Coding agents often retrieve code from an entire repository, but only limited evidence can fit into the final model input. Conventional retrieval-augmented generation (RAG) for coding agents treats fragments from the same code object as separate results, so redundant views can occupy multiple context positions and crowd out useful code. Grouping fragments by code object reduces this redundancy, but can discard local information needed for the task. We describe this tension as an invariance race: allocation should stay stable under redundant renderings but change when a fragment adds task-relevant semantics. To address this race, we introduce VITAL-RAG, which organizes evidence by canonical code object, keeps one query-relevant companion only when it adds semantics not already represented, and renders selected evidence under per-object and global token budgets. On RepoBench, VITALRAG improves Recall@4K from 39.59

Radio frequency (RF) is an emerging technology for rapeseed pretreatment, offering a comparison to the established microwave (MW) technique. This study investigated the effects of RF and MW pretreatment combined with different oil extraction methods on the oil yield, quality characteristics and lipid concomitant contents of rapeseed and its products. Results indicated that RF combined with pressing extraction yielded the highest tocopherol and canolol contents in rapeseed oil (839.6 and 1316.4 mg/kg, 8.0 % and 7.9 times higher than the control, respectively), and MW combined with supercritical carbon dioxide fluid extraction yielded the highest phytosterol content (8402.0 mg/kg, 16.6 % higher than the control). These results indicate the effectiveness of RF as a novel pretreatment method for rapeseed and its potentially greater advantage than MW. Results also imply that RF could contribute to sustainable and efficient oil extraction processes in the future food industry owing to its high efficiency and energy-saving capability.

In this study, we applied various thermal pretreatment methods (e.g., hot-air, microwave, and stir-frying) to process walnut kernels, and conducted comparative analysis of the physicochemical properties, nutritional components, in vitro antioxidant activity, and flavor substances of the extracted walnut oil (WO). The results indicated that, thermal pretreatment significantly increased the extraction of total trace nutrients (e.g., total phenols, tocopherols, and phytosterols) in WO. The WO produced using microwave had 2316.71 mg/kg of total trace nutrients, closely followed by the stir-frying method, which yielded an 11.22% increase compared to the untreated method. The WO obtained by the microwave method had a higher Oxidative inductance period (4.05 h) and oil yield (2.48%). After analyzing the flavor in WO, we found that aldehydes accounted for 28.77% of the 73 of volatile compounds and 58.12% of the total flavor compound content in microwave-pretreated WO, these percentages were higher than those recorded by using other methods. Based on the comprehensive score obtained by the PCA, microwave-pretreatment might be a promising strategy to improve the quality of WO based on aromatic characteristics.

We present the latest generation of MobileNets, known as MobileNetV4 (MNv4), featuring universally efficient architecture designs for mobile devices. At its core, we introduce the Universal Inverted Bottleneck (UIB) search block, a unified and flexible structure that merges Inverted Bottleneck (IB), ConvNext, Feed Forward Network (FFN), and a novel Extra Depthwise (ExtraDW) variant. Alongside UIB, we present Mobile MQA, an attention block tailored for mobile accelerators, delivering a significant 39% speedup. An optimized neural architecture search (NAS) recipe was also crafted to improve MNv4 search effectiveness. The integration of UIB, Mobile MQA and the refined NAS recipe results in a new suite of MNv4 models that are mostly Pareto optimal across mobile CPUs, DSPs, GPUs, as well as specialized accelerators like Apple Neural Engine and Google Pixel EdgeTPU - a characteristic not found in any other models tested. Our approach emphasizes simplicity, utilizing standard components and a straightforward attention mechanism to ensure broad hardware compatibility. To further boost efficiency, we finally introduce a novel distillation technique. Enhanced by this technique, our MNv4-Hybrid-Large model delivers impressive 87% ImageNet-1K accuracy, with a Pixel 8 EdgeTPU runtime of just 3.8ms.

Walnut oils were obtained by supercritical carbon dioxide extraction (SCB), cold-pressing (CP), hexane extraction (HE), and subcritical butane extraction (SBE), and walnut protein isolates (WPI) from the walnut cakes were performed. The results indicate that SCB has the highest oil yield for walnut oil, which was 62.72%, and the total content of trace nutrients (total tocopherols, total phytosterols, and total phenolic compounds) in SCB-walnut oil was also the highest at 2186.75 mg/kg, approximately 1.05 times higher than CP-walnut oil and 1.21 times higher than SBE-walnut oil. Meanwhile, the treatment of WPI with SCB results in a decrease in β-Sheet and α-Helix structures and an increase in β-Turn and Random coil structures. Thereby increasing its oil-holding capacity (OHC) and solubility by approximately 1.16 times and 1.27 times compared to CP, respectively. Interestingly, SCB as a green oil production technology, also has good prospects for retaining WPI functionality characteristics.

Magnesium silicate and basic magnesium carbonate were characterized and their application in the low temperature adsorption refining of fragrant rapeseed oil was investigated. The results showed that both materials had a loose porous structure, but their particle size distribution, active functional groups, crystallinity, specific surface area, pore volume, and pore size differed. Under the optimal adsorption refining conditions, magnesium silicate and basic magnesium carbonate reduced the phospholipid content from 4.18 to 0.83 and 1.57 mg/g, acid value from 1.67 to 0.88 and 0.90 mgKOH/g, yellow value from 38.2 to 32.1 and 36.2, and red value from 2.9 to 1.7 and 2.2, respectively. The use of these two magnesium salts did not significantly alter the composition of fatty acids and the content of tocopherol. The retention rates of total phytosterol by magnesium silicate and basic magnesium carbonate were 93.6% and 95.9%, respectively. The retention rates of total phenol were 88.9% and 93.9%, respectively. The adsorption refining process proved to be advantageous in purifying fragrant rapeseed oil while maintaining the dominant flavor components relatively unchanged (P < 0.05). As a result, these two magnesium salts have the potential to be used during the low temperature adsorption refining process of fragrant rapeseed oil.

This paper presents a new mechanism to facilitate the training of mask transformers for efficient panoptic segmentation, democratizing its deployment. We observe that due to the high complexity in the training objective of panoptic segmentation, it will inevitably lead to much higher penalization on false positive. Such unbalanced loss makes the training process of the end-to-end mask-transformer based architectures difficult, especially for efficient models. In this paper, we present ReMaX that adds relaxation to mask predictions and class predictions during the training phase for panoptic segmentation. We demonstrate that via these simple relaxation techniques during training, our model can be consistently improved by a clear margin without any extra computational cost on inference. By combining our method with efficient backbones like MobileNetV3-Small, our method achieves new state-of-the-art results for efficient panoptic segmentation on COCO, ADE20K and Cityscapes. Code and pre-trained checkpoints will be available at https://github.com/google-research/deeplab2.

To overcome the issues in the traditional deacidification processes of peony seed oil (PSO), such as losses of neutral oil and trace nutrients, waste discharge, and high energy consumption, adsorption deacidification was developed. The acid removal capacity of adsorbent-alkali microcrystalline cellulose was evaluated using the isothermal adsorption equilibrium and the pseudo-first-order rate equation. The optimized adsorption deacidification conditions included adsorbent-alkali microcrystalline cellulose at 3%, a heating temperature of 50 °C, and a holding time of 60 min. The physicochemical, bioactive properties, antioxidant capacities, and oxidative stabilities of PSO processed by alkali refining and oil-hexane miscella deacidification were compared under the same operating conditions. Fatty acid content was not significantly different across all three methods. The deacidification rates were 88.29%, 98.11%, and 97.76%, respectively, for adsorption deacidification, alkali refining, and oil-hexane miscella deacidification. Among the three deacidification samples, adsorption deacidification showed the highest retention of tocopherols (92.66%), phytosterols (91.96%), and polyphenols (70.64%). Additionally, the obtained extract preserved about 67.32% of the total antioxidant activity. The oil stability index was increased 1.35 times by adsorption deacidification. Overall, adsorption deacidification can be considered a promising extraction technology in terms of quality as compared to alkali refining and oil-hexane miscella deacidification.

This study analyzed and evaluated the basic crude fat contents, crude protein contents, phenolic compounds, lipid compositions (fatty acids, phytosterols, and tocopherols), and amino acid compositions of 26 walnut samples from 11 walnut-growing provinces in China. The results indicate that the oil contents of the samples varied from 60.08% to 71.06%, and their protein contents ranged from 7.26 g/100 g to 19.50 g/100 g. The composition of fatty acids corresponded to palmitic acid at 4.61–8.27%, stearic acid at 1.90–3.55%, oleic acid at 15.50–32.28%, linoleic acid at 53.44–67.64%, and α-linolenic acid at 2.45–12.77%. The samples provided micronutrients in widely varying amounts, including tocopherol, phytosterol, and total phenolic content, which were found in the walnut oil samples in amounts ranging from 356.49 to 930.43 mg/kg, from 1248.61 to 2155.24 mg/kg, and from 15.85 to 68.51 mg/kg, respectively. A comprehensive evaluation of walnut oil quality in the samples from the 11 provinces using a principal component analysis was conducted. The findings revealed that the samples from Henan, Gansu, and Zhejiang had the highest composite scores among all provinces. Overall, Yunnan-produced walnuts had high levels of crude fat, polyunsaturated fatty acids, and total tocopherols, making them more suitable for producing high-quality oil, whereas Henan-produced walnuts, although lower in crude fat, had a higher crude protein content and composite score, thus showing the best walnut characteristics.

为了探明油菜籽微波预处理过程中的水分变化情况,建立油菜籽微波预处理干燥模型,对油菜籽在料层厚度12 mm、不同微波功率(1.0、1.5、2.0 kW)以及微波功率1.5 kW、不同料层厚度(6、12、18 mm)下预处理后的含水率、水分比和失水速率的变化情况进行了研究,并以常用的3种干燥模型指数模型、单项扩散模型和Page模型进行了数据拟合.结果表明:微波功率越高、料层越薄,油菜籽水分流失越快,微波预处理时间越短;微波预处理过程中油菜籽水分变化情况与Page模型拟合度最好.

本文以甘蓝型、白菜型和芥菜型三种油菜籽为原料,探究了油菜籽经过膨爆预处理后菜籽油的品质及挥发性风味成分的变化.研究结果表明,膨爆预处理后甘蓝型、白菜型和芥菜型三种菜籽油的酸值和过氧化值均显著增加(P<0.05),分别增加了0.25、0.49、0.39 mg KOH/g和0.20、0.18、0.18 mmol/kg,均在国际食品法典规定的范围内;三种菜籽油的氧化诱导时间均显著增加(P<0.05),分别增加了17.32、10.06和13.99 h.膨爆预处理对三种菜籽油的脂肪酸组成无显著影响.膨爆预处理后,甘蓝型、白菜型和芥菜型三种菜籽油中苯乙腈、3-甲基巴豆腈、5-己烯腈等含硫化合物含量、杂环类物质以及氧化挥发物的种类和含量均显著增加(P<0.05),杂环类物质和氧化挥发物中分别是以吡嗪类和醛类物质为主.

以微波-压榨富含酚酸菜籽油为研究对象,考察了钠基膨润土的吸附脱磷效果以及处理过程中菜籽油酚酸和其他品质的变化.首先考察了吸附温度、钠基膨润土添加量和吸附时间对脱磷率和菜籽油中酚酸的影响,然后通过正交实验,以脱磷率、总酚保留率和canolol(2,6-二甲氧基-4-2烯基苯酚)保留率的综合加权得分为考察指标,得出钠基膨润土的最优脱磷条件参数为:吸附温度40℃、钠基膨润土添加量0.75％、时间30 min,在此条件下,脱磷率为82.5％,总酚和canolol保留率分别为97.2％和93.2％.同时,钠基膨润土吸附脱磷对菜籽油的脂肪酸组成无显著影响,也可以很好地保留住菜籽油中的生育酚和植物甾醇,总生育酚和总植物甾醇的保留率分别为98.2％和94.2％.

A new method for rapeseed oil preparation named synchronous pressing and refining after solid-phase pre -adsorption technology was investigated. Rapeseed and adsorbents were first mixed evenly to form premixes, and the product oils were then obtained by direct pressing and filtering. The phospholipid content, acid value, a* value, chlorophyll, carotene, polyphenol contents of crude oil increased, while L* and b* values decreased with increasing squeezing chamber temperature. A moderate adsorbent dosage and temperature could improve the refining effects. Silicon dioxide had the highest dephosphorization and deacidification rates of 95% and 42%, respectively, when the dosage was 30 g/kg at 130 degrees C. Activated clay decreased the chlorophyll and carotene contents by 75% and 38%, respectively. Sucrose fatty acid ester had dephosphorization and deacidification rates of 67s% and 41%, respectively. Ascorbic palmitate had a dephosphorization rate of 74% and retained poly-phenols intactly. This technology would not affect the oil yield and could decrease the peroxide value of rapeseed oil. It realized the simultaneous completion of pressing and refining with the advantages of a short preparation time and environmental sustainability.

The objectives of this study were to obtain the optimal technical parameters of steam explosion for rapeseed and investigate the effects of pretreatment on the bioactive components and characteristics of three diverse samples of Brassica seed (Brassica napus, B. juncea and B. rapa) and their products. Under the optimal conditions, Brassica napus formed the largest amount of canolol, namely, 1210.10 mg/kg; B. rapa formed the smallest amount, namely, 82.70 mg/kg; and the canolol content was 2110.00 mg/kg in Brassica napus oil. Meanwhile, compared to the traditional cold-pressed rapeseed oil, the total tocopherol and phytosterol contents in Brassica napus, B. rapa and B. juncea oil increased by 5.3%, 4.8%, and 7.4% and 2.1%, 3.2%, and 5.1%, respectively. Steam explosion pretreatment also significantly affected the total phenolic contents and antioxidant capacities of the three types of rapeseeds and their processed products (P < 0.05). In addition, by steam explosion pretreatment, the EAA contents of B. rapa and B. juncea rapeseed cakes increased by 2.22% and 4.49%, respectively. These findings will provide data support and a reference basis for the application and development of steam explosion technology to rapeseeds of various types.

In this study, we explored the technical parameters of tree peony seeds oil (TPSO) after their treatment with radio frequency (RF) at 0 °C–140 °C, and compared the results with microwave (MW) and roasted (RT) pretreatment in terms of their physicochemical properties, bioactivity (fatty acid tocopherols and phytosterols), volatile compounds and antioxidant activity of TPSO. RF (140 °C) pretreatment can effectively destroy the cell structure, substantially increasing oil yield by 15.23%. Tocopherols and phytosterols were enhanced in oil to 51.45 mg/kg and 341.35 mg/kg, respectively. In addition, antioxidant activities for 2,2-diphenyl-1-picrylhydrazyl (DPPH) and ferric-reducing antioxidant power (FRAP) were significantly improved by 33.26 μmol TE/100 g and 65.84 μmol TE/100 g, respectively (p < 0.05). The induction period (IP) value increased by 4.04 times. These results are similar to those of the MW pretreatment. The contents of aromatic compounds were significantly increased, resulting in improved flavors and aromas (roasted, nutty), by RF, MW and RT pretreatments. The three pretreatments significantly enhanced the antioxidant capacities and oxidative stabilities (p < 0.05). The current findings reveal RF to be a potential pretreatment for application in the industrial production of TPSO.

We present a next-generation neural network architecture, MOSAIC, for efficient and accurate semantic image segmentation on mobile devices. MOSAIC is designed using commonly supported neural operations by diverse mobile hardware platforms for flexible deployment across various mobile platforms. With a simple asymmetric encoder-decoder structure which consists of an efficient multi-scale context encoder and a light-weight hybrid decoder to recover spatial details from aggregated information, MOSAIC achieves new state-of-the-art performance while balancing accuracy and computational cost. Deployed on top of a tailored feature extraction backbone based on a searched classification network, MOSAIC achieves a 5% absolute accuracy gain surpassing the current industry standard MLPerf models and state-of-the-art architectures.

Implementation details. Following HAQ [8], we quantize all the layers, in which the first and the last layers are quantized to 8-bit. Following [4, 2], we introduce weight normalization during training. We use SGD with nesterov [6] for optimization, with a momentum of 0.9. For all models on ImageNet, we first train the full-precision models and then use the pre-trained weights to initialize the quantized models. We then fine-tune for 150 epochs. The learning rate starts at 0.01 and decays with cosine annealing [5]. Main Results. We apply the proposed method to quantize MobileNetV1 [3] and MobileNetV2 [7] to 4-bit. We compare the performance of different methods in Table S1. From the results, our proposed method outperforms other methods by a large margin. For example, compared with HAQ, our proposed method achieve 2.7% and 3.5% higher Top-1 accuracy for 4-bit MobileNetV1 and MobileNetV2.

为了精炼后的油中尽可能多地保留菜籽多酚,首先通过傅里叶红外光谱分析(Fourier infrared spectrum,FT-IR)和差示扫描量热分析(differential scanning calorimetry,DSC)对二氧化硅(SiO2)和磷脂酰乙醇胺(PE)以及它们的复合物进行表征分析,然后以菜籽油为研究对象,分析了它在SiO2吸附脱磷处理中酚酸的变化规律,同时考察了吸附脱磷效果,并通过正交优化实验获得了高脱磷率和高酚酸保留率的工艺参数条件.研究结果表明,SiO2与PE的复合可能是一种弱的相互作用;SiO2对菜籽油中的磷脂有较好的吸附脱除作用,在脱磷温度为35oC、SiO2添加量为0.75％、脱磷时间为15 min的条件下,脱磷率可达86.7％,同时菜籽油中的总酚和Canolol(2,6-二甲氧基-4-乙烯基苯酚)保留率分别高达99.5％和98.7％.

本研究对油菜籽进行了干法炒籽预处理,然后压榨制油,并对油菜籽、菜籽油、菜籽饼、脱脂菜籽粕的总酚含量、酚酸含量和菜籽油的营养指标进行了测定.研究结果表明,随着炒籽温度的升高,菜籽油中的总酚含量增加,在160℃时增加了 27.4倍,芥子酸、芥子碱和Canolol含量先增加后减少,最高分别在120、150、140℃时增加了 80.4％、6.7倍和191.4倍;油菜籽中的总酚含量增加,在160℃时增加了 23.7％,芥子酸和芥子碱含量减少,在160℃时分别下降了 90.5％和26.5％,Canolol含量先增加后减少,在130℃增加了 55.4倍.炒籽预处理使菜籽油β-生育酚含量增加了 52.7％,菜籽甾醇、菜油甾醇和谷甾醇含量分别增加了29.2％、17.7％和18.7％,而对脂肪酸组成无显著影响(P＜0.05).炒籽过程中Canolol的形成与油菜籽的初始芥子酸、芥子碱的含量以及炒籽过程中芥子酸、芥子碱的减少量高度相关,且可能有芥子碱也转化为芥子酸并热脱羧形成了 Canolol.

Hardware-aware neural architecture designs have been predominantly focusing on optimizing model performance on single hardware and model development complexity, where another important factor, model deployment complexity, has been largely ignored. In this paper, we argue that, for applications that may be deployed on multiple hardware, having different single-hardware models across the deployed hardware makes it hard to guarantee consistent outputs across hardware and duplicates engineering work for debugging and fixing. To minimize such deployment cost, we propose an alternative solution, multi-hardware models, where a single architecture is developed for multiple hardware. With thoughtful search space design and incorporating the proposed multi-hardware metrics in neural architecture search, we discover multi-hardware models that give state-of-the-art (SoTA) performance across multiple hardware in both average and worse case scenarios. For performance on individual hardware, the single multi-hardware model yields similar or better results than SoTA performance on accelerators like GPU, DSP and EdgeTPU which was achieved by different models, while having similar performance with MobilenetV3 Large Minimalistic model on mobile CPU. 1