Quantized machine learning models with adapters
Patent Information
- Authority / Receiving Office
- WO · WO
- Patent Type
- Applications
- Current Assignee / Owner
- Filing Date
- 2025-02-10
- Publication Date
- 2026-08-13
Smart Images

Figure CN2025076601_13082026_PF_FP_ABST
Abstract
Description
QUANTIZED MACHINE LEARNING MODELS WITH ADAPTERSTECHNICAL FIELD
[0001] The disclosure relates generally to machine learning models. For example, aspects of the present disclosure include systems and techniques for configuring machine learning models with adapters.BACKGROUND
[0002] Machine learning systems (or models) , such as neural networks (e.g., deep neural networks) are widely used for numerous applications, such as generative operations (e.g., to generate images, language / text outputs, etc. ) , object detection, object classification, object tracking, big data analysis, among others. For example, convolutional neural networks (CNNs) are able to extract high-level features, such as facial shapes, from an input image, and use these high-level features to output a probability that, for example, an input image includes a particular object.
[0003] A generative machine learning system can process data to generate desired output content from an input (e.g., a natural language input, input image (s) or video (s) , a noise input such as for diffusion models, etc. ) . For instance, a language-based generative machine learning model (e.g., a large language model (LLM) ) can generate natural language responses from natural language inputs and can incorporate various forms of data, such as audio, images and text. In some cases, a generative machine learning system can include an encoder that processes input features to generate output features (also referred to as embeddings or encodings) and can use the output features to generate relevant output content in a given form (e.g., in a natural language form, as generated images and / or video, as audio, explanation of the content, etc. ) .
[0004] A generative ML system can perform a wide range of tasks such as answering questions, providing explanations, generating creative content, assisting with coding, and offering recommendations. Various tools may be connected to the generative ML system to allow interaction with external systems, such as browsing the Internet, generating images, executing code, etc. Generative ML systems are designed to assist users in solving problems, learning new information, and enhancing productivity.SUMMARY
[0005] The following presents a simplified summary relating to one or more aspects disclosed herein. Thus, the following summary should not be considered an extensive overview relating to all contemplated aspects, nor should the following summary be considered to identify key or critical elements relating to all contemplated aspects or to delineate the scope associated with any particular aspect. Accordingly, the following summary presents certain concepts relating to one or more aspects relating to the mechanisms disclosed herein in a simplified form to precede the detailed description presented below.
[0006] In some aspects, an apparatus for generating content is provided. The apparatus includes at least one memory and at least one processor coupled to the at least one memory and configured to: process weights associated with one or more layers of a machine learning model to generate quantized base model weights; process, using the machine learning model configured with the quantized base model weights and an adapter connected to at least one layer of the one or more layers, data to generate output values; modify adapter weights (of the adapter) based on the output values; process the modified adapter weights to generate quantized adapter weights; process activations of the adapter to generate quantized adapter activations; and process, using the machine learning model with the quantized base model weights and the adapter with the quantized adapter weights and adapter quantized activations, input data to generate output content.
[0007] In some aspects, a method for generating content is provided. The method includes: processing weights associated with one or more layers of a machine learning model to generate quantized base model weights; processing, using the machine learning model configured with the quantized base model weights and an adapter connected to at least one layer of the one or more layers, data to generate output values; modifying adapter weights (of the adapter) based on the output values; processing the modified adapter weights to generate quantized adapter weights; processing activations of the adapter to generate quantized adapter activations; and processing, using the machine learning model with the quantized base model weights and the adapter with the quantized adapter weights and adapter quantized activations, input data to generate output content.
[0008] In some aspects, a non-transitory computer-readable medium is provided that has stored thereon instructions that, when executed by one or more processors, cause the one or more processors to: process weights associated with one or more layers of a machine learning model to generate quantized base model weights; process, using the machine learning model configured with the quantized base model weights and an adapter connected to at least one layer of the one or more layers, data to generate output values; modify adapter weights (of the adapter) based on the output values; process the modified adapter weights to generate quantized adapter weights; process activations of the adapter to generate quantized adapter activations; and process, using the machine learning model with the quantized base model weights and the adapter with the quantized adapter weights and adapter quantized activations, input data to generate output content.
[0009] In some aspects, an apparatus for generating content is provided. The apparatus includes: means for processing weights associated with one or more layers of a machine learning model to generate quantized base model weights; means for processing, based on the quantized base model weights and an adapter connected to at least one layer of the one or more layers, data to generate output values; means for modifying adapter weights (of the adapter) based on the output values; processing the modified adapter weights to generate quantized adapter weights; means for processing activations of the adapter to generate quantized adapter activations; and means for processing, based on the quantized base model weights and the adapter with the quantized adapter weights and adapter quantized activations, input data to generate output content.
[0010] In some aspects, one or more of the apparatuses described herein is, is part of, and / or includes a mobile device (e.g., a mobile telephone or other mobile devices) or other wireless communication device, a vehicle or a computing device or component of a vehicle, an extended reality (XR) device or system (e.g., a virtual reality (VR) device, an augmented reality (AR) device, or a mixed reality (MR) device) , a wearable device, a camera, a personal computer, a laptop computer, a server computer or server device (e.g., an edge or cloud-based server, a personal computer acting as a server device, a mobile device such as a mobile phone acting as a server device, an XR device acting as a server device, a vehicle acting as a server device, a network router, or other device acting as a server device) , another device, or a combination thereof. In some aspects, each apparatus can include a camera or multiple cameras for capturing one or more images. In some aspects, each apparatus can include a display or multiple displays for displaying one or more images, notifications, and / or other displayable data. In some aspects, each apparatus can include one or more sensors (e.g., one or more inertial measurement units (IMUs) , such as one or more gyroscopes, one or more gyrometers, one or more accelerometers, or any combination thereof, and / or other sensor.
[0011] This summary is not intended to identify key or essential features of the claimed subject matter, nor is it intended to be used in isolation to determine the scope of the claimed subject matter. The subject matter should be understood by reference to appropriate portions of the entire specification of this patent, any or all drawings, and each claim.
[0012] The foregoing, together with other features and aspects, will become more apparent upon referring to the following specification, claims, and accompanying drawings.BRIEF DESCRIPTION OF THE DRAWINGS
[0013] Illustrative aspects of the present application are described in detail below with reference to the following drawing figures:
[0014] FIG. 1 illustrates an example implementation of a system, which may include a central processing unit (CPU) , configured to perform one or more of the functions described herein;
[0015] FIG. 2 is a block diagram illustrating a multimodal generative machine learning system for generating natural language responses based on natural language input from a prompt and any additional information, in accordance with aspects of the disclosure;
[0016] FIG. 3 is a block diagram illustrating a generative response engine, in accordance with aspects of the disclosure;
[0017] FIG. 4 is a block diagram illustrating an example of a system including a base model and various adapters, in accordance with aspects of the disclosure;
[0018] FIG. 5 is a block diagram of a process for providing a base model weight activation quantization aware Low Rank Adapter (LoRA) , in accordance with aspects of the disclosure;
[0019] FIG. 6 is a block diagram of a process for providing a base model weight quantization aware LoRA, in accordance with aspects of the disclosure;
[0020] FIG. 7 is a flow diagram illustrating an example of a process for configuring adapters for machine learning models, in accordance with aspects of the disclosure;
[0021] FIG. 8 is a block diagram of an example transformer in accordance with some aspects of the disclosure; and
[0022] FIG. 9 shows an example of a computing system, which may be for example any computing device that may implement components of the system.DESCRIPTION
[0023] Certain aspects of this disclosure are provided below. Some of these aspects may be applied independently and some of them may be applied in combination as would be apparent to those of skill in the art. In the following description, for the purposes of explanation, specific details are set forth in order to provide a thorough understanding of aspects of the application. However, it will be apparent that various aspects may be practiced without these specific details. The figures and description are not intended to be restrictive.
[0024] The ensuing description provides example aspects only, and is not intended to limit the scope, applicability, or configuration of the disclosure. Rather, the ensuing description of the example aspects will provide those skilled in the art with an enabling description for implementing an example aspect. It should be understood that various changes may be made in the function and arrangement of elements without departing from the spirit and scope of the application as set forth in the appended claims.
[0025] Machine learning models can be trained to perform various functions and / or provide various types of outputs. For instance, some generative machine learning models can provide a conversational interface that uses natural language prompts as inputs, such as text or voice. In some examples, a user can provide an input prompt in natural language to the generative machine learning model, and the generative machine learning model can provide a response in natural language form. The input prompt and the output response can optionally be combined with one or more other types of information or data, such as images or files.
[0026] Previously trained machine learning models, such as generative machine learning models, can be fine-tuned to improve performance of the machine learning models for specific tasks. Fine-tuning can involve adapting a pre-trained machine learning model (e.g., a pre-trained generative machine learning model or other type of machine learning model) to a specific task by updating parameters (e.g., weights and / or other parameters) of the machine learning model using task-specific training data. The fine-tuning process allows the machine learning model to leverage its pre-existing knowledge while specializing in new tasks. Fine-tuning typically requires less data and computing than fully training a previously untrained machine learning model. Fine-tuning can thus be useful for improving performance of machine learning models on domain-specific tasks without sacrificing the benefits of large-scale pretraining.
[0027] One technique for fine tuning machine learning models is to use an adapter, such as a Low-Rank Adapter (LoRA) . An adapter is a lightweight trainable model that connects to different layers (or blocks) of a machine learning model. Adapters can be used to efficiently adapt large pre-trained models to specific tasks with minimal computational cost and resource usage. Adapters (e.g., a LoRA) can be trained to generalize task-specific information to a frozen (e.g., immutable) pre-trained machine learning model.
[0028] For instance, LoRA introduces lightweight trainable modules parameterized as low-rank matrices into an original (e.g., pre-trained) machine learning model to capture task-specific information while the original machine learning model parameters remain frozen, preserving the pre-trained knowledge of the machine learning model. Training a LoRA can include injecting low-rank layers into certain layers of the base model (e.g., attention layers in a transformer model, feedforward layers in the transformer model, convolutional layers in a transformer and / or convolutional model, etc. ) to optimize the layers during training and reduce the number of overall parameters that need to be updated. Such a technique can be performed to overcome the computational challenges of fine-tuning large machine learning models, such as when working with a smaller dataset.
[0029] While machine learning models (e.g., neural networks) are powerful architectures capable of a wide range of useful tasks, such as recognizing objects in image data, they are likewise highly resource dependent. For example, neural networks may require significant compute, memory, power, and / or time resources for training and / or for inferencing. These resource requirements may significantly limit the ability to train and deploy neural networks to certain types of devices and for certain use cases. For instance, training of machine learning models may be a computationally intensive process that can take a relatively long time, a large quantity of training data, and many operations.
[0030] Quantization is a method of mapping continuous values to a smaller set of discrete finite values. For example, quantization approximates real-world values (e.g., floating point values) with representative values (e.g., integer values) that limit the precision and range of the original input. When applied to machine learning, such as to a neural network, quantization may significantly reduce resource usage for both training and inferencing. For example, performing massive numbers of integer operations during training of or inferencing with a quantized (e.g., reduced-precision) neural network may be significantly more efficient in terms of resource usage as compared to performing floating point operations with an unquantized (e.g., full-precision) neural network processing the same input data.
[0031] Quantization of trained machine learning models allows quantized trained machine learning models to be efficiently deployed on various devices. Quantization may allow a quantized trained machine learning model to perform operations (e.g., at the inference phase of operation) on a device relatively quickly. Quantization may include changing a format of parameters (e.g., weights and activations) of a machine learning model from the format in which the machine learning model was trained to a different format.
[0032] Post-Training Quantization (PTQ) is a process by which a trained machine learning model is quantized. For example, during training, a machine learning model may store parameters (e.g., weights) according to a first format that may allow a high degree of precision. For example, during training, the model may store parameters as floating-point numbers, such as 16-bit floating point numbers, which may be referred to as float16 or FP16, or 32-bit floating point numbers, which may be referred to as float32 or FP32. According to PTQ, after training is performed to generate the trained machine learning model and before the trained machine learning model is deployed onto a device for use by a device, the model may be quantized by, in part, changing the format used to store the parameters of the model from the first format to a second format (e.g., an integer format where the parameters are represented as integer numbers instead of floating point numbers) . The second format may use less memory and / or be less computational expensive to use. For example, after quantization, the model may store parameters as integer numbers such as 16-bit integer numbers (which may be referred to as Int16) , 8-bit integer numbers (which may be referred to as Int8) , 4-bit integer numbers (which may be referred to as Int4) , etc. It may be less computationally expensive to store and / or operate using integer numbers than floating point numbers. Thus, a device may conserve power and / or processing time when using a quantized trained machine learning model as compared with using an unquantized trained machine learning model. Accordingly, it may be advantageous to quantize machine learning models.
[0033] Quantized Aware Training (QAT) is another quantization technique that allows a machine learning model to adapt to quantization during training, resulting in better performance after actual quantization is performed as compared to quantizing a fully trained model (e.g., using PTQ) . For example, using QAT, a machine learning model can be trained while simulating effects of actual quantization of weights and / or activations of the machine learning model during training. QAT can minimize accuracy loss when the machine learning model is eventually quantized to a different format, such as from floating point to an integer format, such as INT8.
[0034] In various aspects of the present disclosure, the term “quantize, ” “quantizing, ” and like terms may be used as a verb and may be applied to machine learning models, and / or layers of a trained machine learning model. In such cases the term “quantize, ” “quantizing, ” and like terms may refer to quantizing parameters (e.g., weights and / or activations) as stored in the of the respective machine learning models and / or layers. For example, a trained machine learning model may include weights (e.g., numerical values) . During training, and after training, the weights may be stored in the trained machine learning model in a first format (e.g., float8, float16, float32, float16) . Quantizing the trained machine learning model may include changing the weights from the first format to a second format (e.g., Int8) .
[0035] Further, machine learning models may include activations. Activations can refer to values used as inputs and / or outputs of various functions, operations, and / or layers (or blocks or modules) of the machine learning model. The activations may have a certain format. For example, a function (of a particular layer) may expect to receive values formatted according to the certain format. For instance, the function may read the values from memory according to the certain format. Further, the function may output other values (e.g., based on processing the values read from memory by the function) according to a particular format (which may or may not be the same as the certain format) . For instance, the function may write the other values to memory in the particular format. For example, an unquantized trained machine learning model may use FP16 to pass values between functions (e.g., outputting values from one function and reading the values by another function) . The activations of the machine learning model can also be quantized. For example, quantizing activations may refer to obtaining floating point (or continuous) values calculated by a layer (e.g., an output of an activation function of the layer) and converting the values into a smaller set of discrete values. For instance, the values may be represented with fewer bits, such as 8-bit integer numbers instead of 32-bit floating point numbers. After quantization of the activations the machine learning model is configured to process and / or output, the quantized trained machine learning model may use Int8 to pass the values between functions. The term “quantize, ” “quantizing, ” and like terms may refer to changing how activations are stored and / or used in a quantized machine learning model.
[0036] Quantizing numbers can result in a loss of precision. Quantizing a trained machine learning model can result in degradation of the model. In the present disclosure, the term “noise” is used to describe a degree of difference between outputs of an unquantized trained machine learning model and the trained machine learning model after being quantized (given the same inputs) . For example, a trained machine learning model may be quantized. The unquantized trained machine learning model may be provided an input, the quantized trained machine learning model may be provided the same input. Outputs of the models may be compared. A degree of difference between the outputs may be described as noise. Noise may be a way to describe degradation of a quantized machine learning model.
[0037] As mentioned, adapters such as LoRA can be used to efficiently adapt large pre-trained models to specific tasks with minimal computational cost and resource usage. However, a given adapter typically corresponds to a specific set of base model parameters and activations (e.g., parameters of the layer to which an adapter is attached) , which have different activation encodings. An activation encoding refers to the quantization level for a particular activation (e.g., the quantization step size or grid used for the activation) . In some cases, when using different adapters for a given layer (e.g., for different tasks) , the activation encodings can be different for the different adapters (referred to as a fully updatable encoding) . Such fully updateable encoding allows the activation encodings of the base model to be customized to a given task of the adapter. For instance, if two adapters are attached to a given layer of a base model, the activation encodings of the base model can be customized differently for each of the two adapters (e.g., a first set of base model activation encodings of a layer are selected when a first adapter is used for a first task, a second set of base model activation encodings of the layer are selected when a second adapter is used for a second task, etc. ) . However, allowing customized activation encodings of a given layer from the base model for different adapters requires the network to switch to the different base model activation encodings when switching from one adapter to the other. Given that there may be large numbers of layers in a machine learning model, such switching can result in large computational, memory, and latency overhead.
[0038] In some cases, the activation encodings for a given layer can remain fixed when using different adapters for that layer (referred to as adapter-only updatable encoding) . The adapter-only updatable encoding results in the machine learning model only selecting different adapter activation encodings when switching between different adapters for the layer of the base model, as opposed to also switching between different base model activation encodings as is done in the fully updateable encoding described above. Because the number of adapters (and corresponding parameters and activations) is much less than the number of layers in the base model (which can include thousands of layers) , using adapter-only updateable encoding is computationally and memory efficient. However, adapter-only updatable encoding may result in lower accuracy as compared to the fully updateable encoding due to the fixed base model activation encodings being used across adapters.
[0039] Systems, apparatuses, processes (also referred to as methods) , and computer-readable media (collectively referred to as “systems and techniques” ) are described herein for configuring machine learning models with adapters. The systems and techniques enable deployment of different adapters for a given base model without costly reconfiguration of the base model. For instance, certain aspects involve configuring weights and / or activations (e.g., activation encodings) of a base model and / or one or more adapters. Adapters may be adjusted based on the base model weights and / or activation encodings.
[0040] Various aspects of the disclosure are discussed in detail below. While specific implementations are discussed, it should be understood that this is done for illustration purposes only. A person skilled in the relevant art will recognize that other components and configurations may be used without parting from the spirit and scope of the disclosure.
[0041] FIG. 1 illustrates an example implementation of a system 100, which may include a central processing unit (CPU) , configured to perform one or more of the functions described herein. System 100, which may include a central processing unit (CPU 102) (which may be a multi-core CPU) . Parameters or variables (e.g., neural signals and synaptic weights) , system parameters associated with a computational device (e.g., neural network with weights) , task information, among other information may be stored in a memory block associated with a neural processing unit (NPU 108) , in a memory block associated with a CPU 102, in a memory block associated with a graphics processing unit (GPU 104) , in a memory block associated with a digital signal processor (DSP 106) , in a memory 116, and / or may be distributed across multiple blocks. Instructions executed at the CPU 102 may be loaded from a program memory associated with the CPU 102 or may be loaded from memory 116.
[0042] The system 100 may also include additional processing blocks tailored to specific functions, such as the GPU 104, the DSP 106, a connectivity engine 118, which may include fifth generation (5G) connectivity, fourth generation long term evolution (4G LTE) connectivity, Wi-Fi connectivity, USB connectivity, Bluetooth connectivity, and the like, and a multimedia processor 112 that may, for example, detect and recognize gestures. In one implementation, the NPU is implemented in the CPU 102, the DSP 106, and / or the GPU 104. The system 100 may also include one or more sensor processor (s) 114, one or more image signal processors (ISP (s) 110) , and / or navigation engine 120, which may include a global positioning system. In some examples, the sensor processor (s) 114 can be associated with or connected to one or more sensors for providing sensor input (s) to the sensor processor (s) 114. For example, the one or more sensors and sensor processor (s) 114 can be provided in, coupled to, or otherwise associated with a same computing device.
[0043] The system 100 may be implemented as a system on a chip (SoC) . The system 100 may be based on an Advanced Reduced Instruction Set Computer (RISC) Machine (ARM) instruction set. The system 100 and / or components thereof may be configured to perform machine learning techniques according to aspects of the present disclosure discussed herein. For example, the system 100 and / or components thereof may be configured to implement a machine learning model (e.g., a quantized trained machine learning model) as described herein and / or according to aspects of the present disclosure.
[0044] Machine learning (ML) can be considered a subset of artificial intelligence (AI) . ML systems can include algorithms and statistical models that computer systems can use to perform various tasks by relying on patterns and inference, without the use of explicit instructions. In some examples, an ML system can be a neural network (also referred to as an artificial neural network) , which may include an interconnected group of artificial neurons (e.g., neuron models) . Neural networks may be used for various applications and / or devices, such as image and / or video coding, image analysis and / or computer vision applications, Internet Protocol (IP) cameras, Internet of Things (IoT) devices, autonomous vehicles, service robots, among others.
[0045] Individual nodes in a neural network may emulate biological neurons by taking input data and performing simple operations on the data. The results of the simple operations performed on the input data are selectively passed on to other neurons. Weight values are associated with each vector and node in the network, and these values constrain how input data is related to output data. For example, the input data of each node may be multiplied by a corresponding weight value, and the products may be summed. The sum of the products may be adjusted by an optional bias, and an activation function may be applied to the result, yielding the node’s output signal or “output activation” (sometimes referred to as a feature map or an activation map) . The weight values may initially be determined by an iterative flow of training data through the network (e.g., weight values are established during a training phase in which the network learns how to identify particular classes by their typical input data characteristics) .
[0046] Different types of neural networks exist, such as convolutional neural networks (CNNs) , recurrent neural networks (RNNs) , generative adversarial networks (GANs) , multilayer perceptron (MLP) neural networks, transformer neural networks, diffusion-based neural networks, among others. For instance, convolutional neural networks (CNNs) are a type of feed-forward artificial neural network. Convolutional neural networks may include collections of artificial neurons that each have a receptive field (e.g., a spatially localized region of an input space) and that collectively tile an input space. RNNs work on the principle of saving the output of a layer and feeding this output back to the input to help in predicting an outcome of the layer. A GAN is a form of generative neural network that can learn patterns in input data so that the neural network model can generate new synthetic outputs that reasonably could have been from the original dataset. A GAN can include two neural networks that operate together, including a generative neural network that generates a synthesized output and a discriminative neural network that evaluates the output for authenticity. In MLP neural networks, data may be fed into an input layer, and one or more hidden layers provide levels of abstraction to the data. Predictions may then be made on an output layer based on the abstracted data.
[0047] Deep learning (DL) is an example of a machine learning technique and can be considered a subset of ML. Many DL approaches are based on a neural network, such as an RNN or a CNN, and utilize multiple layers. The use of multiple layers in deep neural networks can permit progressively higher-level features to be extracted from a given input of raw data. For example, the output of a first layer of artificial neurons becomes an input to a second layer of artificial neurons, the output of a second layer of artificial neurons becomes an input to a third layer of artificial neurons, and so on. Layers that are located between the input and output of the overall deep neural network are often referred to as hidden layers. The hidden layers learn (e.g., are trained) to transform an intermediate input from a preceding layer into a slightly more abstract and composite representation that can be provided to a subsequent layer, until a final or desired representation is obtained as the final output of the deep neural network.
[0048] As noted above, a neural network is an example of a machine learning system, and can include an input layer, one or more hidden layers, and an output layer. Data is provided from input nodes of the input layer, processing is performed by hidden nodes of the one or more hidden layers, and an output is produced through output nodes of the output layer. Deep learning networks typically include multiple hidden layers. Each layer of the neural network can include feature maps or activation maps that can include artificial neurons (or nodes) . A feature map can include a filter, a kernel, or the like. The nodes can include one or more weights used to indicate an importance of the nodes of one or more of the layers. In some cases, a deep learning network can have a series of many hidden layers, with early layers being used to determine simple and low-level characteristics of an input, and later layers building up a hierarchy of more complex and abstract characteristics.
[0049] A deep learning architecture may learn a hierarchy of features. If presented with visual data, for example, the first layer may learn to recognize relatively simple features, such as edges, in the input stream. In another example, if presented with auditory data, the first layer may learn to recognize spectral power in specific frequencies. The second layer, taking the output of the first layer as input, may learn to recognize combinations of features, such as simple shapes for visual data or combinations of sounds for auditory data. For instance, higher layers may learn to represent complex shapes in visual data or words in auditory data. Still higher layers may learn to recognize common visual objects or spoken phrases.
[0050] Deep learning architectures may perform especially well when applied to problems that have a natural hierarchical structure. For example, the classification of motorized vehicles may benefit from first learning to recognize wheels, windshields, and other features. These features may be combined at higher layers in different ways to recognize cars, trucks, and airplanes.
[0051] Neural networks may be designed with a variety of connectivity patterns. In feed-forward networks, information is passed from lower to higher layers, with each neuron in a given layer communicating to neurons in higher layers. A hierarchical representation may be built up in successive layers of a feed-forward network, as described above. Neural networks may also have recurrent or feedback (also called top-down) connections. In a recurrent connection, the output from a neuron in a given layer may be communicated to another neuron in the same layer. A recurrent architecture may be helpful in recognizing patterns that span more than one of the input data chunks that are delivered to the neural network in a sequence. A connection from a neuron in a given layer to a neuron in a lower layer is called a feedback (or top-down) connection. A network with many feedback connections may be helpful when the recognition of a high-level concept may aid in discriminating the particular low-level features of an input.
[0052] FIG. 2 is a block diagram illustrating a multimodal generative ML system 200 for generating natural language responses based on natural language input from a prompt 202, in accordance with aspects of the disclosure. A multimodal machine learning system is a machine learning model that receives, processes, and outputs data in multiple forms. For example, the input prompt may include text, images, and audio.
[0053] For example, the multimodal generative ML system 200 includes a plurality of encoders 204 that are each configured to encode different modes of content (e.g., text, images, audio, etc. ) into different tokens within a common embedding space. For example, a text input may be segmented based on different techniques (e.g., paragraph, sentence, etc. ) and encoded by a text encoder (from the encoders 204) into tokens. In another example, one or more images can be provided to an image encoder (from the encoders 204) that extracts features associated with the image and generates tokens representing the visual features. In another example, audio can be provided to an audio encoder (from the encoders 204) that extracts features associated with the image and generates tokens representing the audio features. In the case of audio, the audio encoder can identify features that can include formants that characterize resonant frequencies in speech, rhythmic features related to timing and tempo, and harmonic features that describe the relationship between fundamental frequencies and their harmonics.
[0054] The different tokens from the plurality of encoders are provided to combiner 206. The combiner 206 can combine the tokens based on the order in which they are presented. For example, the input into the encoder may be an array of primitive values. A primitive value is an immutable data type provided by a programming language and includes values that represent a single piece of data (e.g., number, string, Boolean, etc. ) rather than a complex object or reference. A non-limiting example prompt may include a byte array (e.g., an unsigned 8-byte integer array or uint8array) , and another string. The byte array may be audio, images, or other content that can be processed by the encoders 204. In some aspects, the combiner 206 is configured to concatenate the different tokens in order based on the array to preserve the semantic order of features and provide the tokens to the generative machine learning model 208.
[0055] The generative machine learning model 208 is configured to receive the tokens and generate a natural language response 212 based on the tokens and the prompt 202. Generative machine learning model 208 may include one or more models 210 (e.g., transformer neural network (s) , diffusion model (s) , fully connected layer (s) , multilayer perceptrons (MLPs) , any combination thereof, and / or other models) . The one or more models 210 of the generative machine learning model 208 are configured to process the tokens and extract different types of features that are relevant to the prompt 202. For example, the prompt 202 can be a query for a particular type of information. The one or more models of the generative machine learning model 208 can perform different tasks related to the query, such as writing code to perform a particular function, generating an image based on an input image with expressed modifications, generate an image without any input image, and so forth.
[0056] The generative machine learning model 208 may include different components, such as a featurization engine to identify different types of features, an inference engine to identify inferences within the text (e.g., pronoun usage and corresponding disambiguation functions) , data retrieval engines (e.g., to identify features related to a particular concept observed by the generative machine learning model 208) , and so forth. The generative machine learning model 208 may also include different types of models and engines to synthesize a coherent contextual output, such as to synthesize the input content and information that is responsive to tasks embedded within the text. For example, the generative machine learning model 208 may include a predictive output engine (not shown) that is configured to generate a sequence of words that is most likely contextually correct and to provide a coherent and contextually relevant answer. For instance, the predictive output generation engine can generate responses by sampling from the probability distribution of possible words and sequences based on patterns observed during training. The generative machine learning model 208 may also include a predictive output generation engine to generate multiple responses that are potentially relevant and coherent with respect to the prompt 202. The generative machine learning model 208 may also include an output validation engine configured to evaluate the generated responses based on certain criteria. Non-limiting examples of criteria to evaluate generated responses include relevance to the prompt, coherence, fluency, and adherence to specific guidelines or rules. Based on the evaluation, the output validation engine may select and output the most appropriate response.
[0057] As noted above, the generative machine learning model 208 may include various types of models (e.g., machine learning models) , such as a transformer. A transformer is a neural network architecture that can be trained to perform one or more natural language processing (NLP) tasks, such as language translation, sentiment analysis, and text summarization. Conventional traditional recurrent neural networks (RNNs) process data in sequence. A transformer or transformer network can process input in parallel and can thus be faster and more efficient than sequential training and processing. In some aspects, a transformer can use a self-attention mechanism (e.g., one or more self-attention layers) , which allows the transformer to identify the most relevant parts of the input text or content (e.g., audio or video) . In some cases, a transformer can also use a cross-attention mechanism (e.g., one or more cross-attention layers) which uses other content or data to determine the most relevant parts of the input. For example, cross-attention mechanisms are useful in sequential content such as a stream of data, such as optical flow, and other computer vision techniques.
[0058] A transformer neural network can include a multi-layer encoder-decoder architecture. For instance, an encoder of the encoder-decoder architecture can receive text as input, convert the input text into a sequence of hidden representations, and capture the meaning of the text at different levels of abstraction. A decoder of the encoder-decoder architecture can then process the representations output from the decoder to generate an output sequence, such as a text translation or a summary. The encoder and decoder can be trained together using supervised learning, unsupervised learning, or a combination of supervised and unsupervised learning techniques, such as maximum likelihood estimation and self-supervised pretraining. Illustrative examples of transformer engines include a BERT model, a Text-to-Text Transfer Transformer (T5) , biomedical BERT (BioBERT) , scientific BERT (SciBERT) , and the SPECTER model for document-level representation learning. In some aspects, multiple transformer engines may be used to generate different tokens.
[0059] In some aspects, the generative machine learning model 208 may be executed using a neural engine (or multiple neural engines) for on-device execution, such as a neural processing unit (NPU) , a neural signal processor (NSP) , a digital signal processor (DSP) , any combination thereof, and / or other neural engine. The neural engine can include a plurality of neural processing cores that are configured to parallelize operations associated with neural networks. A neural processing core can include arrays of multiply-accumulate (MAC) units and specialized instructions that are optimized for matrix operations, such as convolution and matrix multiplication. The neural processing core can receive input data and perform matrix transformations and nonlinear activation functions to break down and parallelize matrix operations. The neural processing core can perform tasks such as inference (e.g., runtime operation of a machine learning model) or training of deep learning models. The neural processing core can accelerate tasks by parallelization of larger computations that can be performed in parallel (e.g., matrix operations associated with neural networks) . For instance, the neural engine may perform computer vision tasks such as object recognition. In some cases, the neural engine can be implemented based on various ML libraries such as PyTorch, which interfaces with the compute unified device architecture (CUDA) to parallelize operations.
[0060] In some aspects, the generative machine learning model 208 may be a small generative model that has fewer parameters, fewer layers, fewer neurons, or a simpler architecture compared to larger models. A small generative model may not capture the full complexity of the underlying data distribution as effectively as larger models but can still be useful in scenarios where computational resources are limited or where a simpler model is sufficient for the task. Small generative models can also be easier to train and interpret, making them suitable for certain applications. For example, ChatGPT-3.5 has 175 billion parameters that results in a size of 1.4 Terabytes (TB) for a model implemented with double-precision floating point numbers. A smaller model may have a simpler architecture, use fewer parameters (e.g., 10 million) , and use less precise numbers (e.g., single-precision floating point numbers) resulting in a size of 38 Megabytes (MB) .
[0061] In addition, small models benefit from increased training based on local execution and data specific to a local device and a user of that local device. An additional benefit to small models is increased privacy because the information is not transmitted over the network and only relies on information requested by the user or usage at the local device.
[0062] FIG. 3 is a conceptual block diagram of a generative response engine 300 that is configured for multimodal input and output in accordance with some aspects of the disclosure. In some aspects, the generative response engine 300 is configured to receive a prompt 302 (e.g., from a user, an application programming interface (API) request, etc. ) and perform various processing to generate a response 304 that is responsive to the prompt 302 that is contextually accurate. In some aspects, the prompt 302 may be an array of primitive values, such as text, binary data (byte arrays, etc. For example, the prompt may be a union type that combines different types of primitive data (e.g., string uint8array in JavaScript) .
[0063] The generative response engine 300 includes a generative machine learning model 318. In some aspects, the generative response engine 300 includes a plurality of adapters (e.g., LoRA) that are trained to improve performance of a base machine learning model (also referred to as a base model) of the generative response engine 300 during inference. The plurality of adapters can be trained to perform different tasks. For example, a first adapter may be trained to perform a first task, and a second adapter a second task and so forth. In this manner, the independently trained adapters are configured to prevent the generative response engine 300 from defaulting to the application of the base model’s weights (which can be referred to as base weights) and allow the combination of the base model and the adapters to accurately provide specialized responses that correspond to the fine tuning.
[0064] The generative response engine 300 includes various components, such as a segmenter 310, a classifier 312, a plurality of encoders 314, a combiner 316, and the generative machine learning model 318. The segmenter 310 is configured to segment different content into different portions based on various limitations, such as text size, image size, and so forth. In some cases, the segmenter 310 may also reduce the scope of content (e.g., by reducing a size of an input image) based on limitations of one or more machine learning models of the generative response engine 300. For example, the machine learning model (s) may be trained on a particular size input to extract important features. The segmenter 310 may also segment text into different chunks, such as into individual sentences or paragraphs.
[0065] In some aspects, the classifier 312 may be configured to classify the input. For example, the classifier 312 may be configured to generate embeddings that represent features associated with the input (e.g., segmented text, images, etc. ) in a vector space. An embedding is a representation of a discrete object, such as a word, a document, or an image, as a continuous vector in a multi-dimensional space. An embedding captures the semantic or structural relationships between the objects, such that similar objects are mapped to nearby vectors, and dissimilar objects are mapped to distant vectors. Embeddings are commonly used in machine learning, computer vision, and natural language processing tasks, such as language modeling, sentiment analysis, and machine translation. Embeddings can be learned from large corpora of data using unsupervised learning algorithms, such as word2vec, GloVe, or fastText, which optimize the embeddings based on the co-occurrence or context of the objects in the data. Once learned, embeddings can be used to improve the performance of downstream tasks by providing a more meaningful and compact representation of the objects. In some aspects, the classifications are provided to the generative machine learning model 318 to activate specific learned domains.
[0066] In some cases, the classifier 312 may be configured to classify the input based on a plurality of clusters. The plurality of clusters may represent different semantic spaces for the operation of the generative response engine 300. The plurality of clusters may be related to different specialized tasks, data, or other operations and represent different domains of the generative response engine 300. For example, a generative model trained for customer service may need to learn different domains of knowledge, such as identification of products or services, accounting and billing information, a corpus of documents to help diagnose problems or provide information, and so forth. The plurality of clusters can be determined based on the input and output from the generative response engine 300.
[0067] In some cases, the classifier 312 may generate a probability for the prompt 302 (e.g., for each cluster) based on the output from the segmenter 310. The probability represents a likelihood that the prompt 302 is associated with features (e.g., features in a corresponding cluster) . In some aspects, the classifier 312 provides the probabilities associated with each cluster and the embeddings to the generative machine learning model 318.
[0068] The segmented portions of the prompt 302 can then be provided as input to a plurality of encoders 314 (e.g., the encoders 204 in FIG. 2) to extract various features and generate tokens associated with those features. In some aspects, each encoder 314 is configured for a different modality or different type of input. For example, the generative response engine 300 may include an encoder 314 for encoding text, an encoder 314 for encoding images, an encoder 314 for encoding audio, etc. There may be additional encoders based on the different modalities. For example, there may be an encoder 314 for handling different types of documents (e.g., PDF, etc. ) .
[0069] The combiner 316 (e.g., the combiner 206 in FIG. 2) receives the tokens from the encoders 314 and synthesizes tokens that represent the prompt 302. For example, the combiner 316 can be configured to concatenate the tokens from the encoders 314 in the correct order to preserve the semantic context of the content. In some cases, the encoders 314 may be configured to execute the encoding operations in parallel and because different content may be encoded at different rates, the tokens may be received out of order. The combiner 316 may receive information (e.g., from the segmenter 310 and / or from the encoders 314) to identify how to sequence the embeddings into an input suitable for the generative machine learning model 318 (e.g., an array of tokens) .
[0070] In some aspects, the generative machine learning model 318 is configured to use the embeddings during inference to begin inference operation based on the prompt 302. In some aspects, the probabilities from the classifier 312 can be used by the generative machine learning model 318 to activate at least one adapter based on the probability associated with that adapter. In some aspects, because the adapters are tuned independently based on clustered data, the adapters provide a stronger response based on the identification of features that more accurately represent that different classification.
[0071] The generative machine learning model 318 may be configured to synthesize the response 304 based on the probabilities provided by the classifier 312 and the generative machine learning model 318.
[0072] As noted previously, the systems and techniques described herein enable deployment of different adapters for a given base model without costly reconfiguration of activation encodings of the base model when switching between adapters.
[0073] FIG. 4 is a block diagram illustrating an example of a system 400 including a base model 410 and various adapters 420a-420n, in accordance with aspects of the disclosure. In some aspects, the adapters 420a-420n can include Low-Rank Adapters (LoRAs) or other types of adapters.
[0074] The base model 410 and / or the adapters 420a-n may be adjusted or calibrated (e.g., quantized) to generate one or more activation encodings, which represent how parameters of the model (e.g., tensors) are quantized. For instance, an activation encoding refers to the quantization level for a particular activation (e.g., the quantization step size or grid used for the activation) . As described herein, quantization refers to a reduction in precision of values or numbers representing parameters (e.g., weights) or activations. For example, the values or numbers can be quantized by converting the values or numbers from floating point values to integer values, or first floating-point values with a first precision to second floating point values with a second precision.
[0075] Activation encodings may be categorized as base activation encodings (applicable to layers of the base model 410) or adapter activation encodings (applicable to one or more of the adapters 420a-n) . Examples of base activation encodings include 417, 415, and 412. Examples of adapter activation encodings include 431a-n, 429a-n, 427a-n, 414, 425a-n, 423a-n, 421a-n, and 411.
[0076] In some aspects, one or more of the adapters 420a-n can include a LoRA. A LoRA can include trainable rank decomposition matrices. Use of LoRAs to fine-tune the base model 410 can involve freezing the base model weights (also referred to as base weights) and combining outputs of the trainable rank decomposition matrices with outputs of one or more layers of the base model 410 to then tune parameters of the LoRA. A LoRA can include two matrices, referred to as matrix A and matrix B. As shown, each adapter 420a-n includes matrix A (with dimensions d x r) , a scalar value denoted as alpha (α) , and matrix B (with dimensions r x m) . Rank (r) is proportional to a number of trainable parameters in the adapter. Therefore, a higher rank implies a greater number of trainable parameters in the adapter model, whereas lower rank implies fewer trainable parameters. In general, the low rank matrices A and B include a rank (r) having a size that is much smaller than the size of d and m (r<<d, m) . Alpha (α) is a scaling factor that can control how much influence the adapter will have on a combined output resulting from combining an output of a layer of the base model 410 (e.g., the linear layer 416 and / or the linear layer 413) with an output of the adapter. For instance, a higher alpha (α) value can result in the adapter having more influence on the combined output, while a lower alpha (α) value can result in the adapter having less influence on the combined output. Adjustment of “alpha” can balance fitting the data against overfitting. The dimension of the original weight is represented by d.
[0077] The base model 410 in FIG. 4 is shown to include two linear layers 413 and 416 for illustration purposes. In practice, the base model 410 can include many layers in addition to those shown in FIG. 4. Each adapter 420a-n includes a capability of scaling and / or adjusting parameters or weights based on training, such as parameters of the matrices A, B and alpha (α) . Various inputs and outputs of the linear layers 413 and 416 are shared between the base model 410 and adapters 420a-n. Similarly, outputs from each of the adapters 420a-n are shared with the base model 410 and ultimately combined into inputs to various linear layers.
[0078] In the example depicted in FIG. 4, base activation encoding 417 is provided as input to linear layer 416, and to adapter 420a. Linear layer 416 processes base activation encoding 417, outputting base activation encoding 415. The adapter multiplies the base activation encoding 417 with matrix A 432a, resulting in the adapter outputting adapter activation encoding 431a. Adapter activation encoding 431a is scaled by scaling block 430a, which generates adapter activation encoding 429a. The adapter then multiplies the adapter activation encoding 429a by matrix B 428a to generate and output adapter activation encoding 427a. Adapter activation encoding 429a is passed back to base model 410, where adapter activation encoding 429a is combined with (e.g., added to, concatenated with, etc. ) the base activation encoding 415, resulting in adapter activation encoding 414.
[0079] Similarly, adapter activation encoding 414 is passed both as an input to linear layer 413 and to adapter 420a. Linear layer 413 processes the input, outputting base activation encoding 412. Separately, matrix A 426a is multiplied with adapter activation encoding 414, resulting in adapter activation encoding 425a. Adapter activation encoding 425a is scaled by scaling 424a, which results in adapter activation encoding 423a. Adapter activation encoding 423a is multiplied with matrix B 422a, creating adapter activation encoding 421a. Adapter activation encoding 421a is passed back to base model 410, where adapter activation encoding 421a is added to base encoding 412, resulting in adapter activation encoding 411. Different activation encodings may be generated using the techniques described herein.
[0080] As previously described, in some cases, a fully updatable encoding can be performed. For example, when using different adapters (e.g., for different tasks) for a particular layer, the activation encodings can be different for the different adapters. While fully updateable encoding allows the activation encodings of the base model to be customized to the tasks of different adapter, the model needs to switch to the different base model activation encodings when switching from one adapter to the other. Such switching for large base machine learning models can result in large computational, memory, and latency overhead.
[0081] In some cases, adapter-only updatable encoding can be performed, where the activation encodings for a given layer can remain fixed when using different adapters for that layer. The adapter-only updatable encoding results in the machine learning model only needing to select different adapter activation encodings when switching between different adapters for the layer of the base model, in contrast to also switching between different base model activation encodings as in the fully updateable encoding described above. However, adapter-only updatable encoding may result in lower performance as compared to the fully updateable encoding.
[0082] The systems and techniques described herein can provide a middle ground between fully updateable encoding and adapter-only updatable encoding by providing a base model weight activation quantization aware adapter (e.g., LoRA) and / or a base weight quantization aware adapter (e.g., LoRA) .
[0083] FIG. 5 is a block diagram of a process 500 for providing a base model weight activation quantization aware LoRA, in accordance with aspects of the disclosure. The process 500 includes performing weight quantization 520, activation quantization 530, LoRA training / tuning 540, and a post-training quantization (PTQ) 550.
[0084] A machine learning model 510 (abase model) includes parameters, including weights and in some cases other parameters (e.g., biases, etc. ) . The machine learning model can be similar to or the same as the base model 410 of FIG. 4. The weights (and in some cases the other parameters) of the machine learning model 510 may be represented as floating point numbers or values (e.g., as 16-bit floating point numbers, float16 or FP16, 32-bit floating point numbers, float32 or FP32, or other floating-point representation) . Weight quantization 520 can be performed on the weights (and in some cases the other parameters) to convert the floating-point representations to integer numbers or values (e.g., converted to Int4, Int8, Int16, etc. ) . For example, the weight quantization 520 can quantize the parameters (including weights) of the machine learning model 510 in order to generate (e.g., determine, calculate, etc. ) quantized base model weights 525 for the machine learning model 510. Examples of quantized base model weights 525 are those associated with layer 413 and / or layer 416 of the base model 410 of FIG. 4.
[0085] The machine learning model 510 can process data using the quantized base model weights 525 to generate activation data at each layer of the machine learning model 510. Activation quantization 530 can be used to quantize the activation data output from each layer to generate one or more base activation encodings 535. Examples of base model activation encodings 535 include base activation encodings 412, 414, and 417 of FIG. 4.
[0086] After the base model activation encodings 535 are generated, the base model parameters and activations are frozen (are no longer updated or adjusted) . The LoRA can then be trained / fine-tuned conditioned on the quantized base model weights and activation encodings. For instance, as shown in FIG. 5, the quantized base model weights 525 and the quantized base model activation encodings 535 are used for LoRA training / tuning 540 to generate adapter weights 545 (e.g., weights for the matrices A and B, the scalar value alpha (α) , and / or other parameters of the LoRA) . For example, the base model 510 (with the frozen quantized base model weights 525 and the frozen quantized base model activation encodings 535) and the LoRA together can process a set of training data to generate output data. The training data can be used to fine-tune the LoRA for a specific task, such as image segmentation, answering questions specific to a particular document or subject, for generating images of a particular dog, etc. The output data can be compared to ground truth data to determine a loss. Based on the loss, backpropagation can be performed to update or fine-tune the weights of the LoRA. Such a process can be performed a number of times (e.g., until the loss is minimized) , resulting in the adapter weights 545.
[0087] PTQ 550 can then be performed on the adapter weights 545 and on activations resulting from processing data using the adapter weights 545. Examples of adapter activations include, but are not limited to, adapter activations 431a-n, 429a-n, 427a-n, etc. of FIG. 4. For instance, the adapter weights 545 and the activations are represented as floating point numbers or values (e.g., as float16, float32, etc. ) . The PTQ 550 can change the format of the adapter weights 545 and activations by converting the floating-point representations to integer numbers or values (e.g., to Int4, Int8, Int16, etc. ) . The PTQ 550 can output quantized adapter weights and quantized activation encodings 555. Once the process 500 is completed, all parameters and activations of the based machine learning model 510 and all parameters and activations of the LoRA are quantized (shown as quantized weights / activation encodings 560) , in which case a system or device deploying the machine learning model 510 and LoRA (for inference) can store the parameters and activations as integer numbers (e.g., 4-bit integer numbers in Int4 format, 8-bit integer numbers in Int8 format, 16-bit integer numbers in Int16 format, etc. ) . The machine learning model 510 and LoRA may be deployed in a resource-constrained device (e.g., an edge device) . The operations discussed with respect to process 500 may be repeated as necessary for different adapters (e.g., LoRAs) that are used with machine learning model 510 (e.g., adapters for different downstream inference tasks or functionality) .
[0088] Using the process 500, an adapter (e.g., LoRA) can be trained / fine-tuned conditioned on base model weight / activation quantization. Such a solution provides accuracy and is adapter-only updateable (due to the base model parameters and activation encodings being frozen) , which provides the quality of a fully updatable encoding and the performance benefits of adapter-only updateable encoding. Training and / or fine-tuning facilitates updates of the adapter weights, which are designed to minimize a given loss function. Accordingly, such an approach can achieve accuracy higher than traditional PTQ-only methods. As the training and / or fine-tuning is conditioned on base model weight / activation quantization, adapter-only updateable encoding is enabled, which can yield a performance benefit.
[0089] FIG. 6 is a block diagram of a process 600 for providing base model weight quantization aware LoRA, in accordance with aspects of the disclosure. The process 600 includes weight quantization 620, LoRA training / tuning 640, and PTQ 650.
[0090] A machine learning model 610 (abase model) includes parameters, including weights and in some cases other parameters (e.g., biases, etc. ) . The machine learning model can be similar to or the same as the base model 410 of FIG. 4. The weights (and in some cases the other parameters) of the machine learning model 610 may be represented as floating point numbers or values (e.g., as 16-bit floating point numbers, float16 or FP16, 32-bit floating point numbers, float32 or FP32, or other floating-point representation) . Weight quantization 620 can be performed on the weights (and in some cases the other parameters) to convert the floating-point representations to integer numbers or values (e.g., converted to Int4, Int8, Int16, etc. ) . For instance, the weight quantization 620 can quantize the parameters (including weights) of the machine learning model 610 to generate (e.g., determine, calculate, etc. ) quantized base model weights 625 for the machine learning model 610. Examples of quantized base model weights 625 are those associated with layer 413 and / or layer 416 of the base model 410 of FIG. 4.
[0091] After the weights (and in some cases the other parameters) of the machine learning model 610 are quantized, the base model parameters are frozen (are no longer updated or adjusted) . The LoRA can then be trained / fine-tuned conditioned on the quantized base model weights 625 (but not any activation encodings of the machine learning model 610) . As shown in FIG. 5, the quantized base model weights 625 are used for LoRA training / tuning 640 to generate adapter weights 645 (e.g., weights for the matrices A and B, the scalar value alpha (α) , and / or other parameters of the LoRA) . For example, the base machine learning model 610 (with the frozen quantized base model weights 625) and the LoRA can together process a set of training data to generate output data. The training data can be used to fine-tune the LoRA for a specific task, such as image classification, etc. The output data can be compared to ground truth data to determine a loss. Based on the loss, backpropagation can be performed to update or fine-tune the weights of the LoRA. Such a process can be performed a number of times (e.g., until the loss is minimized) , resulting in the adapter weights 645.
[0092] PTQ 650 can then be performed on the adapter weights 645, on activations resulting from processing data using the adapter weights 645, and on activations resulting from processing the data using the quantized base model weights 625. For instance, the adapter weights 645 and the activations of the LoRA and the base machine learning model 610 are represented as floating point numbers or values (e.g., as float16, float32, etc. ) . The PTQ 650 can change the format of the adapter weights 645 and the activations of the LoRA and the model 610 by converting the floating-point representations to integer numbers or values (e.g., to Int4, Int8, Int16, etc. ) . The PTQ 650 can output quantized adapter weights, quantized activation encodings, and quantized base model activation encodings 655. Once the process 600 is completed, all parameters and activations of the based machine learning model 610 and all parameters and activations of the LoRA are quantized (shown as quantized weights / activation encodings 660) . The operations discussed with respect to process 600 may be repeated as necessary for different adapters (e.g., LoRAs) that are used with machine learning model 510 (e.g., adapters for different downstream inference tasks or functionality) . A system or device deploying the machine learning model 510 and LoRA (for inference) can store the parameters and activations as integer numbers (e.g., 4-bit integer numbers in Int4 format, 8-bit integer numbers in Int8 format, 16-bit integer numbers in Int16 format, etc. ) .
[0093] In the example of FIG. 6, the LoRA training / tuning is aware of the base model weight quantization but is not aware of the base model activation encodings. Such an approach may be employed when accuracy loss from activation quantization is not an issue for a particular task. Because the approach discussed in FIG. 6 is unaware of the base model activation encodings, it may be easier to build the adapter training / fine-tuning pipeline. For example, mimicking a base model weight quantization by dequantizing to floating point is possible. In such an approach, the pipeline for adapter training / fine-tuning is agnostic to quantization, although the trained / fine-tuned adapter weights are aware of the quantization noise.
[0094] FIG. 7 is a flow diagram illustrating an example of a process for configuring adapters for machine learning models, in accordance with aspects of the disclosure. The process 700 (or any other described herein) can be performed by a computing device (or apparatus) or a component (e.g., one or more chipsets, a system-on-chip (SoC) , one or more processors such as one or more central processing units (CPUs) , graphics processing units (GPUs) , digital signal processors (DSPs) , neural processing units (NPUs) , neural signal processors (NSPs) , microcontrollers, ASICs, FPGAs, programmable logic devices, discrete gates or transistor logic components, discrete hardware components, etc., an ML system such as a neural network model, any combination thereof, and / or other component or system) of the computing device. The operations of the process 700 may be implemented as software components that are executed and run on one or more processors (e.g., CPU, GPU, DSP, NPU or neural engine, SoC, the processor 902 of FIG. 9, and / or other processor (s) ) .
[0095] In some aspects, at block 702, the computing system may process weights associated with one or more layers of a machine learning model to generate quantized base model weights. For instance, referring back to FIG. 4, the computing system processes weights of one or more of linear layers 413 and 416 to generate quantized weights. In an example, the quantization involves conversion from floating point to integer representation.
[0096] At block 704, the computing system may process, using the machine learning model configured with the quantized base model weights and an adapter connected to at least one layer of the one or more layers, data to generate output values.
[0097] At block 706, the computing system may modify adapter weights (of the adapter) based on the output values. In some cases, modifying the adapter weights is also based on training. Block 706 may involve training. Training involves using a loss function to determine a loss based on comparing the output values to a ground truth. Then, the weights are modifying (or “tuned” ) based on the loss (e.g., using backpropagation) .
[0098] At block 708, the computing system may process the modified adapter weights to generate quantized adapter weights. The quantized adapter weights may have lower precision than before quantization. In some cases, to process the modified adapter weights, the weights are converted from floating point format to integer format.
[0099] In some aspects, the computing system may process activations of the machine learning model to generate quantized base model activation encodings (also referred to as quantized base activation encodings) . Referring back to FIG. 4, examples of a base model activations (also referred to as base activations) that may be subject to quantization include base model activation encodings 412 and 415. The output values may be generated using the machine learning model configured with the quantized base model weights and the quantized base model activations.
[0100] At block 710, the computing system may process activations of the adapter to generate quantized adapter activations. Referring back to FIG. 4, examples of an adapter activations that may be subject to quantization include adapter activations 431a-n, 429a-n, 427a-n, and so forth. In some cases, processing the activations of the adapter may be performed using post-quantization training. In some cases, both the base model activations and the adapter activations may be modified using post-quantization training.
[0101] In some cases, the activations of the machine learning model may be processed to generate quantized base model activations after modifying the adapter weights. In some cases, the output values are generated using the machine learning model configured with the quantized base model weights and floating point base model activation values.
[0102] At block 712, the computing system may process, using the machine learning model with the quantized base model weights and the adapter with the quantized adapter weights and quantized activations, input data to generate output content.
[0103] In some aspects, the machine learning model is configured to receive and / or output multimodal content. For example, to handle multimodal input content, the computing system may generate first tokens associated with a first portion of the content based on a type of the first portion using a first encoder, generate second tokens associated with a second portion of the content based on a type of the second portion using a second encoder, and combine the first tokens and the second tokens into the input tokens. In some examples, the first portion may be text and the second portion may be an image or audio.
[0104] FIG. 8 is a block diagram of an example transformer in accordance with some aspects of the disclosure. In a convolutional neural network (CNN) model, the number of operations required to relate signals from two arbitrary input or output positions grows in the distance between positions, which makes learning dependencies at different distant positions challenging for a CNN model. The transformer 800 reduces the operations of learning dependencies by using an encoder 810 and a decoder 830 that implements an attention mechanism at different positions of a single sequence to compute a representation of that sequence. An attention function can be described as mapping a query and a set of key-value pairs to an output, where the query, keys, values, and output are all vectors. The output is computed as a weighted sum of the values, where the weight assigned to each value is computed by a compatibility function of the query with the corresponding key.
[0105] In some examples of a transformer, the encoder 810 is composed of a stack of six identical layers and each layer has two sub-layers. The first sub-layer is a multi-head self-attention engine 812, and the second sub-layer is a fully connected feed-forward network 814. A residual connection (not shown) connects around each of the sub-layers followed by normalization.
[0106] In this example of a transformer 800, the decoder 830 is also composed of a stack of six identical layers. The decoder also includes a masked multi-head self-attention engine 832, a multi-head attention engine 834 over the output of encoder 810, and a fully connected feed-forward network 826. Each layer includes a residual connection (not shown) around the layer, which is followed by layer normalization. The masked multi-head self-attention engine 832 is masked to prevent positions from attending to subsequent positions and ensures that the predictions at position i can depend only on the known outputs at positions less than i (e.g., auto-regression) .
[0107] In the transformer 800, the queries, keys, and values are linearly projected by a multi-head attention engine into learned linear projects, and then attention is performed in parallel on each of the learned linear projects, which are concatenated and then projected into final values.
[0108] The transformer also includes a positional encoder 840 to encode positions because the model does not contain recurrence and convolution and relative or absolute position of the tokens is needed. For example, the positional encodings are added to the input embeddings at the bottom layer of the encoder 810 and the decoder 830. The positional encodings are summed with the embeddings because the positional encodings and embeddings have the same dimensions. A corresponding position decoder 850 is configured to decode the positions of the embeddings for the decoder 830.
[0109] In some aspects, the transformer 800 uses self-attention mechanisms to selectively weigh the importance of different parts of an input sequence during processing and allows the model to attend to different parts of the input sequence while generating the output. The input sequence is first embedded into vectors and then passed through multiple layers of self-attention and feed-forward networks. The transformer 800 can process input sequences of variable length, making it well-suited for natural language processing tasks where input lengths can vary greatly. Additionally, the self-attention mechanism allows the transformer 800 to capture long-range dependencies between words in the input sequence, which is difficult for RNNs and CNNs. The transformer with self-attention has achieved results in several natural language processing tasks that are beyond the capabilities of other neural networks and has become a popular choice for language and text applications. For example, the various large language models, such as a generative pretrained transformer (e.g., ChatGPT, etc. ) and other current models are types of transformer networks.
[0110] FIG. 9 is a diagram illustrating an example of a system for implementing certain aspects of the present technology. In particular, FIG. 9 illustrates an example of a computing system 900, which can be for example any computing device making up internal computing system, a remote computing system, a camera, or any component thereof in which the components of the system are in communication with each other using a connection 912. The connection 912 can be a physical connection using a bus, or a direct connection into the processor 902, such as in a chipset architecture. The connection 912 can also be a virtual connection, networked connection, or logical connection.
[0111] In some aspects, the computing system 900 is a distributed system in which the functions described in this disclosure can be distributed within a datacenter, multiple data centers, a peer network, etc. In some aspects, one or more of the described system components represents many such components each performing some or all of the function for which the component is described. In some aspects, the components can be physical or virtual devices.
[0112] An example computing system 900 includes at least one processing unit (acentral processing unit (CPU) or processor) 902 and a connection 912 that couples various system components including a system memory 910, such as ROM 908 and RAM 906 to the processor 902. The computing system 900 can include a cache 904 of high-speed memory connected directly with, in close proximity to, or integrated as part of the processor 902.
[0113] The processor 902 can include any general purpose processor and a hardware service or software service, such as services 916, 918, and 920 stored in the storage device 914, configured to control the processor 902 as well as a special-purpose processor where software instructions are incorporated into the actual processor design. The processor 902 may essentially be a completely self-contained computing system, containing multiple cores or processors, a bus, memory controller, cache, etc. A multi-core processor may be symmetric or asymmetric.
[0114] To enable user interaction, the computing system 900 includes an input device 922, which can represent any number of input mechanisms, such as a microphone for speech, a touch-sensitive screen for gesture or graphical input, keyboard, mouse, motion input, speech, etc. The computing system 900 can also include an output device 924, which can be one or more of a number of output mechanisms. In some instances, multimodal systems can enable a user to provide multiple types of input / output to communicate with the computing system 900. The computing system 900 can include communications interface 926, which can generally govern and manage the user input and system output. The communication interface may perform or facilitate receipt and / or transmission wired or wireless communications using wired and / or wireless transceivers, including those making use of an audio jack / plug, a microphone jack / plug, a universal serial bus (USB) port / plug, an port / plug, an Ethernet port / plug, a fiber optic port / plug, a proprietary wired port / plug, a wireless signal transfer, a BLE wireless signal transfer, an wireless signal transfer, an RFID wireless signal transfer, near-field communications (NFC) wireless signal transfer, dedicated short range communication (DSRC) wireless signal transfer, 802.11 WiFi wireless signal transfer, WLAN signal transfer, Visible Light Communication (VLC) , Worldwide Interoperability for Microwave Access (WiMAX) , IR communication wireless signal transfer, Public Switched Telephone Network (PSTN) signal transfer, Integrated Services Digital Network (ISDN) signal transfer, 3G / 4G / 5G / LTE cellular data network wireless signal transfer, ad-hoc network signal transfer, radio wave signal transfer, microwave signal transfer, infrared signal transfer, visible light signal transfer, ultraviolet light signal transfer, wireless signal transfer along the electromagnetic spectrum, or some combination thereof. The communications interface 926 may also include one or more Global Navigation Satellite System (GNSS) receivers or transceivers that are used to determine a location of the computing system 900 based on receipt of one or more signals from one or more satellites associated with one or more GNSS systems. GNSS systems include, but are not limited to, the US-based GPS, the Russia-based Global Navigation Satellite System (GLONASS) , the China-based BeiDou Navigation Satellite System (BDS) , and the Europe-based Galileo GNSS. There is no restriction on operating on any particular hardware arrangement, and therefore the basic features here may easily be substituted for improved hardware or firmware arrangements as they are developed.
[0115] The storage device 914 can be a non-volatile and / or non-transitory and / or computer-readable memory device and can be a hard disk or other types of computer readable media which can store data that are accessible by a computer, such as magnetic cassettes, flash memory cards, solid state memory devices, digital versatile disks, cartridges, a floppy disk, a flexible disk, a hard disk, magnetic tape, a magnetic strip / stripe, any other magnetic storage medium, flash memory, memristor memory, any other solid-state memory, a compact disc read only memory (CD-ROM) optical disc, a rewritable compact disc (CD) optical disc, digital video disk (DVD) optical disc, a blu-ray disc (BDD) optical disc, a holographic optical disk, another optical medium, a secure digital (SD) card, a micro secure digital (microSD) card, a Memory card, a smartcard chip, a EMV chip, a subscriber identity module (SIM) card, a mini / micro / nano / pico SIM card, another IC chip / card, RAM, static RAM (SRAM) , dynamic RAM (DRAM) , ROM, programmable read-only memory (PROM) , erasable programmable read-only memory (EPROM) , electrically erasable programmable read-only memory (EEPROM) , flash EPROM (FLASHEPROM) , cache memory (L1 / L2 / L3 / L4 / L5 / L#) , resistive random-access memory (RRAM / ReRAM) , phase change memory (PCM) , spin transfer torque RAM (STT-RAM) , another memory chip or cartridge, and / or a combination thereof.
[0116] The storage device 914 can include software services, servers, services, etc., that when the code that defines such software is executed by the processor 902, it causes the system to perform a function. In some aspects, a hardware service that performs a particular function can include the software component stored in a computer-readable medium in connection with the necessary hardware components, such as the processor 902, the connection 912, the output device 924, etc., to carry out the function. The term “computer-readable medium” includes, but is not limited to, portable or non-portable storage devices, optical storage devices, and various other mediums capable of storing, containing, or carrying instruction (s) and / or data. A computer-readable medium may include a non-transitory medium in which data can be stored and that does not include carrier waves and / or transitory electronic signals propagating wirelessly or over wired connections. Examples of a non-transitory medium may include, but are not limited to, a magnetic disk or tape, optical storage media such as CD or DVD, flash memory, memory or memory devices. A computer-readable medium may have stored thereon code and / or machine-executable instructions that may represent a procedure, a function, a subprogram, a program, a routine, a subroutine, a module, an engine, a software package, a class, or any combination of instructions, data structures, or program statements. A code segment may be coupled to another code segment or a hardware circuit by passing and / or receiving information, data, arguments, parameters, or memory contents. Information, arguments, parameters, data, etc. may be passed, forwarded, or transmitted via any suitable means including memory sharing, message passing, token passing, network transmission, or the like.
[0117] In some examples, the processes described herein (e.g., process 700, and / or other process described herein) may be performed by a computing device or apparatus. In some examples, the process 700 can be performed by a computing device having a computing architecture of the computing system 900 shown in FIG. 9.
[0118] Specific details are provided in the description above to provide a thorough understanding of the aspects and examples provided herein, but those skilled in the art will recognize that the application is not limited thereto. Thus, while illustrative aspects of the application have been described in detail herein, it is to be understood that the inventive concepts may be otherwise variously embodied and employed, and that the appended claims are intended to be construed to include such variations, except as limited by the prior art. Various features and aspects of the above-described application may be used individually or jointly. Further, aspects may be utilized in any number of environments and applications beyond those described herein without departing from the broader scope of the specification. The specification and drawings are, accordingly, to be regarded as illustrative rather than restrictive. For the purposes of illustration, methods were described in a particular order. It should be appreciated that in alternate aspects, the methods may be performed in a different order than that described.
[0119] For clarity of explanation, in some instances the present technology may be presented as including individual functional blocks including devices, device components, steps or routines in a method embodied in software, or combinations of hardware and software. Additional components may be used other than those shown in the figures and / or described herein. For example, circuits, systems, networks, processes, and other components may be shown as components in block diagram form in order not to obscure the aspects in unnecessary detail. In other instances, well-known circuits, processes, algorithms, structures, and techniques may be shown without unnecessary detail in order to avoid obscuring the aspects.
[0120] Further, those of skill in the art will appreciate that the various illustrative logical blocks, modules, engines, circuits, and algorithm steps described in connection with the aspects disclosed herein may be implemented as electronic hardware, computer software, or combinations of both. To clearly illustrate this interchangeability of hardware and software, various illustrative components, blocks, modules, engines, circuits, and steps have been described above generally in terms of their functionality. Whether such functionality is implemented as hardware or software depends upon the particular application and design constraints imposed on the overall system. Skilled artisans may implement the described functionality in varying ways for each particular application, but such implementation decisions should not be interpreted as causing a departure from the scope of the present disclosure.
[0121] Individual aspects may be described above as a process or method which is depicted as a flowchart, a flow diagram, a data flow diagram, a structure diagram, or a block diagram. Although a flowchart may describe the operations as a sequential process, many of the operations may be performed in parallel or concurrently. In addition, the order of the operations may be re-arranged. A process is terminated when its operations are completed but could have additional steps not included in a figure. A process may correspond to a method, a function, a procedure, a subroutine, a subprogram, etc. When a process corresponds to a function, its termination may correspond to a return of the function to the calling function or the main function.
[0122] Processes and methods according to the above-described examples may be implemented using computer-executable instructions that are stored or otherwise available from computer-readable media. Such instructions may include, for example, instructions and data which cause or otherwise configure a general purpose computer, special purpose computer, or a processing device to perform a certain function or group of functions. Portions of computer resources used may be accessible over a network. The computer executable instructions may be, for example, binaries, intermediate format instructions such as assembly language, firmware, source code. Examples of computer-readable media that may be used to store instructions, information used, and / or information created during methods according to described examples include magnetic or optical disks, flash memory, USB devices provided with non-volatile memory, networked storage devices, and so on.
[0123] In some aspects the computer-readable storage devices, mediums, and memories may include a cable or wireless signal containing a bitstream and the like. However, when mentioned, non-transitory computer-readable storage media expressly exclude media such as energy, carrier signals, electromagnetic waves, and signals per se.
[0124] Those of skill in the art will appreciate that information and signals may be represented using any of a variety of different technologies and techniques. For example, data, instructions, commands, information, signals, bits, symbols, and chips that may be referenced throughout the above description may be represented by voltages, currents, electromagnetic waves, magnetic fields or particles, optical fields or particles, or any combination thereof, in some cases depending in part on the particular application, in part on the desired design, in part on the corresponding technology, etc.
[0125] The various illustrative logical blocks, modules, engines, and circuits described in connection with the aspects disclosed herein may be implemented or performed using hardware, software, firmware, middleware, microcode, hardware description languages, or any combination thereof, and may take any of a variety of form factors. When implemented in software, firmware, middleware, or microcode, the program code or code segments to perform the necessary tasks (e.g., a computer-program product) may be stored in a computer-readable or machine-readable medium. A processor (s) may perform the necessary tasks. Examples of form factors include laptops, smart phones, mobile phones, tablet devices or other small form factor personal computers, personal digital assistants, rackmount devices, standalone devices, and so on. Functionality described herein also may be embodied in peripherals or add-in cards. Such functionality may also be implemented on a circuit board among different chips or different processes executing in a single device, by way of further example.
[0126] The instructions, media for conveying such instructions, computing resources for executing them, and other structures for supporting such computing resources are example means for providing the functions described in the disclosure.
[0127] The techniques described herein may also be implemented in electronic hardware, computer software, firmware, or any combination thereof. Such techniques may be implemented in any of a variety of devices such as general purposes computers, wireless communication device handsets, or integrated circuit devices having multiple uses including application in wireless communication device handsets and other devices. Any features described as modules, engines, or components may be implemented together in an integrated logic device or separately as discrete but interoperable logic devices. If implemented in software, the techniques may be realized at least in part by a computer-readable data storage medium including program code including instructions that, when executed, performs one or more of the methods, algorithms, and / or operations described above. The computer-readable data storage medium may form part of a computer program product, which may include packaging materials. The computer-readable medium may include memory or data storage media, such as random access memory (RAM) such as synchronous dynamic random access memory (SDRAM) , read-only memory (ROM) , non-volatile random access memory (NVRAM) , electrically erasable programmable read-only memory (EEPROM) , FLASH memory, magnetic or optical data storage media, and the like. The techniques additionally, or alternatively, may be realized at least in part by a computer-readable communication medium that carries or communicates program code in the form of instructions or data structures and that may be accessed, read, and / or executed by a computer, such as propagated signals or waves.
[0128] The program code may be executed by a processor, which may include one or more processors, such as one or more digital signal processors (DSPs) , general purpose microprocessors, an application specific integrated circuits (ASICs) , field programmable logic arrays (FPGAs) , or other equivalent integrated or discrete logic circuitry. Such a processor may be configured to perform any of the techniques described in this disclosure. A general-purpose processor may be a microprocessor; but in the alternative, the processor may be any conventional processor, controller, microcontroller, or state machine. A processor may also be implemented as a combination of computing devices, e.g., a combination of a DSP and a microprocessor, a plurality of microprocessors, one or more microprocessors in conjunction with a DSP core, or any other such configuration. Accordingly, the term “processor, ” as used herein may refer to any of the foregoing structure, any combination of the foregoing structure, or any other structure or apparatus suitable for implementation of the techniques described herein.
[0129] One of ordinary skill will appreciate that the less than ( “<” ) and greater than ( “>” ) symbols or terminology used herein may be replaced with less than or equal to ( “≤” ) and greater than or equal to ( “≥” ) symbols, respectively, without departing from the scope of this description.
[0130] Where components are described as being “configured to” perform certain operations, such configuration may be accomplished, for example, by designing electronic circuits or other hardware to perform the operation, by programming programmable electronic circuits (e.g., microprocessors, or other suitable electronic circuits) to perform the operation, or any combination thereof.
[0131] The phrase “coupled to” or “communicatively coupled to” refers to any component that is physically connected to another component either directly or indirectly, and / or any component that is in communication with another component (e.g., connected to the other component over a wired or wireless connection, and / or other suitable communication interface) either directly or indirectly.
[0132] Claim language or other language reciting “at least one of” a set and / or “one or more” of a set indicates that one member of the set or multiple members of the set (in any combination) satisfy the claim. For example, claim language reciting “at least one of A and B” or “at least one of A or B” means A, B, or A and B. In another example, claim language reciting “at least one of A, B, and C” or “at least one of A, B, or C” means A, B, C, or A and B, or A and C, or B and C, A and B and C, or any duplicate information or data (e.g., A and A, B and B, C and C, A and A and B, and so on) , or any other ordering, duplication, or combination of A, B, and C. The language “at least one of” a set and / or “one or more” of a set does not limit the set to the items listed in the set. For example, claim language reciting “at least one of A and B” or “at least one of A or B” may mean A, B, or A and B, and may additionally include items not listed in the set of A and B. The phrases “at least one” and “one or more” are used interchangeably herein.
[0133] Claim language or other language reciting “at least one processor configured to, ” “at least one processor being configured to, ” “one or more processors configured to, ” “one or more processors being configured to, ” or the like indicates that one processor or multiple processors (in any combination) can perform the associated operation (s) . For example, claim language reciting “at least one processor configured to: X, Y, and Z” means a single processor can be used to perform operations X, Y, and Z; or that multiple processors are each tasked with a certain subset of operations X, Y, and Z such that together the multiple processors perform X, Y, and Z; or that a group of multiple processors work together to perform operations X, Y, and Z. In another example, claim language reciting “at least one processor configured to: X, Y, and Z” can mean that any single processor may only perform at least a subset of operations X, Y, and Z.
[0134] Where reference is made to one or more elements performing functions (e.g., steps of a method) , one element may perform all functions, or more than one element may collectively perform the functions. When more than one element collectively performs the functions, each function need not be performed by each of those elements (e.g., different functions may be performed by different elements) and / or each function need not be performed in whole by only one element (e.g., different elements may perform different sub-functions of a function) . Similarly, where reference is made to one or more elements configured to cause another element (e.g., an apparatus) to perform functions, one element may be configured to cause the other element to perform all functions, or more than one element may collectively be configured to cause the other element to perform the functions.
[0135] Where reference is made to an entity (e.g., any entity or device described herein) performing functions or being configured to perform functions (e.g., steps of a method) , the entity may be configured to cause one or more elements (individually or collectively) to perform the functions. The one or more components of the entity may include at least one memory, at least one processor, at least one communication interface, another component configured to perform one or more (or all) of the functions, and / or any combination thereof. Where reference to the entity performing functions, the entity may be configured to cause one component to perform all functions, or to cause more than one component to collectively perform the functions. When the entity is configured to cause more than one component to collectively perform the functions, each function need not be performed by each of those components (e.g., different functions may be performed by different components) and / or each function need not be performed in whole by only one component (e.g., different components may perform different sub-functions of a function) .
[0136] Illustrative aspects of the disclosure include:
[0137] Aspect 1. An apparatus for generating content, comprising: at least one memory; and at least one processor coupled to the at least one memory and configured to:process weights associated with one or more layers of a machine learning model to generate quantized base model weights; process, using the machine learning model configured with the quantized base model weights and an adapter connected to at least one layer of the one or more layers, data to generate output values; modify adapter weights based (of the adapter) on the output values; process the modified adapter weights to generate quantized adapter weights; process activations of the adapter to generate quantized adapter activations; and process, using the machine learning model with the quantized base model weights and the adapter with the quantized adapter weights and adapter quantized activations, input data to generate output content.
[0138] Aspect 2. The apparatus of Aspect 1, wherein the at least one processor is further configured to: process activations of the machine learning model to generate quantized base model activations, wherein the output values are generated using the machine learning model configured with the quantized base model weights and the quantized base model activations.
[0139] Aspect 3. The apparatus of Aspect 2, wherein the at least one processor is further configured to: process the activations of the adapter using post-quantization training.
[0140] Aspect 4. The apparatus of any of Aspects 1-3, wherein the at least one processor is further configured to: process activations of the machine learning model to generate quantized base model activations after modifying the adapter weights; wherein the output values are generated using the machine learning model configured with the quantized base model weights and floating point base model activation values.
[0141] Aspect 5. The apparatus of Aspect 4, wherein the at least one processor is further configured to: process the activations of the machine learning model and the activations of the adapter using post-quantization training.
[0142] Aspect 6. The apparatus of any of Aspects 1-5, wherein to process the modified adapter weights, the at least one processor is configured to convert the modified adapter weights from floating point format to integer format.
[0143] Aspect 7. The apparatus of any of Aspects 1-6, wherein modifying the adapter weights comprises using backpropagation based on a loss, and wherein to modify the adapter weights, the at least one processor is configured to: determine the loss from the output values and a ground truth; and update the adapter weights based on gradient derived from the loss.
[0144] Aspect 8. A method for generating content, the method comprising: processing weights associated with one or more layers of a machine learning model to generate quantized base model weights; processing, using the machine learning model configured with the quantized base model weights and an adapter connected to at least one layer of the one or more layers, data to generate output values; modifying adapter weights (of the adapter) based on the output values; processing the modified adapter weights to generate quantized adapter weights; processing activations of the adapter to generate quantized adapter activations; and processing, using the machine learning model with the quantized base model weights and the adapter with the quantized adapter weights and adapter quantized activations, input data to generate output content.
[0145] Aspect 9. The method of Aspect 8, further comprising: processing activations of the machine learning model to generate quantized base model activations, wherein the output values are generated using the machine learning model configured with the quantized base model weights and the quantized base model activations.
[0146] Aspect 10. The method of any of Aspects 8-9, further comprising processing the activations of the adapter using post-quantization training.
[0147] Aspect 11. The method of any of Aspects 8-10, further comprising: processing activations of the machine learning model to generate quantized base model activations after modifying the adapter weights, wherein the output values are generated using the machine learning model configured with the quantized base model weights and floating point base model activation values.
[0148] Aspect 12. The method of any of Aspect 11, further comprising: processing the activations of the machine learning model and the activations of the adapter using post-quantization training.
[0149] Aspect 13. The method of any of Aspects 8-12, further comprising converting the weights from floating point format to integer format.
[0150] Aspect 14. The method of any of Aspects 8-13, wherein modifying the adapter weights comprises using backpropagation based on a loss, the method further comprising: determining the loss from the output values and a ground truth; and updating the adapter weights based on gradient derived from the loss.
[0151] Aspect 15. A non-transitory computer-readable medium is provided that has stored thereon instructions that, when executed by one or more processors, cause the one or more processors to: process weights associated with one or more layers of a machine learning model to generate quantized base model weights; process, using the machine learning model configured with the quantized base model weights and an adapter connected to at least one layer of the one or more layers, data to generate output values; modify adapter weights (of the adapter) based on the output values; process the modified adapter weights to generate quantized adapter weights; process activations of the adapter to generate quantized adapter activations; and process, using the machine learning model with the quantized base model weights and the adapter with the quantized adapter weights and adapter quantized activations, input data to generate output content.
[0152] Aspect 16. The non-transitory computer-readable medium of Aspect 15, wherein when executed by the one or more processors, the one or more processors are further configured to: process activations of the machine learning model to generate quantized base model activations, wherein the output values are generated using the machine learning model configured with the quantized base model weights and the quantized base model activations.
[0153] Aspect 17. The non-transitory computer-readable medium of Aspect 16, wherein when executed by the one or more processors, the one or more processors are further configured to process the activations of the adapter using post-quantization training.
[0154] Aspect 18. The non-transitory computer-readable medium of any of Aspects 15-17, wherein when executed by the one or more processors, the one or more processors are further configured to: process activations of the machine learning model to generate quantized base model activations after modifying the adapter weights, wherein the output values are generated using the machine learning model configured with the quantized base model weights and floating point base model activation values.
[0155] Aspect 19. The non-transitory computer-readable medium of Aspects 15-18, wherein when executed by the one or more processors, the one or more processors are further configured to: process the activations of the machine learning model and the activations of the adapter using post-quantization training.
[0156] Aspect 20. The non-transitory computer-readable medium of any of Aspects 15-19, wherein processing the modified adapter weights comprises converting the modified adapter weights from floating point format to integer format.
[0157] Aspect 21. A non-transitory computer-readable medium is provided that has stored thereon instructions that, when executed by one or more processors, cause the one or more processors to perform operations according to any of Aspects 8-14.
[0158] Aspect 22. An apparatus for generating content, the apparatus comprising one or more means for performing operations according to any of Aspects 8-14.
Claims
1.An apparatus for generating content, comprising:at least one memory; andat least one processor coupled to the at least one memory and configured to:process weights associated with one or more layers of a machine learning model to generate quantized base model weights;process, using the machine learning model configured with the quantized base model weights and an adapter connected to at least one layer of the one or more layers, data to generate output values;modify adapter weights based on the output values;process the modified adapter weights to generate quantized adapter weights;process activations of the adapter to generate quantized adapter activations; andprocess, using the machine learning model with the quantized base model weights and the adapter with the quantized adapter weights and adapter quantized activations, input data to generate output content.2.The apparatus of claim 1, wherein the at least one processor is configured to:process activations of the machine learning model to generate quantized base model activations, wherein the output values are generated using the machine learning model configured with the quantized base model weights and the quantized base model activations.3.The apparatus of claim 2, wherein the at least one processor is configured to:process the activations of the adapter using post-quantization training.4.The apparatus of claim 1, wherein the at least one processor is configured to:process activations of the machine learning model to generate quantized base model activations after modifying the adapter weights;wherein the output values are generated using the machine learning model configured with the quantized base model weights and floating point base model activation values.5.The apparatus of claim 4, wherein the at least one processor is configured to:process the activations of the machine learning model and the activations of the adapter using post-quantization training.6.The apparatus of claim 1, wherein to process the modified adapter weights, the at least one processor is configured to convert the modified adapter weights from floating point format to integer format.7.The apparatus of claim 1, wherein the at least one processor is configured to modify the adapter weights using backpropagation based on a loss, and wherein to modify the adapter weights, the at least one processor is configured to:determine the loss from the output values and a ground truth; andupdate the adapter weights based on gradient derived from the loss.8.A method for generating content, the method comprising:processing weights associated with one or more layers of a machine learning model to generate quantized base model weights;processing, using the machine learning model configured with the quantized base model weights and an adapter connected to at least one layer of the one or more layers, data to generate output values;modifying adapter weights based on the output values;processing the modified adapter weights to generate quantized adapter weights;processing activations of the adapter to generate quantized adapter activations; andprocessing, using the machine learning model with the quantized base model weights and the adapter with the quantized adapter weights and adapter quantized activations, input data to generate output content.9.The method of claim 8, further comprising:processing activations of the machine learning model to generate quantized base model activations, wherein the output values are generated using the machine learning model configured with the quantized base model weights and the quantized base model activations.10.The method of claim 9, further comprising processing the activations of the adapter using post-quantization training.11.The method of claim 8, further comprising:processing activations of the machine learning model to generate quantized base model activations after modifying the adapter weights, wherein the output values are generated using the machine learning model configured with the quantized base model weights and floating point base model activation values.12.The method of claim 11, further comprising:processing the activations of the machine learning model and the activations of the adapter using post-quantization training.13.The method of claim 8, further comprising converting the weights from floating point format to integer format.14.The method of claim 8, wherein modifying the adapter weights comprises using backpropagation based on a loss, the method further comprising:determining the loss from the output values and a ground truth; andupdating the adapter weights based on gradient derived from the loss.15.A non-transitory computer-readable medium is provided that has stored thereon instructions that, when executed by one or more processors, cause the one or more processors to:process weights associated with one or more layers of a machine learning model to generate quantized base model weights;process, using the machine learning model configured with the quantized base model weights and an adapter connected to at least one layer of the one or more layers, data to generate output values;modify adapter weights the adapter based on the output values;process the modified adapter weights to generate quantized adapter weights;process activations of the adapter to generate quantized adapter activations; andprocess, using the machine learning model with the quantized base model weights and the adapter with the quantized adapter weights and adapter quantized activations, input data to generate output content.16.The non-transitory computer-readable medium of claim 15, wherein the instructions, when executed by the one or more processors, cause the one or more processors to: process activations of the machine learning model to generate quantized base model activations, wherein the output values are generated using the machine learning model configured with the quantized base model weights and the quantized base model activations.17.The non-transitory computer-readable medium of claim 16, wherein the instructions, when executed by the one or more processors, cause the one or more processors to process the activations of the adapter using post-quantization training.18.The non-transitory computer-readable medium of claim 15, wherein the instructions, when executed by the one or more processors, cause the one or more processors to:process activations of the machine learning model to generate quantized base model activations after modifying the adapter weights, wherein the output values are generated using the machine learning model configured with the quantized base model weights and floating point base model activation values.19.The non-transitory computer-readable medium of claim 17, wherein the instructions, when executed by the one or more processors, cause the one or more processors to:process the activations of the machine learning model and the activations of the adapter using post-quantization training.20.The non-transitory computer-readable medium of claim 15, wherein the instructions, when executed by the one or more processors, cause the one or more processors to process the modified adapter weights by converting the modified adapter weights from floating point format to integer format.