


DevSecOps engineer · Kubernetes (CKA/CKAD/KCSA) Local AI on a DGX Spark: benchmarks, recipes, breakages Open to work & collabs · 🇮🇹 in 🇨🇭


This would be really interesting to test on a single Spark (when proper quantization comes out). Could it replace the good old Qwen 3.8-flash-next?
Hy4 Flash is now open weights 🔓 Tencent just put the model from Monday's deleted announcement on Hugging Face: • 220B total parameters, but only ~12B active per token, so it's cheap to run • Reads text and images, with a 1M-token context window • In Tencent's blind test with its own users, it scored 2.79 out of 4 vs 2.81 for Claude Haiku 5.5, at the lowest cost per task of any model tested • About 1/20 the cost per task of full Hy4 • Also out: an FP8 version and the pre-trained base model Tencent Hy License. huggingface.co/tencent/Hy4-f…

Over the past week, I’ve been overwhelmed by the love and support of friends, industry colleagues, and people I've never met. Thank you to everyone who has shown kindness to my family and me. It means the world to me. I came to the United States 53 years ago, and becoming an American citizen is one of the greatest gifts and proudest moments of my life. NVIDIA has been my life’s work. Building it into the company it is today alongside my colleagues has been a privilege and an honor. Nowhere but in America could a company like NVIDIA be possible. I love our country, and I love NVIDIA. I did not live up to the standards expected of everyone at NVIDIA, and I have made serious mistakes in my personal life. I created problems for the Board that I deeply regret. I own my mistakes completely, and I am committed to cooperating with the Board’s independent investigation. Since my departure, I’ve seen an outpouring of calls from NVIDIA’s employees, investors, business partners, and customers for me to return as CEO. Many continue to reach out to me directly as the company navigates complex technology, market, competitive, and geopolitical dynamics. NVIDIA is a once-in-a-generation company. If the Board of Directors decides that my return is in the best interest of our shareholders, our employees, the company, and America, I would be honored to answer these calls and return as CEO. Thank you again to everyone who has shared their good wishes. I am profoundly humbled by your support. To my family: I love you, and I will spend the rest of my life being worthy of yours. Jensen

Jensen is reportedly considering a comeback! 💚 More than ever, our voices matter. Let's show him how many of us stand with him. NVIDIANs, shareholders, and supporters, please take a moment to read the letter and sign it. Thank you! #standwithjensen #bringjensenback standwithjensen.com

To be honest, the performance is a bit disappointing, and the price is a bit off too. Maybe the distillation ban is affecting Chinese models? Waiting for the 27B and Flash.
📢 Meet Qwen4-Max, your new AI teammate for work, creativity, and discovery! Qwen4-Max is built on a new MoE architecture with native vision and a 1M-token context window. Highlights: - Outperforms Qwen3.8-Max with up to ~50% fewer tokens, while average cost per task drops over 60% in our internal evaluation. - Advances real-world productivity through greater proficiency with professional tools, stronger long-horizon execution, and higher-quality deliverables. - Handles real tasks, including 3D modeling, game development, manufacturing, finance, law, healthcare, and frontier scientific discovery. - $1.5/M tokens for input, $4.5/M tokens for output, and $0.075/M tokens for implicit caching — all lower than Qwen3.8-Max, with implicit caching down by 70%. The API is live on QwenCloud. You can also try Qwen4-Max on Qwen Studio, QwenWork and Qoder. Start building today! 🚀 - Blog: qwen.ai/blog?id=qwen4-max - API: qwencloud.com/models/qwen4-m… - Token Plan: qwencloud.com/pricing/token-… More to come! The open weights of smaller Qwen4 models will be released. Stay tuned. 👀

TensorFold 1.2.8 is out, and this was a monster landing. 92 PRs merged, and 19 issues closed since 1.2.7. 🚀 There’s new support for Recursive’s Flint 40B-A2.5B on M5 Macs, experimental DeepSeek-V4.1 on Metal, Qwen3 embeddings and rerankers on CUDA, and OCR for page images on HIP, alongside request cancellation by ID and tighter server limits. Among the measured gains; - Qwen3.8-27B MLX 8-bit cold prompts are about 33% faster on an M5 Ultra. - Flash Next CUDA prefill at TP=2 on 2K to 32K prompts is 26.5% to 26.8% faster on two GB10s and 30.9% to 33.9% faster on two RTX PRO 6000s, with the same output tokens in those tests. Massive thanks to everyone contributing code, testing on their hardware and reporting the things that need fixing, especially @3utsu @arminxnova @Bizuayeu @defilan @devopsfortunato @unigilby @javierbits @teleogen @petruspennanen @Shoestring_Dev @woolsey, and on GitHub AbrahamCete, BobClawblaw, dinadur, EugeneClaw, jschmied, kotevcode, Mirrdhyn, ppardus, SergiioB and shanedertrain; I really appreciate the time you all put into this. @petruspennanen @volatilemarkts @MiaAI_lab @plotarmordev @jayleaton tensorfold.dev


Well said! Long live open source!
This is a difficult moment. It’s hard not to look at the journey this company has taken and not be filled with respect for what Jensen built. The spirit of the company remains strong: "mission is the boss”. That said, as NVIDIA finds its way forward, I want one thing to be clear: NVIDIA's contributions to open source are unwavering. Jensen has been vocal about this, and that commitment runs deep in NVIDIA's people and culture. We all understand the importance of open source. NVIDIA is a huge supporter of open source because we understand something. If you want to diffuse technology, you do so through an ecosystem, not a single company or entity. No one owns the internet, no one owns servers, and no one owns operating systems… but through open source, an ecosystem of closed and open innovation flourishes. When we talk about delivering agents safely and securely, that too will take place because of open source. For example, ensuring people have the best models to secure themselves at machine speed unburdened by arbitrary guardrails decided behind closed doors. Open source & AI are often discussed together, but it’s worth reminding, open source predates AI. It is a mechanism for sharing ideas inspired by the hippies at computer labs of California in the 1960s. Before AI, the largest shift I lived through was the internet. It was remarkable seeing the fight for open source take place there. Between browsers, runtimes, net neutrality, etc. What came as a result was HTML5 and an extremely vibrant ecosystem that built the internet as we know it today. NVIDIA and its partners hold an incredibly important role in delivering this technology. NVIDIA contributes to every layer of the stack. Runtimes, models, harnesses, programming languages, frameworks, etc. None of that changes. Long live open source.

TensorFold 1.2.7 is out. 🔥 A big fix for long conversations on Metal: Qwen3.8-27B now reuses its history even when the client drops earlier reasoning. In the measured M3 Ultra test with 34k history and a 5 GiB prompt cache, follow-up time to first token dropped from about 85s to 0.3s. Same replies in that test. Also in this release: • Responses API over WebSocket. • LightOnOCR-3 0.8B and 4B image chat on CUDA. • Opt-in GLM conversation reuse at turn boundaries on CUDA. • Prompt-cache and disk-tier stats in the dashboard. • Targeted speed and memory improvements across Metal, CUDA and HIP. • Fixes for GPU builds, malformed GLM tool calls, damaged JPEGs and disk-cache recovery. Massive thanks to everyone contributing code, testing on their hardware and reporting issues. I really appreciate the time you put into this. @andersjw_ @Bizuayeu @defilan @devopsfortunato @unigilby @kelchm @teleogen @petruspennanen And on GitHub: AbrahamCete, BobClawblaw, ciprianveg, garvrao80, jschmied, laidick, millaguie, Mirrdhyn, mrv777, tacosta, volfco and wojo. Thank you too to the tester who reported the JPEG fault privately. @volatilemarkts @MiaAI_lab @petruspennanen @plotarmordev

Proud to have contributed to this release 💪 Never stop pushing 🚀🚀
TensorFold 1.2.6 is out. 🚀 The big one here is the prompt cache under multi-agent load. In the 16-agent Qwen3.8-27B + DFlash2 test, median follow-up time to first token dropped: • M3 Ultra: 546s → 15.9s • GB10: 105s → 25.8s Each agent started with a ~17k-token prompt. Same replies throughout that test. Also in this release: • Flash Next cold prompts finish 29 to 41% sooner on HIP, measured on the Radeon 8060S. • GLM-5.3-Flash single-stream code decoding is 32% faster on two GB10s at TP=2. • Opt-in DSpine-b16 drafting from Tencent for Qwen3.8-27B on CUDA. • An image-generation API for image-capable engines, local Unix sockets and tool-call fixes. One upgrade note: kept prompt states need a fresh prefill after updating. Massive thanks to everyone contributing code, testing on their hardware and reporting the things that need fixing. I really appreciate the time you all put into this. @andersjw_ @Bizuayeu @defilan @devopsfortunato @unigilby @jvr0x @teleogen And on GitHub: abm9111, AbrahamCete, akol1, BobClawblaw, chadhurley25075-png, davidtai, ElderOrb, jschmied, kotevcode, Mirrdhyn, mrv777, ppardus, robertung88 and volfco. @petruspennanen @volatilemarkts @plotarmordev @MiaAI_lab tensorfold.dev

@TandemLLM 0.15.3 is out and 11x FASTER. Using qwen3.8 flash next, you can run 5 chats in parallel in the same dgx spark, everyone is above 200k context. when memory ran short, a long chat used to get cancelled or had to read its whole context again. now an idle chat is written to the nvme before its pages go. when it comes back, the rows are read in from disk and only the new tokens get prefilled. on one dgx spark with flash-next 3.05: five chats of 221k to 225k tokens, kv pool has room for four. a turn on each chat took 32 to 68 s with the spill on, 659 to 894 s with it off. also in this release, from contributors on github: @AgentSparko (long-context prefill, drafting past 32k, http fixes), @0xWhiteMage (ada kernels, every gguf quant type on NemotronH), salmanarshad321 (engine context in usage), @devopsfortunato (slot count on every weight card). Thank you. github.com/0xBakeer/TandemLL…

Qwen 4 today
Qwen 4 today

TensorFold 1.2.5 is out. 🚀 This release brings: • TensorFold Web, a local browser app for chat, models, learning, benchmarks and more. • LFM2.5-24B-A2B on CUDA. • Nemotron Parse and LightOnOCR-3 on HIP. • Clef's joint schema head on CUDA. Qwen3.8-27B MLX 8-bit on M5 Ultra also gets faster. In the measured test, cold prompts were 18% faster and decode 7.6% faster, with the same tokens and prompt state. There's lower memory use for Qwen CUDA streams, better conversation-cache handling and a fix for PDF rendering crashes on macOS too. A massive thank you to everyone contributing code, testing builds and reporting problems. These releases take a lot of work, and I really appreciate the time people put into them. @Bizuayeu @devopsfortunato @unigilby @teleogen @dubsthedev @kalin_swiss @Q4_K_M And on GitHub: AbrahamCete, gordonhatusupy, jschmied, mrv777 and yguanrong. Special credit to Stephane on GitHub for TensorFold Studio, which TensorFold Web builds on. @volatilemarkts @plotarmordev @MiaAI_lab @petruspennanen TensorFold.dev

TensorFold 1.2.4 is out. 🚀 Two of the measured wins in this release: • DeepSeek-V4.1 restored a saved 319K-token conversation from NVMe in 1.2 to 1.4 seconds, versus 155 to 178 seconds, on two GB10s at CUDA TP=2 with replay prefill. • Flash Next on M5 Ultra: two overlapping clients finished in 47 seconds instead of 127 in the MLX 4-bit parallel-serving test. Also new: Gemma 4 E4B on CUDA, LFM3-1.7B text on Metal and Clef-Flash decisions on HIP. The HIP runtime now builds and loads on Windows too. There are cold-loading improvements, tool-call fixes and clearer API errors. One upgrade note: existing disk-cached prompt states need a fresh prefill after updating. Thank you to everyone behind the code, testing and bug reports. The community puts an incredible amount of work into these releases. @arminxnova @atasrb @Bizuayeu @BlivionIaG @codengod @defilan @devopsfortunato @unigilby @jayleaton @jishvenugopal @Wi11y010 @teleogen @lasse_ai @petruspennanen @dubsthedev @woolsey And on GitHub: AbrahamCete, AdrianBinDC, BobClawblaw, ElderOrb, gordonhatusupy, goshi, jschmied, laidick, ppardus, shuuul and tolewis. @petruspennanen @plotarmordev @volatilemarkts @MiaAI_lab tensorfold.dev

Everyone says automation kills quality. 2 weeks automating my open source workflow says otherwise. Last week: 24 PRs, mostly Tensorfold with @ashxhart. Proud of MOSS image adoption. Release soon, right? :D I've always been the first to advocate for open source. But I struggled to contribute week after week. Two weeks ago I started polishing my workflow around one rule: offload everything I can from me, without lowering what I ship. I found my place in @DeepSeekHarness. It automates most of the loop. Plugins, task boards, subagent-driven development. Full breakdown soon. What that gave me last week: - 24 PRs on projects I believe in - Tensorfold: pushed MOSS image adoption forward, release coming - A smaller contribution to Tanderllm @0xBakeer Automation only costs quality if you skip verification. Check every step and it doesn't. Local AI gets better when more people ship to it. Pick a project you believe in and contribute.

We're giving away a Hogwarts Legacy custom wrapped NVIDIA GeForce RTX 5090 to celebrate its official release with DLSS 5 🪄 To enter: 🟢 Share this post 🟢 Comment #RTXON Terms and Conditions: nvidia.com/en-us/geforce/con…