Computer Science and Programming
141K subscribers
893 photos
31 videos
37 files
1.19K links
Channel specialized for advanced topics of:
* Artificial intelligence,
* Machine Learning,
* Deep Learning,
* Computer Vision,
* Data Science
* Python

Admin: @otchebuch

Memes: @memes_programming

Ads: @Source_Ads,
https://telega.io/c/computer_science
Download Telegram
What's new in Git 2.55.0?
Git 2.55.0 introduces several notable features: a new `git history fixup` subcommand that amends staged changes into an existing commit and auto-rebases stacked branches; fsmonitor daemon support for Linux using inotify(7); the ability to push to a group of remotes with fsmonitor d a
at's new in Git 2.55.0
option for
in Git 2.55.0?
Gitto cap lane width; batched blob downloads for partial clones in git-grep and git-cherry; and Rust is now a required build dependency unless explicitly disabled.
❀5πŸ‘¨β€πŸ’»4πŸ‘2
Announcing SkiaSharp 4.0
SkiaSharp 4.148.0 is now available on NuGet as the first stable release of SkiaSharp v4. The release brings a Skia engine updated to milestone m148, delivering up to 24% faster GPU rendering (65β†’80 FPS for shadowed UIs), 6x faster CPU Perlin-noise shaders, variable font support, color font palettes, animated WebP encoding, and a cleaned-up API with fixed use-after-free crashes. Uno Platform is a co-maintainer alongside Microsoft's .NET team, meaning faster issue resolution and a predictable release cadence tied to Chrome's Skia milestones. Uno Platform users can opt in by setting the SkiaSharpVersion MSBuild property to 4.148.0. The next preview (4.150.0 Preview 2) is already available and introduces the Graphite next-gen GPU backend.
❀7πŸ‘3πŸ‘¨β€πŸ’»1
GigaChat 3.5 Ultra Publicly Released β€” The New Generation of the Flagship Model

The GigaChat team has released GigaChat 3.5 Ultra as open sourceβ€”a new 432B model under the MIT license. This is the first open-source hybrid of GatedDeltaNet and MLA scaled to hundreds of billions of parameters, featuring a proprietary training recipe we refined through more than 1,500 experiments. The model has grown in terms of code, mathematics, agent scenarios, and application domainsβ€”yet it’s 40% smaller than GigaChat 3.1 Ultra.

What’s inside:

πŸ”˜A proprietary hybrid MLA + Gated DeltaNet architecture with a dedicated stabilization framework, without which this hybrid setup would not train reliably at this scale;
πŸ”˜ Gated Attention: the model can locally down-weight overly strong signals from the attention layer;
πŸ”˜GatedNorm: normalization with an explicit gate that controls signal magnitude across features;
πŸ”˜Approximately 4x lower KV cache per token: with the same memory budget, the model can support 2.14x longer context and deliver a 20% throughput increase under load;
πŸ”˜Two MTP heads, enabling up to 2.2x faster generation;
πŸ”˜FP8 across all training stages with no quality degradation compared with bf16, enabled by custom Triton and CUDA kernels;
πŸ”˜A new online RL stage after SFT and DPO.

Results:

πŸ”˜ GigaChat-3.5-Ultra-Base outperforms DeepSeek V3.2 Exp Base and DeepSeek V4 Flash Base on average across a set of general, math, and code benchmarks:
πŸ”˜ GigaChat-3.5-Ultra-Instruct is comparable to DeepSeek V3.2 in terms of average score, despite having half the size;
πŸ”˜ According to the MiniMax-M2.7 LLM judge, the average win rate against GigaChat 3.1 Ultra is 75.9%, and against GPT-5 is 68.7%.

The entire stack β€” data (our own LLM-filtered Common Crawl, 600+ programming languages in the code), architecture, training methodology, and infrastructure β€” was built end-to-end by GigaChat team.

➑️ HuggingFace
Please open Telegram to view this post
VIEW IN TELEGRAM
❀22πŸ‘6πŸ‘Ž2πŸ‘¨β€πŸ’»1
DuckDuckGo browser now blocks YouTube video ads
DuckDuckGo's browser now blocks most YouTube video ads, including pre-roll and mid-roll ads, across iOS, Mac, and Windows by default. Android users must enable it manually in Settings. The feature uses community-maintained filter lists from uBlock Origin, supplemented by DuckDuckGo's own compatibility rules. It works separately from the existing Duck Player feature, and both can be used simultaneously. DuckDuckGo joins Brave and Opera in offering built-in YouTube ad blocking without third-party extensions. Minor buffering delays may occur, and the feature may occasionally break as YouTube changes its ad-serving methods.
πŸ‘28πŸ—Ώ10😁5πŸ‘¨β€πŸ’»4
Hello everyone some of you may not know but now we started to expand to WhatsApp and we are inviting you to join our WhatsApp channel link here πŸ‘‡πŸ‘‡πŸ‘‡πŸ‘‡πŸ‘‡

WhatsApp Computer Science and Technology

We welcome you all to be part of this family
πŸ‘Ž12πŸ”₯4❀3😁1
SkiaSharp for .NET
SkiaSharp 4.0 is now stable, marking the biggest release in years for the cross-platform rendering library that underpins .NET MAUI, WinUI 3, and Uno Platform. Built on Google's Skia engine (which also powers Chrome and Android), SkiaSharp provides a single GPU-accelerated, pixel-perfect rendering API across Windows, macOS, Linux, iOS, Android, and WebAssembly. Version 4.0 brings the native engine current through Skia milestone m148 (28 upstream milestones), a reworked object lifecycle that eliminates use-after-free crashes, variable font support, color font palettes, animated WebP encoding, and up to 24% faster GPU rendering (30% for Uno Platform apps). Uno Platform is now a formal co-maintainer alongside Microsoft's .NET team, ensuring predictable release cadence
❀16πŸ‘¨β€πŸ’»3πŸ‘1
Announcing .NET Modernization for Beginners
Microsoft has released a free, open-source, hands-on course called '.NET Modernization for Beginners' to help developers upgrade legacy ASP.NET applications to .NET 10 using the GitHub Copilot modernization agent. The four-chapter course covers assessment, planning, upgrade execution, and cloud deployment to Azure App Service. Unlike typical agentic tools, the GitHub Copilot modernization agent produces transparent, editable artifacts (assessment.md, plan.md, tasks.md) that developers review and adjust before any code changes occur. A GitHub Copilot subscription and Visual Studio 2022 (17.10+) are required.
❀8πŸ—Ώ2πŸ‘1
The Joy of Old-School Coding After a Stupid Mistake
A short personal reflection on the frustration and satisfaction of debugging without AI assistance. The author made a mistake using a collection to map a one-to-one relationship in Entity Framework Core, which only integration tests caught. After resisting the urge to use Copilot, they found the fix themselves β€” a one-line change β€” and reflect on the emotional cycle of coding: frustration followed by joy.
❀9πŸ‘5
Hello everyone some of you may not know but now we started to expand to WhatsApp and we are inviting you to join our WhatsApp channel link here πŸ‘‡πŸ‘‡πŸ‘‡πŸ‘‡πŸ‘‡

WhatsApp Computer Science and Technology

We welcome you all to be part of this family
πŸ‘Ž11❀8πŸ‘2πŸ‘¨β€πŸ’»2
Locally great, globally drifting
An evaluation of AI-generated frontend code in a real product reveals a recurring pattern: AI performs well in isolated, localized contexts but struggles to maintain consistency at the system level. The post argues that AI should be used as a candidate generator and validator rather than as the final decision-maker in code review.
❀8πŸ”₯2
How I Learned to Structure React Projects Around Change, Not Folders
Technical folder structures (components, hooks, services, utils) make React repos easy to browse but hard to change, because a single feature's logic is scattered across unrelated directories. The key insight is organizing code around ownership of change rather than file type. This means colocating a feature's components, state, validation, API calls, and business rules under one feature directory. State placement should reflect lifecycle and source of truth, not convenience. Business rules belong near the feature that owns them, not in generic utils. Shared code should only be extracted when callers truly share the same meaning and change together. Finally, dependency direction matters: features should depend on stable shared primitives, not each other's internals. The real test of an architecture is how confidently a developer can locate the owner of the next change.
πŸ”₯8❀3πŸ‘2
Hello everyone some of you may not know but now we started to expand to WhatsApp and we are inviting you to join our WhatsApp channel link here πŸ‘‡πŸ‘‡πŸ‘‡πŸ‘‡πŸ‘‡

WhatsApp Computer Science and Technology

We welcome you all to be part of this family
πŸ‘Ž8πŸ”₯3❀2