Abstract: Certain aspects of the present disclosure provide techniques and apparatus for machine learning. In an example method, a memory associated with a generative machine learning model is accessed while data is processed using the generative machine learning model. The memory stores, for each token of a set of tokens for, or generated by, the generative machine learning model, a respective key tensor and a respective value tensor corresponding to the token. An average key tensor of the key tensors corresponding to the set of tokens is determined. A respective cosine similarity metric is generated for each key tensor, based on the key tensor and the average key tensor. At least one key tensor corresponding to at least one token of the set of tokens is evicted from the memory in response to determining that the respective cosine similarity metric for the at least one key tensor satisfies a condition.

Abstract: A storage device may include a storage controller and a non-volatile memory device. The storage controller may assign a first locking domain for first user data of a first namespace and second user data of a second namespace based on a first assignment request of a host device, generate a first KEK dedicated to the first locking domain, store first encrypted user data, the first encrypted user data generated by encrypting the first user data using a first media encryption key (MEK), store second encrypted user data, the second encrypted user data generated by encrypting the second user data using a second MEK, store a first encrypted MEK generated by encrypting the first MEK using the first KEK, and store a second encrypted MEK generated by encrypting the second MEK using the first KEK.

Abstract: The memory device includes a clock receiver receiving an external clock signal, a transmitter receiving first to Nth data in parallel and sequentially outputting the first to Nth data based on first to Nth clock signals including different phases, and a QEC circuit correcting a skew between the first to Nth clock signals, wherein the external clock signal includes a same frequency as the first to Nth clock signals, and the QEC circuit selectively receives the first clock signal among the first to Nth clock signals, generates the second clock signal including a phase different from a phase of the first clock signal based on a delay operation with respect to the first clock signal, and corrects the skew between the first to Nth clock signals by performing a phase comparison between the first to Nth clock signals generated based on the first and second clock signals.

Abstract: Disclosed herein are a circuit device for testing a semiconductor device and a method of operating the same. The circuit device includes: a printed circuit board (PCB); a chip socket configured to electrically connect a first semiconductor device to be tested to the PCB in a detachable manner; a second semiconductor device configured to be fixed to the PCB, such that a test code is uploaded thereto, and to apply a test signal to the first semiconductor device according to an operation of the test code and analyze an output signal received from the first semiconductor device; a communication module configured to be fixed to the PCB, to connect an external computing device and the second semiconductor device, and to selectively transmit and receive the test code and analysis results; and a power supply module configured to be fixed to the PCB, and to supply power.

Abstract: Disclosed are systems, apparatuses, processes, and computer-readable media for model training. A device may process, using a linear layer, an embedding generated from a first output token and input features to generate first features, wherein the first output token is generated by a previous iteration of a token predictor and wherein the input features are generated by a previous iteration of a decoding layer. A device may process, using the decoding layer, the first features to generate second features having first dimensions. A device may process, using a down-projection layer, the second features to generate third features having second dimensions smaller than the first dimensions. A device may generate, using the token predictor and the third features, a second output token.

Abstract: Disclosed are systems, apparatuses, processes, and computer-readable media for model training. A device may process, using a linear layer, an embedding generated from a first output token and input features to generate first features, wherein the first output token is generated by a previous iteration of a token predictor and wherein the input features are generated by a previous iteration of a decoding layer. A device may process, using the decoding layer, the first features to generate second features having first dimensions. A device may process, using a down-projection layer, the second features to generate third features having second dimensions smaller than the first dimensions. A device may generate, using the token predictor and the third features, a second output token.

Abstract: A method for repairing a steel alloy article is provided. The method includes depositing, using a first welding process, a first weld repair layer across a surface of the steel alloy article to be repaired. The first weld repair layer forms a heat-affected zone (HAZ) in the steel alloy article beneath the surface. The method also includes tempering the heat-affected zone by depositing, using a second welding process, a second weld repair layer across the first weld repair layer. The first welding technique is different than the second welding process and induces a lower heat input into the steel alloy article than the second welding process, wherein the thickness of the first weld repair layer is thinner than the thickness of the second weld repair layer.

Abstract: Certain aspects of the present disclosure provide techniques and apparatus for machine learning. In an example method, an input prompt comprising a set of tokens is accessed as input to a generative machine learning model. A first key tensor and a first value tensor are generated for a first token of the set of tokens, and the first key tensor and the first value tensor are stored in a memory. A first retention score is generated, for the first token, based on the first key tensor, the first value tensor, and a second token of the set of tokens. The first key tensor and the first value tensor are evicted from the memory in response to determining that the first retention score is a lowest retention score of the memory.

Abstract: Certain aspects of the present disclosure provide techniques and apparatus for generating a response to a query input in a generative artificial intelligence model. An example method generally includes receiving an input prompt for processing; generating a set of forecasted parameters for the input prompt using a parameter prediction model; generating, using a generative artificial intelligence model, a response to the input prompt based on the input prompt and the set of forecasted parameters; and outputting the generated response.

Abstract: Certain aspects of the present disclosure provide techniques and apparatus for efficient inferencing using a machine learning model. An example method generally includes receiving an input including a set of tokens for processing by a transformer neural network. The set of tokens for processing by the transformer neural network is partitioned into a first set of tokens and a second set of tokens. Using at least one state space model, at least one compressed token representing the first set of tokens is generated. An output token is generated, using the transformer neural network, based on the compressed token and the second set of tokens. A response to the input is generated based on the output token.

Abstract: A method and an apparatus for controlling communication parameters in multiple communication are provided. The apparatus includes a memory and a processor operatively coupled to the memory, wherein the memory includes instructions for allowing the processor to, set a first service period duration and a first wake interval, related to first communication, on the basis of the quality of service of the first communication, set a second service period duration and a second wake interval, related to second communication, on the basis of the quality of service of the second communication when a second communication connection having a frequency overlapping with that of the first communication is detected, determine whether the time difference between the first wake interval and the second wake interval occurs, and change, on the basis of the determination result, the first wake interval and the second wake interval to correspond to each other.

Abstract: Certain aspects of the present disclosure provide techniques and apparatus for machine learning. In an example method, a machine learning model comprising a plurality of layers, and a set of input data for the machine learning model, are accessed. A combination of hyperparameters for the machine learning model is selected based on the set of input data, comprising selecting, for each respective layer of the plurality of layers, a respective cache size based on the input data. The machine learning model is deployed according to the combination of hyperparameters.

Abstract: Certain aspects of the present disclosure provide techniques and apparatus for machine learning. In an example method, an input prompt comprising a set of tokens is accessed as input to a generative machine learning model. A first key tensor and a first value tensor are generated for a first token of the set of tokens, and the first key tensor and the first value tensor are stored in a memory. A first retention score is generated, for the first token, based on the first key tensor, the first value tensor, and a second token of the set of tokens. The first key tensor and the first value tensor are evicted from the memory in response to determining that the first retention score is a lowest retention score of the memory.

Abstract: A method of generating a multi-level signal having one of three or more voltage levels that are different from each other, the method including: performing a first voltage setting operation in which first and second voltage intervals are adjusted to be different from each other, wherein the first voltage interval represents a difference between a first pair of adjacent voltage levels and the second voltage interval represents a difference between a second pair of adjacent voltage levels; performing a second voltage setting operation in which a voltage swing width is adjusted, the voltage swing width representing a difference between a lowest and a highest voltage level among the three or more voltage levels; and generating an output data signal that is the multi-level signal based on input data including two or more bits, a result of the first voltage setting operation and a result of the second voltage setting operation.

Abstract: Certain aspects of the present disclosure provide techniques and apparatus for machine learning. In an example method, an input prompt comprising a set of tokens is accessed as input to a generative machine learning model. A first key tensor and a first value tensor are generated for a first token of the set of tokens, and the first key tensor and the first value tensor are stored in a memory. A first retention score is generated, for the first token, based on the first key tensor, the first value tensor, and a second token of the set of tokens. The first key tensor and the first value tensor are evicted from the memory in response to determining that the first retention score is a lowest retention score of the memory.

Abstract: A transmitter configured to receive first to N-th data in parallel and sequentially output the first to N-th data in response to first to N-th clock signals having different phases from each other, where N is an integer of at least 2, the transmitter including first to N-th data selectors including a first data selector and a second data selector in correspondence to the first to N-th data, each of the first to N-th data selectors being configured to perform a logical operation on one of the first to N-th data and the first to N-th clock signals and output a plurality of data selection signals, a first pre-driver in correspondence to at least two data selectors among the first to N-th data selectors, the first pre-driver being configured to receive the plurality of data selection signals from the at least two data selectors.

Abstract: Certain aspects of the present disclosure provide techniques and apparatus for improved machine learning. In an example method, a sequence of tokens is accessed as input to an attention operation. For a first token, an attention output is generated based on a window of tokens relative to the first token, comprising generating a first positional embedding for an influential token, generating a second positional embedding for the first token, and generating the attention output based on the first and second positional embeddings. For a second token, an attention output is generated based on a window of tokens relative to the second token, where the second window of tokens includes the first token, comprising generating a third positional embedding for the influential token, generating a fourth positional embedding for the second token, and generating the attention output based on the second, third, and fourth positional embeddings.

Abstract: A strain of Yarrowia lipolytica was engineered to overexpress ?-hydroxy ?-methylglutaryl-CoA reductase and farnesyl pyrophosphate synthase, as rate-limiting enzymes in the mevalonate and sesquiterpenoid synthesis pathways respectively. Metabolite extracts from this strain were run on LC-MS and showed a number of novel compounds being produced, including terpenoids varying in lengths and oxidation states. Upon NMR and MS/MS structure validation as well as biochemical assays, these compounds were determined as a new class of non-natural compounds, bifunctional terpenoids. Studies on the overexpression of P450 enzymes, alcohol oxidase, aldehyde dehydrogenase, and alcohol dehydrogenase showed that expression of these enzymes in addition to ?-hydroxy ?-methylglutaryl-CoA reductase and farnesyl pyrophosphate synthase increase the production of bifunctional terpenoids. Bioactivity assays demonstrate the application of bifunctional terpenoids.

Abstract: A substrate processing apparatus according to an embodiment includes an index module including a load port configured to receive a substrate carrier, and a process module connected to the index module, the process module including a transfer chamber, a process chamber, and a buffer chamber. The process chamber is connected to the transfer chamber and the buffer chamber is disposed between the index module and the transfer chamber. The transfer chamber includes a transfer robot configured to transfer a substrate between the buffer chamber and the process chamber. The buffer chamber includes a buffer frame including a plurality of buffer slots, a spray nozzle disposed on the buffer frame and configured to spray a status improvement gas into a buffer slot of the plurality, and at least one buffer side sensor disposed on the buffer frame and configured to detect a state of a substrate positioned in the buffer slot.

Abstract: A method and a system enables a temperature to which a component within a turbomachine was exposed, to be determined. The method includes operating the turbomachine including the component for an elapsed period of time, determining a thickness of an oxide layer that has formed on a coupon attached to the component, and determining a temperature offset based on the thickness measurement of the oxide layer. The method also includes determining the temperature to which the component was exposed using the temperature offset, and outputting the temperature to which the component was exposed.