Image-based 3D Object Reconstruction State-of-the-Art and trends in the Deep Learning Era

3D reconstruction is one of the most complex issues of deep learning systems. There have been multiple types of research in this field, and almost everything has been tried on it — computer vision, computer graphics and machine learning, but to no avail. However, that has resulted in CNN or convolutional neural networks foraying into this field, which has yielded some success.

The Main Objective of the 3D Object Reconstruction

Developing this deep learning technology aims to infer the shape of 3D objects from 2D images. So, to conduct the experiment, you need the following:

  • Highly calibrated cameras that take a photograph of the image from various angles.
  • Large training datasets can predict the geometry of the object whose 3D image reconstruction needs to be done. These datasets can be collected from a database of images, or they can be collected and sampled from a video.

By using the apparatus and datasets, you will be able to proceed with the 3D reconstruction from 2D datasets.

State-of-the-art Technology Used by the Datasets for the Reconstruction of 3D Objects

The technology used for this purpose needs to stick to the following parameters:

  • Input

Training with the help of one or multiple RGB images, where the segmentation of the 3D ground truth needs to be done. It could be one image, multiple images or even a video stream.

The testing will also be done on the same parameters, which will also help to create a uniform, cluttered background, or both.

  • Output

The volumetric output will be done in both high and low resolution, and the surface output will be generated through parameterisation, template deformation and point cloud. Moreover, the direct and intermediate outputs will be calculated this way.

  • Network architecture used

The architecture used in training is 3D-VAE-GAN, which has an encoder and a decoder, with TL-Net and conditional GAN. At the same time, the testing architecture is 3D-VAE, which has an encoder and a decoder.

  • Training used

The degree of supervision used in 2D vs 3D supervision, weak supervision along with loss functions have to be included in this system. The training procedure is adversarial training with joint 2D and 3D embeddings. Also, the network architecture is extremely important for the speed and processing quality of the output images.

  • Practical applications and use cases

Volumetric representations and surface representations can do the reconstruction. Powerful computer systems need to be used for reconstruction.

Given below are some of the places where 3D Object Reconstruction Deep Learning Systems are used:

  • 3D reconstruction technology can be used in the Police Department for drawing the faces of criminals whose images have been procured from a crime site where their faces are not completely revealed.
  • It can be used for re-modelling ruins at ancient architectural sites. The rubble or the debris stubs of structures can be used to recreate the entire building structure and get an idea of how it looked in the past.
  • They can be used in plastic surgery where the organs, face, limbs or any other portion of the body has been damaged and needs to be rebuilt.
  • It can be used in airport security, where concealed shapes can be used for guessing whether a person is armed or is carrying explosives or not.
  • It can also help in completing DNA sequences.

So, if you are planning to implement this technology, then you can rent the required infrastructure from E2E Networks and avoid investing in it. And if you plan to learn more about such topics, then keep a tab on the blog section of the website

Reference Links

This is a decorative image for: Comprehensive Guide to Deep Q-Learning for Data Science Enthusiasts
October 18, 2022

A Comprehensive Guide To Deep Q-Learning For Data Science Enthusiasts

For all data science enthusiasts who would love to dig deep, we have composed a write-up about Q-Learning specifically for you all. Deep Q-Learning and Reinforcement learning (RL) are extremely popular these days. These two data science methodologies use Python libraries like TensorFlow 2 and openAI’s Gym environment.

So, read on to know more.

What is Deep Q-Learning?

Deep Q-Learning utilizes the principles of Q-learning, but instead of using the Q-table, it uses the neural network. The algorithm of deep Q-Learning uses the states as input and the optimal Q-value of every action possible as the output. The agent gathers and stores all the previous experiences in the memory of the trained tuple in the following order:

State> Next state> Action> Reward

The neural network training stability increases using a random batch of previous data by using the experience replay. Experience replay also means the previous experiences stocking, and the target network uses it for training and calculation of the Q-network and the predicted Q-Value. This neural network uses openAI Gym, which is provided by taxi-v3 environments.

Now, any understanding of Deep Q-Learning   is incomplete without talking about Reinforcement Learning.

What is Reinforcement Learning?

Reinforcement is a subsection of ML. This part of ML is related to the action in which an environmental agent participates in a reward-based system and uses Reinforcement Learning to maximize the rewards. Reinforcement Learning is a different technique from unsupervised learning or supervised learning because it does not require a supervised input/output pair. The number of corrections is also less, so it is a highly efficient technique.

Now, the understanding of reinforcement learning is incomplete without knowing about Markov Decision Process (MDP). MDP is involved with each state that has been presented in the results of the environment, derived from the state previously there. The information which composes both states is gathered and transferred to the decision process. The task of the chosen agent is to maximize the awards. The MDP optimizes the actions and helps construct the optimal policy.

For developing the MDP, you need to follow the Q-Learning Algorithm, which is an extremely important part of data science and machine learning.

What is Q-Learning Algorithm?

The process of Q-Learning is important for understanding the data from scratch. It involves defining the parameters, choosing the actions from the current state and also choosing the actions from the previous state and then developing a Q-table for maximizing the results or output rewards.

The 4 steps that are involved in Q-Learning:

  1. Initializing parameters – The RL (reinforcement learning) model learns the set of actions that the agent requires in the state, environment and time.
  2. Identifying current state – The model stores the prior records for optimal action definition for maximizing the results. For acting in the present state, the state needs to be identified and perform an action combination for it.
  3. Choosing the optimal action set and gaining the relevant experience – A Q-table is generated from the data with a set of specific states and actions, and the weight of this data is calculated for updating the Q-Table to the following step.
  4. Updating Q-table rewards and next state determination – After the relevant experience is gained and agents start getting environmental records. The reward amplitude helps to present the subsequent step.  

In case the Q-table size is huge, then the generation of the model is a time-consuming process. This situation requires Deep Q-learning.

Hopefully, this write-up has provided an outline of Deep Q-Learning and its related concepts. If you wish to learn more about such topics, then keep a tab on the blog section of the E2E Networks website.

Reference Links

This is a decorative image for: GAUDI: A Neural Architect for Immersive 3D Scene Generation
October 13, 2022

GAUDI: A Neural Architect for Immersive 3D Scene Generation

The evolution of artificial intelligence in the past decade has been staggering, and now the focus is shifting towards AI and ML systems to understand and generate 3D spaces. As a result, there has been extensive research on manipulating 3D generative models. In this regard, Apple’s AI and ML scientists have developed GAUDI, a method specifically for this job.

An introduction to GAUDI

The GAUDI 3D immersive technique founders named it after the famous architect Antoni Gaudi. This AI model takes the help of a camera pose decoder, which enables it to guess the possible camera angles of a scene. Hence, the decoder then makes it possible to predict the 3D canvas from almost every angle.

What does GAUDI do?

GAUDI can perform multiple functions –

  • The extensions of these generative models have a tremendous effect on ML and computer vision. Pragmatically, such models are highly useful. They are applied in model-based reinforcement learning and planning world models, SLAM is s, or 3D content creation.
  • Generative modelling for 3D objects has been used for generating scenes using graf, pigan, and gsn, which incorporate a GAN (Generative Adversarial Network). The generator codes radiance fields exclusively. Using the 3D space in the scene along with the camera pose generates the 3D image from that point. This point has a density scalar and RGB value for that specific point in 3D space. This can be done from a 2D camera view. It does this by imposing 3D datasets on those 2D shots. It isolates various objects and scenes and combines them to render a new scene altogether.
  • GAUDI also removes GANs pathologies like mode collapse and improved GAN.
  • GAUDI also uses this to train data on a canonical coordinate system. You can compare it by looking at the trajectory of the scenes.

How is GAUDI applied to the content?

The steps of application for GAUDI have been given below:

  • Each trajectory is created, which consists of a sequence of posed images (These images are from a 3D scene) encoded into a latent representation. This representation which has a radiance field or what we refer to as the 3D scene and the camera path is created in a disentangled way. The results are interpreted as free parameters. The problem is optimized by and formulation of a reconstruction objective.
  • This simple training process is then scaled to trajectories, thousands of them creating a large number of views. The model samples the radiance fields totally from the previous distribution that the model has learned.
  • The scenes are thus synthesized by interpolation within the hidden space.
  • The scaling of 3D scenes generates many scenes that contain thousands of images. During training, there is no issue related to canonical orientation or mode collapse.
  • A novel de-noising optimization technique is used to find hidden representations that collaborate in modelling the camera poses and the radiance field to create multiple datasets with state-of-the-art performance in generating 3D scenes by building a setup that uses images and text.

To conclude, GAUDI has more capabilities and can also be used for sampling various images and video datasets. Furthermore, this will make a foray into AR (augmented reality) and VR (virtual reality). With GAUDI in hand, the sky is only the limit in the field of media creation. So, if you enjoy reading about the latest development in the field of AI and ML, then keep a tab on the blog section of the E2E Networks website.

Reference Links

Build on the most powerful infrastructure cloud

A vector illustration of a tech city using latest cloud technologies & infrastructure