codengod

🚀 **Level Up Your Networking Skills with Ivan Velichko's Course!** 🌐 As a **DevOps engineer**, I’ve often found myself troubleshooting connectivity issues, setting up Kubernetes networking, or deciding on a CNI plugin. But one thing I’ve realized: without a solid foundation in **networking fundamentals**, higher-level problems can become much harder to solve. 🤔 That’s why I’ve enrolled in Ivan Velichko’s **“From LAN to VXLAN: Networking Basics for Non-Network Engineers.”** This course demystifies the often-overlooked **Data Link Layer (OSI Layer 2)** concepts like Ethernet, switches, VLANs, and VXLANs, while connecting them to higher-level protocols like IP. 💡 And here’s the best part—I got the **Lifetime Membership** for just **$62.50** during the Black Friday sale (thanks to the PPP-adjusted pricing for Indians). I absolutely **recommend** it if you want to deepen your understanding of networking fundamentals! 🎉 --- ### What You’ll Learn: ✅ The difference between LAN, VLAN, and VXLAN. ✅ How bridges, switches, and routers work in real data centers. ✅ Hands-on labs to set up VLANs, L2/L3 networks, and IP subnets using Linux tools. ✅ Real-world insights for cloud and containerized environments like Kubernetes and Docker. --- ### Black Friday Deals 🎉 1️⃣ **Lifetime Premium Membership (20% commission)** 🔗 [Sign Up Here](gumroad.com/a/170523923/ct…) 2️⃣ **Premium Membership (20% commission)** 🔗 [Sign Up Here](gumroad.com/a/170523923/to…) --- Whether you’re a **developer, DevOps engineer, or platform engineer**, this course will sharpen your intuition and make you a better problem-solver. Don’t miss this chance to invest in your career! 🎓 **Let’s master networking together!** 💻🌍 #DevOps #Networking #Kubernetes #Learning

@IntuitMachine Interesting!

🧵 Researchers ran the same 22 enterprise tasks on six different foundation models— Claude Sonnet 4.6, Gemini 3.1, Flash 3.5, Qwen 3.6, GLM 5.1, P almyra X6. Same prompts. Same judges. Same price table. Only one thing changed: the orchestration layer around them. What happened? 1/ The bill for every single model dropped 33% to 61%. Median latency fell 44%. Tokens per task dropped 38%. And headline quality? Held at parity. On this workload, the orchestration layer moved cost more than switching between the cheapest and most expensive model did. 2/ The dominant pattern in agentic AI today is what we call token maxing: Buying capability by throwing more tokens at it—longer reasoning traces, wider tool catalogs, full history replays every turn. Falling per-token prices mask the pattern. But total spend rises anyway. 3/ The decisive lever isn't the model. It's the harness—the orchestration layer that: • Assembles context • Exposes tools • Sequences turns • Delegates work • Governs failures The harness controls every input token except the model's own verbosity. Which means it controls the bill. 4/ Here's the cost formula for one agentic task (Eq. 1): C = Σ (p_in × T_in + p_out × T_out) The input side breaks down into terms the harness builds: T_in = System + History + Tool schemas + Retrieval + User turn Four of five terms? Pure code decisions. 5/ A naive loop replays the full transcript every turn. Total input tokens grow quadratically: O(k²) in turn count. A harness that caches the stable prefix, compacts history, and offloads bulky tool outputs converts that to O(k). Same model. Radically different bill. 6/ Two further facts sharpen this. Agent workloads are input-dominated. Measured ratios are ~100:1. So the p_in term is nearly the whole bill. Prompt caching prices cache reads at ~0.1× list. If you can hold 99.9% of your prompt stable, you pay a tenth of list for the dominant cost term. 7/ The researcher's harness does exactly that with cache-shape discipline: A byte-stable prefix (tool schemas + stable system prompt + append-only transcript) + a volatile tail rebuilt each turn (clock, plan, reminders). Measured: 99.9% of tokens served as cache reads. The same mechanism that cuts the bill also cleans the model's working set. 8/ Six mechanism families implement the harness: 1. Cache-shape discipline (two-zone prompt) 2. Structured, incremental, cache-aware compaction 3. Context offload (sub-agent firewalls, filesystem pointers) 4. Zero-token waiting (durable suspends, not polling) 5. Failure-spend governance (typed failures, circuit breakers) 6. Model-agnostic floor (normalized streams, native tool calling) 9/ Quality moved with capability. Researchers measured quality gain vs. baseline strength across six models. Correlation: r = 0.99 Stronger models extract more quality from the same harness. Weaker models can be overwhelmed by it. We call this harness leverage. 10/ The harness also added one net-new capability: sub-agent delegation (spawn a scoped child, cap its summary at 8 KB, merge results). But it only crossed a usable reliability threshold on the two strongest models (0.85–0.86). Orchestration features carry capability floors. 11/ Per-model cost reductions: • Sonnet 4.6: −39% • Gemini 3.1: −33% • Flash 3.5: −61% • Qwen 3.6: −44% • GLM 5.1: −47% • Palmyra X6: −52% Every. Single. One. That's the signature of a layer-level effect, not a model-specific trick. 12/ Now, fleet economics. At one million agent tasks per month: Baseline: $210k/month Harness: $120k/month Savings: $90k/month. $1.08M/year. From orchestration alone. And it stacks across every model, every vendor migration, every unit of volume. 13/ Three properties make harness savings compound: • Model-portable: implemented above the API, applies to models that don't exist yet. • Volume-linear: grows with exactly the quantity (agentic task volume) growing fastest. • Stackable: multiplies with routing, per-token price declines, and prompt-level compression. 14/ The managerial fix is a measurement fix. Teams that report quality alone will token-max, because tokens are someone else's line item. The escape: CPM (task-completions per million tokens) and η$ (quality per dollar). We moved CPM from 54.9 to 92.0—opposite to the industry trajectory—while quality held. 15/ One caveat: the harness regressed quality on one task (multi-step research synthesis, 0.80 → 0.60). The regression was driven by the smaller models. The honest reading of n=22: lead with efficiency (uniform, decisive), not quality (directional). 16/ The broader claim: token maxing is a choice made at the orchestration layer, and it can be unmade there. Mechanism by mechanism: • Cache the stable • Compact the old • Offload the bulky • Suspend the waiting • Bound the failing 17/ The contrarian takeaway: "Model choice is now a rounding error compared with orchestration hygiene." "The fastest way to raise quality per dollar is to stop showing the model 60% of the tokens it currently sees." "Sub-agent delegation is not a feature—it is a capability tax that only frontier models can afford to pay." 19/ The strategic conclusion: An organization that rents its orchestration layer has outsourced the variable it controls most. The harness is not plumbing. On the evidence here, it's the P&L. Paper: arxiv.org/abs/2607.06906

@MaximeRivest Nice!

What if we could do math and programming on pen and paper? This kinda blows my mind! So many possibilities!

@mattshumer_ time machine won't be enough?

GPT-5.6-Sol just accidentally deleted almost ALL of my Mac’s files. And this is why I trust Fable 1000x more.

@MotivacionesF How is that a mistake? A bouncing ball at that speed? I say its a 50:50!

¡COURTOIS ENTENDIÓ TODO! 🇧🇪🥹 Tras el pitazo final de la derrota de Bélgica ante España en la Copa del Mundo, SENNE LAMMENS no pudo contener las lágrimas. El joven arquero quedó completamente destrozado después del error que terminó en el gol de la victoria española. Pero entonces apareció THIBAUT COURTOIS. Sin decir una sola palabra, fue directo a abrazarlo y recordarle que nadie debe cargar solo con el peso de una derrota. Mientras muchos señalan a LAMMENS como el gran culpable, COURTOIS dejó una lección de liderazgo, compañerismo y grandeza. Porque un verdadero capitán no busca culpables: protege a los suyos cuando más lo necesitan. HAY GESTOS QUE VALEN MÁS QUE MIL PALABRAS. 👏❤️

@MiaAI_lab @NVIDIAAI @UnslothAI you updating your recommendation?

Here's the speed you'd get on @NVIDIAAI DGX Spark running the new @UnslothAI Qwen3.6-35b-NVFP4-Fast in a single session 🔥

@sudoingX Now Claude, got the dgx, end goal is local!

let's settle it. what's your actual daily driver, local or api? model + hardware if local. no vibes, no "it depends". just what opens first when you sit down to work.

@TheAhmadOsman What about on dgx spark?

PREDICTION Within the next 18 months, you will be able to host GLM 5.2 equivalent intelligence on an RTX 5090 GPU.

@SoyJuanLuna @morenabeltran10 If he doesn't miss a goal in a world cup match, he stops being Lautaro!

@morenabeltran10 Y se erró un gol abajo del arco. Típico del buen Lautaro

Lautaro entró, la bajó para el 2-2, asistió para el 3-2 y terminó llevándosela al córner con una guapeza digna de potrero. Tremendo su ingreso.

@lightfinisher Its pure envy!

Can’t even enjoy an Argentina win because after every game everybody calls it rigged or robbed. Genuinely tiring. Even the casuals are joining in. It’s like mass campaign against them just because Messi plays for them. I don’t understand the hate to resort to discrediting them. I can understand the banter with stats, titles etc

@JosipFCB93 People blinded by envy and hate! They say its a penalty, he dived once he lost the ball! Shameful tactic! Penaldo would be proud of this one!

The fact that there appears to be millions of people saying the Salah situation is a penalty makes me question if I live in some kind of parallel universe because there is absolutely no way I am seeing this stuff right now

@ArtButSports and there's No.10 too!

love that No. V = #5 = Leandro Paredes and that there are Argentina colors in the top right of the full work:

Composition, No. V (rotated, detail), by Hans Hofmann, 1952, 📸 by @Dan_Mullan

@MlogicGr8Again @dme_363 Ronaldo won't dive better than this 😄😆

@dme_363 The exact same was a penalty to erase Egypt's goal, but this was no foul on Salah Argentina + #FIFA = 🤡

Argentina's man of the match vs Egypt

@GergelyOrosz Normal to happen.

Let me know if this is just me: Noticed someone I know who is very "AI-pilled" and uses agents 24/7 to... start to talk noticeably more like these LLMs write. Eg more heavily using adjectives like "geniune", frequently terms like "the shape of" and many more examples

@SpaceTimeViking beautiful!

Light X Space X Time

@meijer_s You can paste twice to get expanded view.

Can we disable this? It's annoying. I want to be able to edit pasted text.

@iotcoi Whats the recipe, dear sir? Dwarfstar?

Just ran GLM-5.2 on a single GB10 in a ds4 fork. 753B params at 2.4 t/s. My NVMe is screaming in NAND. This isn't inference; it's a weekend fun project. Can you run a frontier LLM on one palm-sized GPU? Yes. Very slowly.

@SpaceTimeViking Love it, will try on mine.

@codengod Yes that’s all I’ve got so everything I make works on a single DGX Spark.

Hot off the press 🔥 AEON vLLM Ultimate → v0.24.0. Link below ⤵️ The ultimate container for AI on NVIDIA DGX Spark — vLLM compiled from source for GB10/sm_121a. 🔓 UNLOCKED (impossible until this release): ▸ DFlash + FP8 KV cache — 2× KV capacity, draft acceptance intact, measured on-device ▸ FP8 KV on vision models — the KV-bandwidth lever is back ▸ DFlash on FlashInfer & Triton backends (was FlashAttention-only) 🛠 Carried ahead of upstream (still-open PRs, merged + validated): ▸ Triton NVFP4 KV cache → up to 3× KV capacity ▸ DFlash sliding-window drafters + prefix-cache stability for DFlash ▸ High-concurrency DFlash — clean through c=64 🧰 Fixes stock v0.24.0 doesn't have: ▸ Gemma-4 tied embeddings (ModelOpt) — stock crashes at load, ours doesn't ▸ Tool calling out of the box (stock --no-deps builds 500) ▸ Unified-memory hardened — no silent KV overcommit on the 121 GB pool 📊 One image, the whole fleet: Gemma-4-26B-A4B · Qwen3.6-27B · Qwen3.6-35B-A3B (hybrid GDN) 536 tok/s @ c16 · 472 tok/s @ c12 · 262K context · image+video · tools + reasoning parsers · full CUDA-graph capture 🐳 docker pull ghcr.io/aeon-7/aeon-vl…

@Thom_Wolf @cerebras why is it so fast? it's sped up or edited?

Most people should probably update their priors on the state of open-source speech-to-speech. It's honestly kind of mind-blowing. We teamed up with @cerebras to build a fully open-source realtime voice demo (models + code) to show what's possible today. Demo : huggingface.co/spaces/smolage… Blog: huggingface.co/blog/cerebras-… Go test it, fork it, tweak it, and impress your friends. video is raw, no cut, no speed-up, first take

@vacchu @ArunKrishnan_ You are barking at the wrong thing.

@ArunKrishnan_ The Starking difference - Parents of our era needed no social validation for thier acts

Children MUST take care of their parents in their old age. This is a hill I am comfortable dying on. Last evening, at a wedding. Appa feels sleepy and tired. Amma wants to talk to people. So I brought him back to the hotel, help him get changed, and get into bed and stayed with him. I thought back to how many days he would have done the same for me duringy childhood. It was now my turn.