Screenshot of my website circa 2011
目前，MiniMax在模型层与应用层形成了较完整的产品矩阵。基础模型包括大语言模型M1与M2系列、文生视频模型Hailuo-02、语音生成模型Speech-02；应用侧则包括集合图像与视频生成能力的“海螺AI”，以及主打互动体验的“星野/Talkie”（Talkie为星野海外版本）等。。关于这个话题，体育直播提供了深入分析
1）半导体解决方案：主要受益于 AI 收入的增长，主要受益于谷歌、meta 和字节跳动等客户对定制 ASIC 的需求。而其他非 AI 业务基本持平。，更多细节参见谷歌浏览器【最新下载地址】
回望过去十年，白酒行业处于高速扩张期，各大酒企纷纷加码产能建设，比拼窖池数量、基酒储备，竞争焦点集中于“量”的层面。然而，随着行业进入存量竞争时代，单纯的产能优势已无法支撑企业持续发展，而数智化转型则成为实现生产端“提质”的核心路径。，推荐阅读Safew下载获取更多信息
sRGB↔XYZ conversionBy Michał ‘mina86’ NazarewiczUpdated on 21st of March 2021Share on BlueskyIn an earlier post, I’ve shown how to calculate an RGB↔XYZ conversion matrix. It’s only natural to follow up with a code for converting between sRGB and XYZ colour spaces. While the matrix is a significant portion of the algorithm, there is one more step necessary: gamma correction.What is gamma correction?Human perception of light’s brightness approximates a power function of its intensity. This can be expressed as \(P = S^\alpha\) where \(P\) is the perceived brightness and \(S\) is linear intensity. \(\alpha\) has been experimentally measured to be less than one which means that people are more sensitive to changes to dark colours rather than to bright ones.Based on that observation, colour space’s encoding can be made more efficient by using higher precision when encoding dark colours and lower when encoding bright ones. This is akin to precision of floating-point numbers scaling with value’s magnitude. In RGB systems, the role of precision scaling is done by gamma correction. When colour is captured (for example from a digital camera) it goes through gamma compression which spaces dark colours apart and packs lighter colours more densely. When displaying an image, the opposite happens and encoded value goes through gamma expansion.1.00.90.80.70.60.50.40.30.20.10.0EncodedIntensityMany RGB systems use a simple \(S = E^\gamma\) expansion formula, where \(E\) is the encoded (or non-linear) value. With decoding \(\gamma\) approximating \(1/\alpha\), equal steps in encoding space correspond roughly to equal steps in perceived brightness. Image on the right demonstrates this by comparing two colour gradients. The first one has been generated by increasing encoded value in equal steps and the second one has been created by doing the same to light intensity. The former includes many dark colours while the latter contains a sudden jump in brightness from black to the next colour.sRGB uses slightly more complicated formula stitching together two functions: $$ \begin{align} E &= \begin{cases} 12.92 × S & \text{if } S ≤ S_0 \\ 1.055 × S^{1/2.4} - 0.055 & \text{otherwise} \end{cases} \\[.5em] S &= \begin{cases} {E \over 12.92} & \text{if } E ≤ E_0 \\ \left({E + 0.055 \over 1.055}\right)^{2.4} & \text{otherwise} \end{cases} \\[.5em] S_0 &= 0.00313066844250060782371 \\ E_0 &= 12.92 × S_0 \\ &= 0.04044823627710785308233 \end{align} $$The formulæ assume values are normalised to [0, 1] range. This is not always how they are expressed so a scaling step might be necessary.sRGB encodingMost common sRGB encoding uses eight bits per channel which introduces a scaling step: \(E_8 = ⌊E × 255⌉\). In an actual implementation, to increase efficiency and accuracy of gamma operations, it’s best to fuse the multiplication into aforementioned formulæ. With that arguably obvious optimisation, the equations become: $$ \begin{align} E_8 &= \begin{cases} ⌊3294.6 × S⌉ & \text{if } S ≤ S_0 \\ ⌊269.025 × S^{1/2.4} - 14.025⌉ & \text{otherwise} \end{cases} \\[.5em] S &= \begin{cases} {E_8 \over 3294.6} & \text{if } E_8 ≤ 10 \\ \left({E_8 + 14.025 \over 269.025}\right)^{2.4} & \text{otherwise} \end{cases} \\[.5em] S_0 &= 0.00313066844250060782371 \\ \end{align} $$This isn’t the only way to represent colours of course. For example, 10-bit colour depth changes the scaling factor to 1024; 16-bit high colour uses five bits for red and blue channels while five or six for green producing different scaling factors for different primaries; and HDTV caps the range to [16, 235]. Needless to say, correct formulæ need to be chosen based on the standard in question.The implementationAnd that’s it. Encoding, gamma correction and the conversion matrix are all the necessary pieces to get the conversion implemented. Like before, Rust programmers can take advantage of the srgb crate which implemented full conversion. However, to keep things interesting, in addition, here’s the conversion code written in TypeScript:type Tripple = [number, number, number];