ChatCrawlersearch across public Telegram Open the app
P

Python Daily Group

30 members
2 August 2026
id 1083679662
webpage
Kimi K3 Deep Dive — Architecture, Training & Benchmarks of the 2.78-Trillion-Parameter Open-Weight Model D Hi everyone! 👋 I wrote an extensive technical deep-dive into Moonshot AI's Kimi K3, analyzing its architectural innovations, training stability tricks, and benchmark performance. The blog covers: Kimi Delta Attention (KDA) Attention Residuals Stable LatentMoE Quantile Balancing NoPE 1M-token context RL training pipeline Infrastructure and serving optimizations I’d love to get your feedback and discuss any specific parts of the architecture! Blog: https://imrancoder786.github.io/blog-post.html?post=kimi-k3-deep-dive https://redd.it/1vdndys @pythondaily
17 ·
I
webpage
Turned a folder of ad-hoc scripts into a proper installable CLI (pyproject.toml entry point) Had a data pipeline that was just `python build_db.py`, `python clean_validate.py`, `streamlit run streamlit_app.py` — worked, but not something you'd hand someone as "the tool." Packaged it as a real CLI: `[project.scripts]` entry point in `pyproject.toml`, `pip install -e .` → `promo-toolkit build / validate / serve`. Also added: * Incremental rebuilds (content-hash diff + scoped upsert instead of full rewrite each run) * pytest suite + ruff, GitHub Actions CI across two Python versions * A separate eval harness for an AI-agent piece of the same project (real API calls scored against golden answers) Not a library, just a real end-to-end tool — sharing mostly for feedback on the packaging/CLI structure since that's the part I have the least experience with. Repo: \[https://github.com/SpiliosDimakopoulos/retail-promo-analytics-toolkit\]. https://redd.it/1vdrvh6 @pythondaily
16 ·
I
id 1083679662
webpage
neurips 2026: ACs and reviewers have disappeared D we submitted our rebuttal via the "Rebuttal" button before the author/reviewer/AC discussion period officially opened (Jul 27 AoE). since then, we've gotten complete silence from all four reviewers and the AC several of us are also reviewing this cycle. when the discussion period opened on Jul 27 AoE, we got no email notification for rebuttals on papers we're reviewing, specifically for the papers whose authors had also posted early via the "Rebuttal" button. so it feels like anything submitted before the window opened may simply have never triggered any notification we also tried: (1) meta-comments visible to everyone (PCs + ACs + reviewers), (2) individual nudges to reviewers, and (3) sending an email to the PCs given there's about 1 day left in the discussion period, what do we do? we honestly thought we had a shot at an oral or spotlight given our initial scores. this is completely messed up. https://redd.it/1vdu92a @pythondaily
14 ·
3 August 2026
I
id 1083679662
webpage
Complete Python Course for Beginners (2026) |Python from Scratch to Real Projects Master Python from absolute beginner to confident programmer with this complete Python course! Whether you’re new to programming or looking to build a strong foundation for Artificial Intelligence, Machine Learning, Data Science, Automation, Web Development, or Software Engineering, this step-by-step course is designed to help you succeed. Every lesson builds on the previous one using clear explanations, practical coding examples, hands-on exercises, and real-world projects. No prior programming experience is required. https://youtu.be/UQ-A2ACLcaM?si=3ysTbKGPb_qW0jh2 https://redd.it/1vdx5y7 @pythondaily
10 ·
id 1083679662
webpage
Monday Daily Thread: Project ideas! # Weekly Thread: Project Ideas 💡 Welcome to our weekly Project Ideas thread! Whether you're a newbie looking for a first project or an expert seeking a new challenge, this is the place for you. ## How it Works: 1. **Suggest a Project**: Comment your project idea—be it beginner-friendly or advanced. 2. **Build & Share**: If you complete a project, reply to the original comment, share your experience, and attach your source code. 3. **Explore**: Looking for ideas? Check out Al Sweigart's ["The Big Book of Small Python Projects"](https://www.amazon.com/Big-Book-Small-Python-Programming/dp/1718501242) for inspiration. ## Guidelines: * Clearly state the difficulty level. * Provide a brief description and, if possible, outline the tech stack. * Feel free to link to tutorials or resources that might help. # Example Submissions: ## Project Idea: Chatbot **Difficulty**: Intermediate **Tech Stack**: Python, NLP, Flask/FastAPI/Litestar **Description**: Create a chatbot that can answer FAQs for a website. **Resources**: [Building a Chatbot with Python](https://www.youtube.com/watch?v=a37BL0stIuM) # Project Idea: Weather Dashboard **Difficulty**: Beginner **Tech Stack**: HTML, CSS, JavaScript, API **Description**: Build a dashboard that displays real-time weather information using a weather API. **Resources**: [Weather API Tutorial](https://www.youtube.com/watch?v=9P5MY_2i7K8) ## Project Idea: File Organizer **Difficulty**: Beginner **Tech Stack**: Python, File I/O **Description**: Create a script that organizes files in a directory into sub-folders based on file type. **Resources**: [Automate the Boring Stuff: Organizing Files](https://automatetheboringstuff.com/2e/chapter9/) Let's help each other grow. Happy coding! 🌟 https://redd.it/1vdxnhc @pythondaily
15 ·
I
webpage
No rebuttals from neurips authors D I know there’s a lot of frustration around no response from reviewers, which I also got only one so yeah what a bummer, but I was wondering if no rebuttal from the authors was just as common or not. I got no rebuttal so far, so I’m here scratching my head what might have happened to the authors lol especially when at least one paper was pretty much on the borderline with somewhat of a positive AC comment https://redd.it/1vdz19e @pythondaily
18 ·
id 1083679662
webpage
Neurips 2026: does every metareview recommend accept/reject? D I see some people say their metareview already contains a decision/recommendation (all of them were rejections). Ours doesn’t. Even though our avg score is 3, the metareview seems optimistic and finishes with “a convincing response would be an important consideration while discussing the paper.” I wonder how to interpret that. We did a strong rebuttal, but none of the reviewers engaged. So I wonder whether there’s any point to keep hope due to the AC review or just give up. https://redd.it/1vdvkp5 @pythondaily
15 ·
webpage
Is it too late regain some coherence in the ML research space in our life time? D Was just looking at the list of preprints on Arxiv cs.LG https://arxiv.org/list/cs.LG/recent?skip=0&show=500 Everyday 100 - 400 new machine learning papers gets uploaded on this server. Looking at this unending list of preprints is as if you stepped into a crowded room, like the stock trading floor on wall st. in the 1980s. Everyone is shouting over each other. Nobody is talking to each other. Everyone's trying to prove something, to someone, to themselves, to build some credentials in the ML/AI space to meet those job requirements, or dying to get their truth out. Every title contains some new terminology invented by the authors that feels not worth the effort in keeping it in your working memory. Burn-out by endless novelty. Frontier research are now corporate trade secrets that politicians and military are watching closely. Research papers are ir/unreproducible he-said-she-saids. Marketing material are research paper and vice versa. Extremely major breakthroughs are announced via tweets, whereas extremely minor results are unannounced via journals. Everything feels simultaneously mostly true and possibly false (because nobody is seriously checking). Nobody knows what's going on, and people who knows what's going on has a non-disclosure clause in their job contract. Is the theory of generalization that we learned in school true or false? It feels false, why hasn't there been any retractions? Many questions like these. Is it too late to regain some coherence in this field?? https://redd.it/1ve7chh @pythondaily
14 ·
webpage
Confused on writing User model. I am making a project in which users can access my backend service via their own API keys. While the models I got confused how many classes i should make. generally i make 2 classes for the user 1. User(AbstractUser) class which have my customusermanager object , this generally have email and password and sometimes registration method (Google/ basic email and password) 2. and UserProfile(RLSModel/models.Model) which have foreign key User, this model have all the details of the user like name , phone , email , role etc. i give permissions based on role type. but in this project i am confused weather to store the APIKey(hased sha256) in userprofile model, user model or make a new model like UserApi key which will have user as foreign key. can someone please tell me what's the industry standard way to write these type of user models. and whats the best way to writing models in django rest framework. please dont judge me by this question, i know its stupid to ask such silly questions but i really wanna learn building good and reliable backends https://redd.it/1ve7pjr @pythondaily
13 ·
webpage
Python learning I'm a complete beginner to Python and I'm looking for good learning resources.Could you recommend, Youtube channel that's beginner friendly and explains concepts clearly and udemy course that's worth buying (preferably with hands-on projects). Thank you https://redd.it/1vebvzf @pythondaily
12 ·
I
webpage
Confused on writing User model. I am making a project in which users can access my backend service via their own API keys. While the models I got confused how many classes i should make. generally i make 2 classes for the user 1. User(AbstractUser) class which have my customusermanager object , this generally have email and password and sometimes registration method (Google/ basic email and password) 2. and UserProfile(RLSModel/models.Model) which have foreign key User, this model have all the details of the user like name , phone , email , role etc. i give permissions based on role type. but in this project i am confused weather to store the APIKey(hased sha256) in userprofile model, user model or make a new model like UserApi key which will have user as foreign key. can someone please tell me what's the industry standard way to write these type of user models. and whats the best way to writing models in django rest framework. please dont judge me by this question, i know its stupid to ask such silly questions but i really wanna learn building good and reliable backends https://redd.it/1ve7pjs @pythondaily
7 ·
id 1083679662
webpage
How I turned my Django CRM into a modular platform with runtime feature flags Part 1 | Part 2 | Part 3 — production CRM for truck service center, built with Django + DRF. So, by v2.3, TruckMaster had grown to 10+ Django apps: orders, clients, inventory, ALPR, apointments, invoices, Telegram bot, and more. Owner started asking things like "can we turn off the appointment booking now, we are not ready for it " or "client X should have access to invoices but client Y shouldn't." I could've just hardcoded some if-statements and call it a day. Instead I built module system. Probably overengineered it, but it turned out to be one of the best architecture decision in whole project. # The idea Every Django app registers itself as a "module" at startup. Owner can flip modules on/off from the admin panel. There are no code changes, no redeployment, no touching the server. Disable ALPR? The entire /api/alpr/ prefix returns 503. Re-enable it? Everything back's in 60 seconds (cache TTL). On top of that, there is per-client feature access. Even if module is globally enabled individual clients can have different features turn on or off. Client A get Telegram bot and invoice access, client B only get basic cabinet. # How apps register themselves Each Django app that want to be a module declares a MODULE_INFO dict in its AppConfig and call register_module() in ready(): class AlprConfig(AppConfig): name = 'alpr' verbosename = 'License Plate Recognition' MODULEINFO = { 'name': 'alpr', 'label': 'License Plate Recognition', 'description': 'ALPR camera: plate recognition, arrival log.', 'iscore': False, 'urlprefixes': '/api/alpr/', 'dependencies': 'clients', 'order': 9, } def ready(self): from core.registry import registermodule registermodule(self.MODULEINFO) Registry is just a plain dict in memory. I considered to us
2 ·
u/admin.display(description='Status') def toggleswitch(self, obj): if obj.iscore: return formathtml( '<label class="module-toggle module-toggle--core">' '<input type="checkbox" checked disabled>' '<span class="module-togglelabel">Core</span>' '</label>', ) checked = 'checked' if obj.isenabled else '' return formathtml( '<label class="module-toggle" data-pk="{}">' '<input type="checkbox" {}>' '<span class="module-togglelabel">{}</span>' '</label>', obj.pk, checked, 'Enabled' if obj.isenabled else 'Disabled', ) Custom URL endpoint handles the POST, toggles is_enabled, clear the cache. Core modules get a disabled checkbox with a tooltip explaining why. I also disabled add/delete permissions on the Module admin — modules are created automatically, not manually. # Auto-sync with postmigrate New module appears in the database automatically. When you run migrate, a post\migrate signal fires and syncs the in-memory registry with the DB: def syncmodulesaftermigrate(sender, kwargs): registry = getregistry() if not registry: return for name, info in registry.items(): module, created = Module.objects.getorcreate( name=name, defaults={ 'label': info['label'], 'isenabled': True, 'iscore': info.get('iscore', False), 'urlprefixes': info.get('urlprefixes', ), 'dependencies': info.get('dependencies', ), }, ) if not created: # Update metadata but don't touch isenabled module.urlprefixes = info.get('urlprefixes', []) module.dependencies = info.get('dependencies', []) module.save() The important l
4 ·
I
webpage
second 59, it still goes through. For a truck service center this is fine. For anything time-sensitive, you would want cache.delete() on save (which I do) plus maybe a shorter TTL as a safety net. Don't let users delete Module records. First version allowed it. Owner accidentally deleted the "orders" module from admin. The middleware stopped checking it, which meant the endpoints still worked, but the admin toggle was gone. Took me a while to figure out why the module list looked wrong. Now has_delete_permission returns False. # What is next Part 5 — Nova Poshta integration. That is Ukraine's biggest postal service and I builded tracking for spare parts deliveries directly into the CRM. The Telegram bot can check shipment status too. Also auto-closing of completed orders and auto-deducting of parts from warehouse stock. Previous posts: Part 1 | Part 2 | Part 3 GitHub (demo repo): github.com/VNmagistr/truckmaster\_demo — branch demo/v2.3 To be continued... # https://redd.it/1vefehr @pythondaily
6 ·
I
id 1083679662
webpage
NeurIPS 2026: If the rebuttal addresses your concern, please raise your score D Potentially a hot take? I am not sure why our community is plagued with reviewers who, after acknowledging that their concerns were addressed by a rebuttal, decide to maintain their score because they don't vibe with the paper. So here is my plea to all reviewers: If you list a set of concerns in your review and these concerns are addressed during the rebuttal, please adjust your score accordingly. This should apply whether or not you like the paper and/or its methodology. The beauty of scientific research is that we each get to explore ideas that we find meaningful whose value may not be immediately obvious to every individual reviewer. https://redd.it/1vefwvh @pythondaily
4 ·
Archive by month
August 2026July 2026June 2026May 2026April 2026March 2026February 2026January 2026December 2025November 2025October 2025September 2025
Каталог площадок Искать в ChatCrawler

A snapshot of an open public feed from the search index ChatCrawler — “Google for public Telegram”; refreshed as the venue is crawled. Times are UTC.

Public content only, official Telegram API. About · FAQ · What we do not do · Remove a page · Catalog