Abstract: Embodiments described herein provide systems and techniques for training large language models. In one aspect, a process for performing in-context training of a language model is disclosed. This process may begin by receiving a language model that includes a context window of a predetermined size, as well as receiving a set of in-context prompt/completion pairs prepared for a target task. The process then constructs a first token sequence based on the set of in-context prompt/completion pairs. Next, the process fits the first token sequence into the context window. The process subsequently performs a first in-context training pass using the first token sequence to train the language model to generate a next token in accordance with the target task.

Abstract: The present disclosure discloses a quantitative dispensing device and method for biological agents. The quantitative dispensing device includes a storage container, a dispensing container, a first non-contact sensor, and a second non-contact sensor. The biological agents are stored in the storage container. A liquid inlet of the dispensing container communicates with a liquid outlet of the storage container through a delivery tube. Both the first and second non-contact sensors are arranged on the delivery tube between the liquid inlet of the dispensing container and the liquid outlet of the storage container and are separated by a preset distance. By arranging the first and second non-contact sensors on the delivery tube between the liquid inlet of the dispensing container and the liquid outlet of the storage container, non-contact quantitative dispensing of the biological agents can be achieved, effectively enhancing the dispensing efficiency and accuracy of the biological agents.

Abstract: Embodiments described herein provide systems and techniques for training large language models. In one aspect, a process for performing in-context few-shot training for a transformer-based language model is disclosed. This process may begin by receiving the transformer-based language model having a context window of a predetermined size, as well as a training dataset comprising a set of prompt/completion examples. The process then constructs a training sequence based on the training dataset. Next, the process performs a single forward pass using the training sequence as input. The process subsequently performs a set of backward passes from a subset of examples in the training sequence, wherein each backward pass is conditioned on a selected subset of prompt/completion examples in the training sequence.

Abstract: Embodiments of this application disclose a route planning method and apparatus. The method includes obtaining request failure information in response to detecting a route planning request failure event, obtaining a failure event occurrence time, and generating a request failure log based on the request failure information and the failure event occurrence time, and sending the request failure log to a cloud, to cause an application deployed in the cloud to perform signal blind area marking on a road for route planning based on the request failure log. The technical solutions can identify a road in a signal blind area, to provide a planned route with covered communication and a more stable positioning network.

Abstract: A molecular marker combination linked to quantitative traits of tea plant (+)-catechin content, including a SNP site 1, a SNP site 2, a SNP site 3, a SNP site 4, a SNP site 5, a SNP site 6, a SNP site 7 and a SNP site 8, which are located in tea genomes Scaffold4239:309117, Scaffold3614: 66549, Scaffold349: 3413816, Scaffold1989: 2316385, Scaffold451: 940283, Scaffold3727:442660, Scaffold115:803980 and Scaffold920:281727, respectively, and genotypes thereof are extremely significantly correlated with the (+)-catechin content is provided. A detection method for detecting each site, and one or more molecular marker site is used to evaluate the tea plant (+)-catechin content.

Abstract: A high-resolution display that is a small size suitable for mobile applications is disclosed. The display is color and so includes different color subpixels arranged in subpixel groups. To minimize power consumption, each subpixel includes its own power rail supplying the subpixel with a rail voltage that is based on a forward voltage of a corresponding light emitting diode. To minimize the area of the subpixel group, transistors of the subpixels are fabricated within a common well and the body terminals of the transistors are connected to a common well rail that is separate from the power rails.

Abstract: A method includes receiving a sequence of acoustic frames characterizing one or more utterances. At each of a plurality of output steps, the method also includes generating, by an encoder network of a speech recognition model, a higher order feature representation for a corresponding acoustic frame of the sequence of acoustic frames, generating, by a prediction network of the speech recognition model, a hidden representation for a corresponding sequence of non-blank symbols output by a final softmax layer of the speech recognition model, and generating, by a first joint network of the speech recognition model that receives the higher order feature representation generated by the encoder network and the dense representation generated by the prediction network, a probability distribution that the corresponding time step corresponds to a pause and an end of speech.

Abstract: An enclosure for an electronic device includes a metal substrate, a porous anodic oxide layer overlaying the metal substrate, and a hydrothermal sealant disposed on the anodic oxide layer, the hydrothermal sealant including a molten salt.

Abstract: Processing circuitry (e.g., GPU) obtains a first MSAA intermediate buffer from a rendering of a first image, the first MSAA intermediate buffer includes a plurality of subsamples for each pixel of the first image. The processing circuitry determines, for a first pixel of the first image, a first surrounding subsample pattern based on first subsamples in a surrounding region of the first pixel. Further, the processing circuitry obtains, based on the first surrounding subsample pattern, a first blending weight vector for a first location among a plurality of locations that define output pixel locations within the first pixel for upscaling the first pixel to a plurality of output pixels in an upscaled image of higher resolution than the first image. The first blending weight vector includes first blending weights respectively for first candidate subsamples associated with the first pixel. The processing circuitry generates the upscaled image accordingly.

Abstract: A backplane including bit lines and word lines to update a radiative state of a pixel is disclosed. The back plane includes a first delay line that replicates delays caused by a bit line and a second delay line that replicates delays caused by a word line. The replicated delays cause word line signals traveling along a first path to arrive at a pixel at approximately the same time as bit line signals traveling along a second path. The delays are added incrementally row-by-row and column-by-column so that each pixel can receive synchronized signals regardless of its position in the array of pixels.

Abstract: An ammonium acetate selected from the group of pentylammonium acetate, phenylethylammonium acetate, 2-([1,1?-biphenyl]-4-yl) ethan-1-amine acetate, butanammonium acetate, hexylammonium acetate, octylammonium acetate, phenylbutanammonium acetate, and any combination thereof, is used to modify a perovskite layer in a perovskite solar cell. A perovskite solar cell includes a perovskite layer and an interface modification layer that is in contact with the perovskite layer and includes at least one ammonium acetate provided herein. A method of preparing the inverted perovskite solar cell.

Abstract: This application discloses example device-to-device (D2D) transmission methods and communication apparatuses. One example method includes receiving, by a station, a trigger frame sent by an access point, where the trigger frame includes a user information field, the user information field includes an identifier of the station, resource allocation information, first indication information, and second indication information, the first indication information indicates that a resource unit indicated by the resource allocation information is used for D2D transmission, and the second indication information indicates that the station is a sender or a receiver of the D2D transmission. The D2D transmission is performed by the station on the resource unit indicated by the resource allocation information.

Abstract: A method for directly preparing a trivalent chromium compound by electrochemical oxidation of ferrochrome is provided. The method includes: putting ferrochrome as an anode, and placing the anode into an electrolyte solution containing a complexing agent together with a cathode, then turning on a power supply for electrolysis reaction so that chromium and iron in ferrochrome are directly converted into free Cr3+ and Fe3+ respectively, allowing one of Cr3+ and fe3+ to form a stable soluble metal complex together with the complexing agent, and allowing the other of Cr3+ and Fe3+ to form a metal hydroxide solid together with OH? generated by electrolysis reaction, so as to obtain an electrolysis completion slurry. Compared with the prior art, the present application has no hexavalent chromium salt stage, thereby shortening the process flow and avoiding the generation of chromium-containing waste residue.

Abstract: An azimuth ambiguity suppression method for a spaceborne SAR in spatially non-uniform sampling relates to the technical field of radars. The method includes the following steps: step 1: compensating for a phase difference between the channels according to the locations of transmitting and receiving stations for channel echoes; step 2: calculating an equivalent azimuth sampling location of each channel at the moment of each pulse transmission; step 3: sorting the equivalent azimuth sampling locations of all channels and their corresponding echoes after the phase difference between the channels is compensated in ascending order; and step 4: calculating reconstructed echoes from the sorted equivalent azimuth sampling locations and corresponding echoes after the phase difference between the channels is compensated, by using an adaptive weight-conjugate gradient-Toeplitz matrix (ACT) algorithm.

Abstract: A square bar grinding device includes a loading rack, a clamping device, a grinding device, and a grinding tool grinding device. The loading rack is provided with stations, the stations include a first station and a second station, the first station is used for placing a to-be-machined square bar, the grinding tool grinding device includes an oil stone assembly, the oil stone assembly is disposed at the second station, and the clamping device is used for clamping and transferring the to-be-machined square bar or the oil stone assembly in the stations of the loading rack to the grinding device. According to the square bar grinding device, a grinding tool grinding device is disposed at a loading rack, and can be clamped and transferred by a clamping device to be quickly moved to a grinding device for grinding of a grinding tool.

Abstract: A word line regulator that can control an SRAM cell during a write process is disclosed. The SRAM cell is coupled to bit lines that are driven using a reduced voltage in order to conserve power. The SRAM cell includes a latch that is powered by a power supply in a power domain that is different from the bit lines. The word line regulator is configured to output a voltage that is based on the different power domains to ensure the proper operation of transistors in the SRAM cell during the write process.

Abstract: Embodiments of the disclosure provide a method, a device, an apparatus, and a storage medium for video production. The method includes: presenting a text setup interface in a displayed video producing window; receiving a first trigger operation; displaying object attribute information corresponding to the video production object in the first information display area; receiving a second trigger operation; and displaying, in the second information display area, video related text corresponding to the video production object. According to the method, a simple and easy operation video production platform is provided for a video producer. A text setting interface of a video to be manufactured is presented on the video production platform, and a function item for video text generation is provided for the video producer.

Abstract: The present disclosure relates to an array substrate, a display panel, and an electronic device. The array substrate includes a plurality of sub-pixel regions, a gate line layer, and a common electrode layer. Each sub-pixel region of the sub-pixel regions has two sub-pixel units disposed in a same row and two thin film transistors connected to the two sub-pixel units respectively. The gate line layer has gate lines and gate electrodes. Control electrodes of two thin film transistors of each sub-pixel region are respectively connected to two gate lines located on two ends of the sub-pixel region through corresponding gate electrodes.

Abstract: An automatic support device for resisting rock burst in mines is provided, which includes bases symmetrically distributed, sliding frames are slidably connected to the bases, hydraulic rods are fixedly connected to the bases, and telescopic ends of the hydraulic rods are fixedly connected to the adjacent sliding frames. The sliding frames are fixedly connected to brackets, and the brackets symmetrically distributed are provided with a main support plate, on which pushing rods are provided. The device can slow down the impact force of the device when rock burst occurs in the mine, at the same time, it plays a supporting role in the mine, ensuring the stability of the roof of the mine.

Abstract: The present disclosure discloses a method for suppressing the azimuth ambiguity of multi-channel SAR systems based on channel cancellation, which falls within the field of radar signal processing. The method includes: imaging echoed of the multi-channel SAR system based on a linear mapping reconstruction algorithm to obtain a reconstructed high-resolution wide-swath image; estimating positions of the azimuth ambiguity regions of the reconstructed high-resolution wide-swath image; extracting ambiguity images in the azimuth ambiguity regions based on channel cancellation and the refocusing algorithm; and suppressing a repetitive ambiguity image in imaging the echo of the multi-channel SAR system.