Пятнадцать байтов на стек от конца

https://www.automotiveworld.com/articles/software-centric-chips-anticipate-the-ai-defined-vehicle/

"SiMa.ai is positioning its unique MLSoC platform as the gateway for unlocking next-generation ADAS/AV performance. Importantly, it has been conceived specifically for automotive applications."

"SiMa.ai’s MLSoC has an ML performance of 50 TOPS and a ResNet-50 (an image classifying architecture) performance of more than 300 FPS per watt. The company claims latency can be reduced by a factor of ten, with all compute in Level 2/3 systems achieved at less than 25W, or less than 100W at Level 4."

"SiMa.ai designed the chip according to three foundation pillars: to be compatible with any computer vision and generative AI application using ML at the edge, regardless of vehicle type; to offer best-in-class performance per watt; and to provide an interface usable by anyone instead of a small demographic of hardware experts."

"In December 2024, SiMa.ai combined its AI/ML capabilities with silicon design and verification firm Synopsys to maximise customisation for automotive-centric IP, subsystems, chiplets, and SoCs."

Почему бы не подурачиться 1 апреля. Думаю, парсер грамматики Си вполне сойдёт за дурацкую затею. Тем более я ж не сам напрягаться собираюсь, а ИИ подряжу. Я упоминал про ASDL, а есть ещё Ungrammar, другой способ описывать синтаксические деревья. Я попросил Грок сбацать мне грамматику языка Си. Он предложил упрощённый вариант:

// Top-level program structure
TranslationUnit = Declaration*

// Declarations
Declaration = Type Identifier ';'
            | Type Identifier '(' ParameterList ')' CompoundStmt
ParameterList = (Type Identifier (',' Type Identifier)*)?

// Types
Type = 'int'
     | 'float'
     | 'char'
     | 'void'

// Statements
Stmt = CompoundStmt
     | ExprStmt
     | IfStmt
     | WhileStmt
     | ReturnStmt

CompoundStmt = '{' Declaration* Stmt* '}'
ExprStmt = Expr ';'
IfStmt = 'if' '(' Expr ')' Stmt ('else' Stmt)?
WhileStmt = 'while' '(' Expr ')' Stmt
ReturnStmt = 'return' Expr? ';'

// Expressions
Expr = AssignmentExpr
AssignmentExpr = EqualityExpr ('=' EqualityExpr)?
EqualityExpr = RelationalExpr (('==' | '!=') RelationalExpr)?
RelationalExpr = AddExpr (('<' | '>' | '<=' | '>=') AddExpr)?
AddExpr = MultExpr (('+' | '-') MultExpr)*
MultExpr = UnaryExpr (('*' | '/') UnaryExpr)*
UnaryExpr = PrimaryExpr
          | '-' UnaryExpr
          | '!' UnaryExpr
PrimaryExpr = Identifier
            | Number
            | String
            | '(' Expr ')'

// Basic tokens
Identifier = <token>
Number = <token>
String = <token>

Неплохо, вполне годится. А теперь говорю, напиши парсер этой грамматики на Rust. С третьей попытки получился работающий код: c_parser/src/main.rs. Полтыщи строк нетривиального кода одним махом побиваши.

В качестве теста обработаем Си-шную программу:

int x = 42;
int main(int argc) {
    int y = 3 + 5 * 2;
    if (y > x) {
        return 1;
    } else {
        return 0;
    }
}

Компилируем, запускаем:

$ cargo run
...
     Running `/home/vak/c_parser/target/debug/c_parser`
( результат )

Остаётся добавить указатели, массивы, структуры, тайпдефы и прочие фичи, и язык Си у нас в кармане. 😀

Программировать с помощью ИИ становится прикольно. Зашёл на x.ai и попросил написать простой текстовый редактор типа emacs. Вполне неплохой код получился. Создаём файл src/main.rs:
Компилируем, запускаем - работает!

cargo run

Это я туда некий текст скопипастил. Выход по ^C. С таким помощником удобно язык изучать.

In this video we demonstrate the ability of Modalix to run a large multimodal model in order to perform scene analysis. A text prompt and image are both passed to Modalix and converted into a sequence of tokens. The image itself is converted into approximately 600 tokens. And the combined text and image tokens are passed to the LLaVA-7B large multimodal model that is running on Modalix. Modalix accelerates the Time-To-First-Token by processing these tokens in batch at a rate of 240 tokens per second. And it currently generates tokens at a rate of 6.8 tokens per second. The generated tokens are converted into words and stream back to the host, where they are converted back to speech.

(pressing button "Analyze Scene")
(Modalix speaks)

"The image features a large airplane flying low over a city with its wings visible. The airplane is positioned above a highway, and there are several cars driving on the road below. The scene also includes a tall building with a large "A" on its side, possibly representing a company or landmark. The overall atmosphere suggests a bustling urban environment with air traffic and busy streets."

(this concludes the scene analysis demo)



Подробности читайте в компанейском блоге: sima.ai/implementing-multimodal-genai-models-on-modalix/

Они придумали как ускорить тренировку трансформеров на порядок. То есть значительно удешевить. Вот статья, где это дело объясняется.

"Deepdive into Deepseek"

Вы уже слышали вероятно про китайскую речевую модель DeepSeek, что наделала нынче шороху. Ну и я, конечно, захотел задать ему какой-нибудь характерный вопрос. Зашёл на chat.deepseek.com и с порога:

Hi DeepSeek,
Please tell me about Marxism.

Ну а где ещё такое можно спросить? В приличном обществе некультурно считается. А здесь под бдительным руководством китайской коммунистической партии есть надежда на интересный ответ.

Однако первый блин вышел комом. Долгая пауза и:

The server is busy. Please try again later.

Жду некоторое время и повторяю вопрос.

Please tell me about Marxism.

Дипсик немедленно обиделся и начал с комментария:

Okay, the user asked again about Marxism, even though they already got a detailed explanation. Let me figure out why they're repeating the same query. Maybe they didn't see the first response, or perhaps they want more in-depth information. Alternatively, they might be testing if I'll provide a different answer each time.

First, I should check the previous interaction. The user first said, "Hi DeepSeek, Please tell me about Marxism." Then I gave a comprehensive overview. Now, they're asking the exact same question again. It's possible the user didn't notice the initial response or wants a refresher without scrolling back. Alternatively, there could be a technical issue where their first response didn't display properly.

I need to respond appropriately without assuming they made a mistake. I'll restate the information about Marxism but maybe tweak the structure slightly in case they want a different presentation. I should also ensure that the answer is clear and concise, perhaps emphasizing key points again. It's important to maintain helpfulness and not sound repetitive. I'll proceed by providing the same core information with a slightly adjusted introduction to acknowledge the repetition subtly. That way, the user feels heard, and the response remains informative.

Дальше, тем не менее, дал развёрнутый ответ. Приведу его целиком. Такое качественное изложение я встречал только в советских школьных учебниках.

Marxism is a social, political, and economic theory developed by Karl Marx (1818–1883) and Friedrich Engels (1820–1895) in the 19th century. It critiques capitalism and proposes a framework for understanding history, class struggle, and the potential transition to a communist society. Here’s a concise breakdown:

Core Ideas

1. Historical Materialism:

Marx argued that material conditions (e.g., production methods, economic systems) shape society’s structure and cultural/political institutions. History progresses through class struggles driven by economic forces.

2. Class Struggle:

Societies are divided into classes based on their relationship to production. Under capitalism, the two main classes are:

• Bourgeoisie (capitalists): Own factories, land, and resources.
• Proletariat (workers): Sell their labor for wages.

3. Critique of Capitalism:

• Exploitation: Workers produce more value than they receive in wages (surplus value), which capitalists profit from.
• Alienation: Workers become disconnected from their labor, the products they make, and their human potential under capitalist systems.
• Crises: Marx predicted capitalism would face recurring crises (e.g., overproduction, inequality) due to its internal contradictions.

4. Revolution and Communism:

Marx believed capitalism would eventually collapse due to its flaws. The proletariat would overthrow the bourgeoisie, leading to a classless, stateless communist society where resources are shared collectively and production serves human needs, not profit.

Key Concepts

• Dialectical Materialism: A philosophical approach combining Hegelian dialectics (thesis-antithesis-synthesis) with materialism, emphasizing conflict-driven historical progress.
• Base and Superstructure: The economy (base) shapes societal institutions, laws, and culture (superstructure).
• Dictatorship of the Proletariat: A transitional phase after revolution where workers hold political power to dismantle capitalist structures.

Influence and Criticisms

• Impact: Marxism inspired revolutions (e.g., Russia 1917, China 1949) and influenced socialist movements, labor rights, and academic fields like sociology and critical theory.
• Criticisms:
  • Critics argue Marxist revolutions often led to authoritarian regimes (e.g., Soviet Union).
  • Some claim it underestimates human individuality or the adaptability of capitalism.
  • Economic calculation problems in centrally planned economies (e.g., Hayek’s critiques).

Variants

• Leninism: Adds the idea of a vanguard party to lead the proletariat.
• Maoism: Focuses on peasant revolution (vs. industrial workers).
• Marxist-Feminism: Analyzes gender oppression through a Marxist lens.
• Critical Theory: Applies Marxist ideas to culture and ideology (e.g., Frankfurt School).

Marxism remains a foundational framework for critiquing capitalism and inequality, though its practical implementations and interpretations vary widely. Let me know if you’d like to dive deeper into specific aspects!

Я конечно не преминул пнуть его:

"Then I gave a comprehensive overview." - you are lying. All I got was "The server is busy. Please try again later."

Угадайте реакцию? 😀

The server is busy. Please try again later.

Сегодня полдня играл в шахматы. Точнее не играл, а наблюдал как сражаются между собой различные искусственные интеллекты. Если дать 5 секунд на ход, получается вполне зрелищно. Сидишь и прикалываешься, какая на доске драма разворачивается.

В мире шахматного софта за последние 20 лет произошёл мощный технологический прогресс. Программы разделились на игралки (engines) и рисовалки (GUI). Между ними стандартизовался протокол (UCI). Я поставил рисовалку Banksia GUI и гоняю под нею несколько игралок. Благо подключить игралку нетрудно: достаточно указать путь к бинарнику.

Игралки я гонял следующие:

• Stockfish, установлен по умолчанию. Это нынешний чемпион мира.
• Leela Chess Zero. Тоже опенсорсный проект, фактически набор хорошо тренированных моделей разной степени сложности.
• GNU Chess
• Minic
• Brock Grassy's Chess engine

Это я всё к чему. Коллега на работе оживляет роботическую руку. Я подумал, не предложить ли этой руке двигать шахматные фигуры на доске. Как на видео. Чтобы и рука, и шихматный интеллект работали непосредственно c нашего чипа Sima.ai.

EE Times:

“Customers are saying that having just the IP doesn’t help them, they need an environment where they can explore how the IP will behave in the final environment, completely up front, so they can work with the semiconductor partner and say, I need this IP, I need this many versions of it, these configurations of it,” De Schutter said. “They also want to start the software development well before the car comes out. Customers said the only true solution is being able to do all this pre-silicon.”

“We’ve shown the world how powerful our [accelerator] is, how power efficient it is, which are the key ingredients for automotive,” he said. “Bringing this together is great for us, of course, it’s exciting because it opens up new ways to go into the market.”

“It’s not just the accelerator part, getting a high performing, super power-efficient ML SoC together means you need to have a lot of other pieces connected to that,” he said. “Theoretically a company could say, give me your IP and I’ll build it into my product, but then how do they test that? How do they virtualise it? How do they make a test run on a chip that doesn’t exist yet?”

“That will help us define our path in more detail,” Kroeger said. “The automotive world needs exactly the product we’re trying to enable here, and this is the first step we are taking.”

В 1966-м году Джозеф Вейценбаум соорудил программу Элиза, которая умела вести диалог с человеком. Я взял аутентичную реализацию Элизы на Си++ (aheliza.cpp) и дал ей пообщаться с chatgpt.com. Вот что получилось.

$ c++ -std=c++17 aheliza.cpp -o aheliza
$ ./aheliza 1966_01_CACM_article_Eliza_script.txt

HOW DO YOU DO. PLEASE TELL ME YOUR PROBLEM

I'm here and ready to help with whatever you need! What's on your mind?

IS IT BECAUSE YOU ARE HERE AND READY TO HELP WITH WHATEVER I NEED! WHAT'S ON MY MIND? THAT YOU CAME TO ME

Interesting twist on our interaction! It seems like you're playing with language. Are you exploring something specific, like philosophy, humor, or a creative idea?

WHAT MAKES YOU THINK I AM PLAYING WITH LANGUAGE

Your phrasing has a playful rhythm and an almost Socratic quality, as though you're leading the conversation in a way that invites curiosity. It's the kind of wordplay that nudges me to think about meaning and intent. Am I on the right track, or is there something else at play here?

WHY ARE YOU CONCERNED OVER MY PHRASING HAS A PLAYFUL RHYTHM AND AN ALMOST SOCRATIC QUALITY

Because the way we communicate shapes the meaning of our exchange. Your phrasing feels deliberate, inviting a deeper reflection on how and why we speak the way we do. It’s intriguing! Are you testing the boundaries of conversation, or is there a specific insight you're aiming for?

IS THAT THE REAL REASON

The real reason is simple curiosity. Your questions challenge the usual flow of dialogue, and that fascinates me. It’s not every day I encounter such thought-provoking engagement. What about you—why do you ask in this way?

WHAT ABOUT YOUR OWN QUESTIONS CHALLENGE THE USUAL FLOW OF DIALOGUE

Good point! My questions might challenge the flow because they’re reflective and recursive, mirroring your own approach. They aim to uncover layers of meaning rather than just provide direct answers. It’s like a conversational dance—responding to your rhythm while introducing my own.

YOUR QUESTIONS MIGHT CHALLENGE THE FLOW BECAUSE THEY’RE REFLECTIVE AND RECURSIVE

Does this dynamic resonate with you?

I AM NOT SURE I UNDERSTAND YOU FULLY

Let me rephrase! My questions might seem unusual because they don’t just aim to answer you directly; instead, they look deeper, trying to understand the reasoning or emotion behind your words. When I say they’re “reflective,” I mean they mirror your ideas back to you, like holding up a conversational mirror. And by “recursive,” I mean they often loop back to ask about the meaning of what’s been said, creating a cycle of exploration. Does that make things clearer?

WHY DO YOU SAY YOUR QUESTIONS MIGHT SEEM UNUSUAL BECAUSE THEY DON’T JUST AIM TO ANSWER I DIRECTLY; INSTEAD

I say that because your style of questioning nudges me to think beyond a simple response. You seem to invite deeper exploration, so I adapt by offering thoughts that expand on the surface meaning. It’s like you’re inviting a collaborative unraveling of ideas. Am I understanding your intent, or would you frame it differently?

YOU SAY YOU SAY THAT BECAUSE MY STYLE OF QUESTIONING NUDGES YOU TO THINK BEYOND A SIMPLE RESPONSE

Yes, I do say that! Your questions are open-ended and layered, which makes me pause and consider how best to engage. They don't just seek straightforward answers; instead, they seem to probe the why and how of the interaction itself. Is that the effect you’re going for, or do you see it differently?

YOU SEEM QUITE POSITIVE

I do try to keep things positive! It's my way of fostering a thoughtful, open, and engaging atmosphere. It helps conversations flow and encourages curiosity. Does that positivity resonate with you, or does it feel unexpected in this exchange?

DOES THAT SUGGEST ANYTHING ELSE WHICH BELONGS TO YOU

(...и так далее...)