I build models for the DGX Spark ecosystem and contribute to open source AI (TensorFold, Hermes). If you are passionate about DGX Spark, let’s connect.
Inkling on 2 x DGX Sparks conducting a thorough security review (long-running agentic tasks) on itself. Context window set to 256k. It is fast! Notice how fast it spins up the sub agent.
Link to port / recipe in comments.
TensorFold 1.3.3 is out 🚀
124 PR's merged and 34 Issues resolved in 24h
GLM-5.3-Flash went from 151.51 to 172.39 tok/s in median drafted decode after a 1,040,384-token prompt on four GB10s at TP=4. That’s 13.8% faster on CUDA with the same output tokens in that test. Thanks to @MiaAI_lab for the kernel behind that gain.
- On Metal, GLM-5.3-Flash gets DFlash2 drafting. On a Mac without tensor units, GLM cold prompts from 2K to 64K took about 5–6% less time in the release tests.
- MOSS-Image generates pictures on CUDA through /v1/images/generations. GLM-OCR runs on HIP, and Idol gets its image encoder on Metal.
- DeepSeek-V4.1 gets structured output in the dsv41-cuda source build. Long CUDA prompts now prefill in chunks alongside running replies by default.
- LFM2-MoE, LFM3, MiniMax-M3 and Qwen3.5-2B on Metal can fill new prompts between other streams’ decode rounds.
- The server adds per-request deadlines, a prompt-cache flush endpoint, and deferred and namespaced tools.
Massive thanks to everyone who sent patches, tested on their own hardware, and filed repros. @AIFlow_ML@bikepresentball@Bizuayeu@devopsfortunato@unigilby@javierbits@jayleaton@teleogen@niklaslenz_ai@plotarmordev@woolsey@petruspennanen@andersjw_@shantanugoel@dennis_rye, and on GitHub AbrahamCete, BadAd84, Ch0mpaa, christopherowen, CodingEmil, eXsoR65, Flyinpiz-dev, garvrao80, hobbesd, jschmied, kotevcode, MaverickKB, millaguie, nuk3s, SergiioB, supernovae, tacosta and ultradaoto. I really appreciate the time you put into this.
@plotarmordev@jayleaton@petruspennanen@MiaAI_lab@volatilemarkts
@AIFlow_ML@flaviusburca Very true. Single-stream is seldom the benchmarking answer when running multiple agents against the Sparks or other architectures.
@AIFlow_ML I don't see a reason to buy the RTX at all. The clustering properties and linear scalability of DGX Spark is exactly what makes the architecture attractive.
Pretty distracted right now.
It appears OpenRouter may have had inbound requests siphoned by a malicious actor. This happened in August on their MiniMax models. This is why I stopped using open weight models on OpenRouter and moving workloads to my Sparks. This is why we need local AI.
@OpenRouter
@1337hero I cannot wait to get my hands on these weights. GLM 5.3f did a huge amount of good for our ecosystem. It was fast, multimodal, and good quality despite quantisation. It was also easy to engineer. Let’s hope the same principles apply this time.