[paper][GC][state machine]
https://arxiv.org/html/2405.11182v1
In this paper, the authors quantify the overhead of running a state machine replication system for cloud systems written in a language with garbage collection (GC). To this end, they (1) design a canonical cloud system—a distributed, consensus-based, linearizable key-value store—from scratch, (2) implement it in C++, Java, Rust, and Go, and (3) evaluate the implementations under update-heavy and read-heavy workloads on AWS with different resource constraints, aiming to maximize throughput while maintaining low tail latency. The results show that GC incurs a non-trivial cost, even with ample memory. With limited memory, languages with manual memory management can achieve an order of magnitude higher throughput than those with GC on the same hardware. A key observation is that if a cloud system is expected to scale significantly, building it in a language with manual memory management, despite the higher development cost, may lead to substantial cloud cost savings in the long run.
https://arxiv.org/html/2405.11182v1
In this paper, the authors quantify the overhead of running a state machine replication system for cloud systems written in a language with garbage collection (GC). To this end, they (1) design a canonical cloud system—a distributed, consensus-based, linearizable key-value store—from scratch, (2) implement it in C++, Java, Rust, and Go, and (3) evaluate the implementations under update-heavy and read-heavy workloads on AWS with different resource constraints, aiming to maximize throughput while maintaining low tail latency. The results show that GC incurs a non-trivial cost, even with ample memory. With limited memory, languages with manual memory management can achieve an order of magnitude higher throughput than those with GC on the same hardware. A key observation is that if a cloud system is expected to scale significantly, building it in a language with manual memory management, despite the higher development cost, may lead to substantial cloud cost savings in the long run.
🔥2
[ethz][computer architecture][lectures]
https://safari.ethz.ch/architecture/fall2022/doku.php?id=schedule
https://safari.ethz.ch/architecture/fall2022/doku.php?id=schedule
🔥1
[paper][ClickHouse]
This paper presents an overview of ClickHouse, a popular open- source OLAP database designed for high-performance analytics over petabyte-scale data sets with high ingestion rates. Its storage layer combines a data format based on traditional log-structured merge (LSM) trees with novel techniques for continuous trans- formation (e.g. aggregation, archiving) of historical data in the background. Queries are written in a convenient SQL dialect and processed by a state-of-the-art vectorized query execution engine with optional code compilation. ClickHouse makes aggressive use of pruning techniques to avoid evaluating irrelevant data in queries. Other data management systems can be integrated at the table function, table engine, or database engine level. Real-world bench- marks demonstrate that ClickHouse is amongst the fastest analyti- cal databases on the market.
https://www.vldb.org/pvldb/vol17/p3731-schulze.pdf
This paper presents an overview of ClickHouse, a popular open- source OLAP database designed for high-performance analytics over petabyte-scale data sets with high ingestion rates. Its storage layer combines a data format based on traditional log-structured merge (LSM) trees with novel techniques for continuous trans- formation (e.g. aggregation, archiving) of historical data in the background. Queries are written in a convenient SQL dialect and processed by a state-of-the-art vectorized query execution engine with optional code compilation. ClickHouse makes aggressive use of pruning techniques to avoid evaluating irrelevant data in queries. Other data management systems can be integrated at the table function, table engine, or database engine level. Real-world bench- marks demonstrate that ClickHouse is amongst the fastest analyti- cal databases on the market.
https://www.vldb.org/pvldb/vol17/p3731-schulze.pdf
👍2
[cicd][uber]
https://www.uber.com/en-NL/blog/continuous-deployment/
“Uber’s business runs on a myriad of microservices. Ensuring that changes to all of these services are deployed safely and in a timely manner is critical. By utilizing continuous deployment to automate this process, we ensure that new features, library updates, and security patches are all delivered to production without unnecessary delays, improving the overall quality of code serving our business.
In this article, we share how we reimagined continuous deployment of microservices at Uber to improve our deployment automation and the user experience of managing microservices, while tackling some of the peculiar challenges of working with large monorepos with increasing commit volumes.
“
https://www.uber.com/en-NL/blog/continuous-deployment/
“Uber’s business runs on a myriad of microservices. Ensuring that changes to all of these services are deployed safely and in a timely manner is critical. By utilizing continuous deployment to automate this process, we ensure that new features, library updates, and security patches are all delivered to production without unnecessary delays, improving the overall quality of code serving our business.
In this article, we share how we reimagined continuous deployment of microservices at Uber to improve our deployment automation and the user experience of managing microservices, while tackling some of the peculiar challenges of working with large monorepos with increasing commit volumes.
“
👍3❤1
[ai][moshi]
Moshi is made of three main components: Helium, a 7B language model trained on 2.1T tokens, Mimi, a neural audio codec that models semantic and acoustic information, and a new multi-stream architecture that jointly models audio from the user and Moshi on separate channels.
https://kyutai.org/Moshi.pdf
https://github.com/kyutai-labs/moshi
https://huggingface.co/kmhf
Moshi is made of three main components: Helium, a 7B language model trained on 2.1T tokens, Mimi, a neural audio codec that models semantic and acoustic information, and a new multi-stream architecture that jointly models audio from the user and Moshi on separate channels.
https://kyutai.org/Moshi.pdf
https://github.com/kyutai-labs/moshi
https://huggingface.co/kmhf
[llm][comparison]
https://github.com/rasbt/LLMs-from-scratch/blob/main/ch05/07_gpt_to_llama/converting-gpt-to-llama2.ipynb
https://github.com/rasbt/LLMs-from-scratch/blob/main/ch05/07_gpt_to_llama/converting-gpt-to-llama2.ipynb
GitHub
LLMs-from-scratch/ch05/07_gpt_to_llama/converting-gpt-to-llama2.ipynb at main · rasbt/LLMs-from-scratch
Implement a ChatGPT-like LLM in PyTorch from scratch, step by step - rasbt/LLMs-from-scratch
[virtual machines][hypervisor]
https://x.com/chessman786/status/1855562661074968729?s=46&t=eNN3Y-GKeBSlFyyj1ozvgg
https://x.com/chessman786/status/1855562661074968729?s=46&t=eNN3Y-GKeBSlFyyj1ozvgg
🔥4
[ai][transformers.js]
Are we gonna run models in the browsers?
Here a great overview of what folks achieved so far:
https://www.youtube.com/watch?v=n18Lrbo8VU8
Are we gonna run models in the browsers?
Here a great overview of what folks achieved so far:
https://www.youtube.com/watch?v=n18Lrbo8VU8
YouTube
Transformers.js: State-of-the-art Machine Learning for the web
Join Joshua Lochner from HuggingFace to learn about Transformers.js, an exciting new JavaScript library that empowers developers to build never-before-seen web applications. It is designed to be functionally equivalent to Hugging Face's Python transformers…
[hardware][ai][network]
Quick overveiew on the hardware for AI infra (swtiches) at Meta
https://engineering.fb.com/2024/10/15/data-infrastructure/open-future-networking-hardware-ai-ocp-2024-meta/
Quick overveiew on the hardware for AI infra (swtiches) at Meta
https://engineering.fb.com/2024/10/15/data-infrastructure/open-future-networking-hardware-ai-ocp-2024-meta/
Engineering at Meta
OCP Summit 2024: The open future of networking hardware for AI
At Open Compute Project Summit (OCP) 2024, we’re sharing details about our next-generation network fabric for our AI training clusters. We’ve expanded our network hardware portfolio and are contrib…
[competitive programming][book]
As we are on the open market and looking for a job we aim to be good enough to pass the gates. But what if we are thinking about perfection?
https://cses.fi/book/book.pdf
As we are on the open market and looking for a job we aim to be good enough to pass the gates. But what if we are thinking about perfection?
https://cses.fi/book/book.pdf
👍2
[leadership][it’s not ai]
Great to see how leaders express their thoughts on making things done as well as how they treat theirselves in the times when you need to make product successfull. Some things you can read through the lines. 3 mins read and it’s worth it.
https://www.notion.so/blog/5-principles-for-effective-ai-leadership-without-deep-expertise
Great to see how leaders express their thoughts on making things done as well as how they treat theirselves in the times when you need to make product successfull. Some things you can read through the lines. 3 mins read and it’s worth it.
https://www.notion.so/blog/5-principles-for-effective-ai-leadership-without-deep-expertise
Notion
5 principles for effective AI leadership without deep expertise
In leadership roles, especially technical-leadership roles, there are few subjects you will be asked about more often than AI. But what if, like me until recently, you have lots of technical experience but have yet to dive meaningfully into AI development?
[data structures][paper]
Cache-Oblivious Algorithms
and Data Structures
https://erikdemaine.org/papers/BRICS2002/paper.pdf
Cache-Oblivious Algorithms
and Data Structures
https://erikdemaine.org/papers/BRICS2002/paper.pdf
[genAI][clone your c-lvl]
The promise of human behavioral simulation—general-purpose computational agents that replicate human behavior across domains—could enable broad applications in policymaking and social science. We present a novel agent architecture that simulates the attitudes and behaviors of 1,052 real individuals—applying large language models to qualitative interviews about their lives, then measuring how well these agents replicate the attitudes and behaviors of the individuals that they represent. The generative agents replicate participants' responses on the General Social Survey 85% as accurately as participants replicate their own answers two weeks later, and perform comparably in predicting personality traits and outcomes in experimental
replications.
https://arxiv.org/pdf/2411.10109
The promise of human behavioral simulation—general-purpose computational agents that replicate human behavior across domains—could enable broad applications in policymaking and social science. We present a novel agent architecture that simulates the attitudes and behaviors of 1,052 real individuals—applying large language models to qualitative interviews about their lives, then measuring how well these agents replicate the attitudes and behaviors of the individuals that they represent. The generative agents replicate participants' responses on the General Social Survey 85% as accurately as participants replicate their own answers two weeks later, and perform comparably in predicting personality traits and outcomes in experimental
replications.
https://arxiv.org/pdf/2411.10109