Abstract: The subject technology provides memory-efficient differentiable weight clustering for large language model compression. An apparatus determines a tensor including an attention map between learned weights of a trained machine learning model and corresponding centroids. The apparatus also determines a compressed attention table and a plurality of index lists during compression of the trained machine learning model based on an uniquification of the attention map and sharding of an associated index list. The apparatus determines whether the tensor exists at a destination device during compression of the trained machine learning model using a marshaling layer. The apparatus refrains from copying the tensor to the destination device when the tensor exists at the destination device, or copies the tensor to the destination device when the tensor does not exist at the destination device. The apparatus deploys a compressed machine learning model based on the compression of the trained machine learning model.

Abstract: An example process includes: receiving a speech input representing a user utterance; determining, based on a textual representation of the speech input, a first score corresponding to a type of the user utterance; determining, based on the textual representation of the speech input, a second score representing a correspondence between the user utterance and a domain recognized by a digital assistant; determining, based on the first score and the second score, whether the speech input is intended for the digital assistant; in accordance with a determination that the speech input is intended for the digital assistant: initiating, by the digital assistant, a task based on the speech input; and providing an output indicative of the initiated task.

Abstract: Systems and processes for operating a digital assistant are provided. An example method for processing an image include receiving an image, generating, based on the image, a question corresponding to a first object in the image, generating, based on the image, a caption corresponding to a second object of the image, receiving an utterance from a user, and determining a plurality of speech recognition results from the utterance based on the question and the caption.

Abstract: Systems and processes for operating a digital assistant are provided. An example method for processing an image include receiving an image, generating, based on the image, a question corresponding to a first object in the image, generating, based on the image, a caption corresponding to a second object of the image, receiving an utterance from a user, and determining a plurality of speech recognition results from the utterance based on the question and the caption.

Abstract: Systems and processes for operating a digital assistant are provided. An example process for performing a task includes, at an electronic device having one or more processors and memory, receiving a spoken input including a request, receiving an image input including a plurality of objects, selecting a reference resolution module of a plurality of reference resolution modules based on the request and the image input, determining, with the selected reference resolution module, whether the request references a first object of the plurality of objects based on at least the spoken input, and in accordance with a determination that the request references the first object of the plurality of objects, determining a response to the request including information about the first object.

Abstract: In an exemplary process, a speech input including a referenced virtual object is received. Based on the speech input, a first reference set is obtained. The first reference set is then compared to a plurality of second reference sets. Based on the comparison, a second reference set from the plurality of second reference sets is obtained. The second reference set may be identified based on a matching score between the first reference set and the second reference set. An object is then identified based on the second reference set. Based on the identified object, the reference virtual object is displayed.

Abstract: An example process includes: while displaying a portion of an extended reality (XR) environment representing a current field of view of a user: detecting a user gaze at a first object displayed in the XR environment, where the first object is persistent in the current field of view of the XR environment; in response to detecting the user gaze at the first object, expanding the first object into a list of objects including a second object representing a digital assistant; detecting a user gaze at the second object; in accordance with detecting the user gaze at the second object, displaying a first animation of the second object indicating that a digital assistant session is initiated; receiving a first audio input from the user; and displaying a second animation of the second object indicating that the digital assistant is actively listening to the user.

Abstract: An example process includes: receiving a speech input representing a user utterance; determining, based on a textual representation of the speech input, a first score corresponding to a type of the user utterance; determining, based on the textual representation of the speech input, a second score representing a correspondence between the user utterance and a domain recognized by a digital assistant; determining, based on the first score and the second score, whether the speech input is intended for the digital assistant; in accordance with a determination that the speech input is intended for the digital assistant: initiating, by the digital assistant, a task based on the speech input; and providing an output indicative of the initiated task.

Abstract: An example process includes: receiving a speech input representing a user utterance; determining, based on a textual representation of the speech input, a first score corresponding to a type of the user utterance; determining, based on the textual representation of the speech input, a second score representing a correspondence between the user utterance and a domain recognized by a digital assistant; determining, based on the first score and the second score, whether the speech input is intended for the digital assistant; in accordance with a determination that the speech input is intended for the digital assistant: initiating, by the digital assistant, a task based on the speech input; and providing an output indicative of the initiated task.

Abstract: Systems and processes for operating a digital assistant are provided. An example process for determining a response includes, at an electronic device having one or more processors and memory, receiving a spoken input including a request, performing a semantic analysis on the spoken input, determining, based on the semantic analysis, a likelihood that the electronic device requires additional contextual data to satisfy the request, and in accordance with the determined likelihood exceeding a threshold, enabling a camera of the electronic device and determining a response to the request based on data captured by the camera of the electronic device.

Abstract: Systems and processes for operating a digital assistant are provided. An example process for performing a task includes, at an electronic device having one or more processors and memory, receiving a spoken input including a request, receiving an image input including a plurality of objects, selecting a reference resolution module of a plurality of reference resolution modules based on the request and the image input, determining, with the selected reference resolution module, whether the request references a first object of the plurality of objects based on at least the spoken input, and in accordance with a determination that the request references the first object of the plurality of objects, determining a response to the request including information about the first object.

Abstract: Systems and processes for operating an intelligent automated assistant are provided. An example process includes receiving an utterance including a user request and determining whether at least a portion of the user request is ambiguous. If at least the portion of the user request is ambiguous then a set of context data based on the ambiguous portion of the user request is determined, metadata is extracted from the context data and a response to the user request is determined based on the extracted metadata.

Abstract: Configuration data for an integrated circuit may be generated using logic design equipment to implement an circuit design on the integrated circuit. Implementing the circuit design may include placing functional blocks at optimal locations that increase the maximum operating frequency of the integrated circuit implementing the optimal circuit design. Logic design equipment may perform timing analysis on an initially placed circuit design that includes initially placed functional blocks. The timing analysis may identify one or more critical paths that may be shortened by moving the critical functional blocks within the circuit design to candidate placement locations. A levelized graph representing possible candidate locations and paths between the possible candidate locations may be traversed in a breadth-first search to generate a shortest updated critical path. The critical functional blocks may be moved to candidate locations corresponding to the updated critical path.

Abstract: A method for designing a system on a target device includes placing the system on the target device. A netlist retiming is performed on the placed system. A clock allocation and a clock region optimization are performed utilizing information from the placing and the netlist retiming.

Abstract: Configuration data for an integrated circuit may be generated using logic design equipment to implement an optimal design on the integrated circuit. Implementing the optimal design may include placing hardware resources within the integrated circuit to decrease or remove overlaps between corresponding hardware resources. A given hardware resource may be defined as a rectangular region, an adjacent hardware resource may be defined as another rectangular region, and together, they may be defined as a hardware resource pair. The hardware resource pair may define an overlap region, with which a cost function may be associated. The cost function may be minimized in conjunction with other types of cost functions using a solver. The solver may generate coordinates that minimize or remove overlap to be implemented in the optimal design.

Abstract: A method for designing a system on a target device includes identifying a candidate cluster for a node in the system based on a gain value that quantifies utility for the candidate cluster. The candidate cluster is designated as a final cluster for the node when the candidate cluster has a highest gain value among other candidate clusters for each node in the candidate cluster.

Abstract: Configuration data for an integrated circuit may be generated using logic design equipment to implement an circuit design on the integrated circuit. Implementing the circuit design may include placing functional blocks at optimal locations that increase the maximum operating frequency of the integrated circuit implementing the optimal circuit design. Logic design equipment may perform timing analysis on an initially placed circuit design that includes initially placed functional blocks. The timing analysis may identify one or more critical paths that may be shortened by moving the critical functional blocks within the circuit design to candidate placement locations. A levelized graph representing possible candidate locations and paths between the possible candidate locations may be traversed in a breadth-first search to generate a shortest updated critical path. The critical functional blocks may be moved to candidate locations corresponding to the updated critical path.

Abstract: A clock-tree construction method for a configurable clock grid structure having a plurality of sectors and a plurality of wire segments includes defining a clock region within the clock grid structure and constructing an H-tree that has a smallest size to cover the clock region. The method further includes aligning the clock region within the H-tree, pruning the H-tree and removing an unused segment from the H-tree. The method further includes performing a tree height reduction procedure to the pruned H-tree, and generating a clock tree with a reduced size or a reduced height from the tree height reduction procedure.

Abstract: A method implemented on a data processing system for circuit design is described. The method comprises identifying one or more first portions of a design of a circuit, each of the one or more first portions containing a set of elements interconnected via nets, and determining whether timing of an element of a particular first portion is sensitive to degradation of a signal from a net associated with the particular first portion. In one embodiment, the method further comprises reporting the determination.

Abstract: In one aspect of an embodiment, a method implemented on a data processing system for circuit design, includes identifying one or more first portions of a design of a circuit, each of the one or more first portions containing a set of elements interconnected via timing nets and generating weights for the timing critical nets, the weights being generated after identifying the one or more first portions and executing a placer algorithm which uses the weights for the timing critical nets to place the set of elements on a representation of the design. In this method, in one embodiment, the weights for the timing critical nets can be generated to have values that differ from weights for non-critical nets. The placer algorithm can be any one of a variety of conventional placer algorithms such as a weighted wire length driven placer algorithm or a force directed timing driven placer algorithm or a min-cut placer algorithm.