Search: “nvidia”

Running Qwen 3.8 next on 16vram+32ram - A useful/fun post for the gpu poors

Hello Reddit. Posting this for fun. I thought it was a lonely and silly journey to set up Qwen 3.8 Next on a system that doesn't really run it properly—it was a challenge that might help the community

14 Sept 2026

Human

Content type

All Blog X Post Paper YouTube Reddit GitHub

Clear filters

Model Releases

UkisAI Swift-Qwen3.8-27B / -58.3% thinking, x1.95 speed while keeping the accuracy of xhigh

DGX agent

Hi everybody, we post-trained Qwen 3.8 27B to be more efficient by figuring out which tokens were linked to overthinking and penalizing them without 'attacking' the reasoning length directly then fixe

model-releasesr-localllama

14 Sept 2026

Agents

Vera Rubin NVL72 Agentic Inference: 67x better Performance per Dollar

**Summary for AI knowledge base** Vera Rubin NVL72 is a co‑designed platform combining the Rubin GPU, Vera CPU, NVLink 6 Switch, ConnectX‑9, BlueField‑4, and Spectrum‑6, targeting agentic inference wo

agentssemianalysis

14 Sept 2026

What are the current best retail GPUs for max VRAM at a reasonable price?

I am considering dumping my ChatGPT Plus subscription and go full local, but to do so I would first need to reach a decent result for quality (and reasonable speed). My 4090 fried itself out of nowher

14 Sept 2026

3k$ 128GB VRAM + 256GB RAM DDR4 Server

I finished my home inference server. First I tried Lenovo p620 workstation and while it’s a good value overall it pissed me off with a ton of proprietary Lenovo shit to deal with and I return it in th

13 Sept 2026

Wild 24 hours for AI and lots of different proposals have been made. TLDR; the only tangible new fact is that OpenAI and Anthropic are goi…

model-releaseselon-musk--x

Wild 24 hours for AI and lots of different proposals have been made. TLDR; the only *tangible* new fact is that OpenAI and Anthropic are going to have embedded 3rd party evaluators from unknown organi

13 Sept 2026

For those of you forced to only use open models from Western labs in production, what are you deploying?

First off, I know that GLM, Qwen, and DeepSeek absolutely dominate in terms of SOTA Open Source models, and that’s what I use in my personal projects and for school, however, I’m also responsible for

12 Sept 2026

Safety

It's now clear that: - alignment is critical to making AI safe - alignment won't be solved behind the closed doors of a handful of frontier …

It's now clear that: - alignment is critical to making AI safe - alignment won't be solved behind the closed doors of a handful of frontier labs So we're taking two steps: - launching the Open Alignme

safetyclem-delangue--x

12 Sept 2026

THE CURE IS THE PITCH Dario’s 'Pacing' Essay Is Quiet-Period Illegal Stock Promotion Wrapped in Regulatory Capture Dario Amodei's 'We Must P…

model-releasesyann-lecun--x

THE CURE IS THE PITCH Dario’s 'Pacing' Essay Is Quiet-Period Illegal Stock Promotion Wrapped in Regulatory Capture Dario Amodei's 'We Must Pace the Frontier' is not a safety paper. It is a pre-roadsho

12 Sept 2026

7900 XTX + 32/64GB RAM for Qwen 3.8 Flash Next?

Planning to build a PC mainly for local LLMs/coding agents. I keep seeing 3090 + Qwen 3.8 Flash Next benchmarks, but could not find enough info for the 7900 XTX 24GB. 3090s are hard to find where I li

CodeFinetuner: Fine-tune a local code autocomplete model on your own codebase

Hi everyone, I was interested in learning LoRA fine-tuning, and ended up building CodeFinetuner over the past few months, a full pipeline that fine-tunes a small code autocomplete model (e.g. Qwen2.5-

Does YOLO26 Truly Offer Advantages Over Its Predecessors for Edge Deployment? A Benchmark Study in Aquaculture

model-releasesarxiv-cs-cv

arXiv:2607.09835v2 Announce Type: replace Abstract: The recently introduced YOLO26 architecture incorporates NMS-free end-to-end inference and is optimized for deployment on resource-constrained CPU-b

Improving 5G AI-RAN MCS Selection by Predicting Retransmissions

model-releasesarxiv-cs-ai

arXiv:2609.09324v1 Announce Type: cross Abstract: Link Adaptation (LA) in 5G NR is inherently reactive, relying on channel measurements and HARQ feedback that may become quickly obsolete when the chan

Introducing Fugu Max and Fugu Ultra v2: the next evolution of Sakana Fugu’s multi-agent orchestration system. Try: https://sakana.ai/fugu Bl…

model-releasesdavid-ha--x

Introducing Fugu Max and Fugu Ultra v2: the next evolution of Sakana Fugu’s multi-agent orchestration system. Try: https://sakana.ai/fugu Blog: https://sakana.ai/fugu-max-release/ The frontier that ac

Is a ZIMA Board 2 + RTX 2000 ADA the cheapest path to a decent Qwen-3.8 27b self-contained endpoint?

I just watched a YouTube from Luke’s Dev Lab where he literally just plugged a RTX 2000 ADA Into the side of the Zima Board 2’s PCIE socket and it just friggin worked and had great token speed despite

Orukeet, new ASR model based on Parakeet

I haven't seen this mentioned yet, so I thought it deserves a post. I was trying out OpenWhispr when this model came up as the recommendation. So I don't have personal experience yet, but it's suppose

Phase-Decoupled, Model-Calibrated Power Control for Disaggregated LLM Serving

arXiv:2609.11133v1 Announce Type: new Abstract: Datacenter GPU power is the binding constraint on LLM serving capacity, and production serving has shifted to prefill/decode (PD) disaggregation. Deploy

Reification as a Transferable Vocabulary: Zero-Shot Link Prediction with Vanilla GNNs

arXiv:2609.11347v1 Announce Type: new Abstract: Knowledge graph foundation models such as ULTRA achieve zero-shot link prediction on unseen graphs through dedicated architectures that hard-code a tran

This is why we need open-source harnesses + local models

i've been thinking about this more after trying different agent setups. the model isn't the only thing that determines how well an agent performs. The harness around the model matters a lot too. With

Characterizing Contention-Induced Reliability Collapse in KV-Cache Timing Side Channels for Multi-Tenant LLM Serving

model-releasesarxiv-cs-ai

arXiv:2609.06853v1 Announce Type: cross Abstract: Shared key--value (KV) cache reuse improves large language model (LLM) serving, but it can also create a timing side channel that reveals whether a pr

Safety

Learning Terrain-Adaptive Humanoid Locomotion on Granular Terrain

arXiv:2609.10286v1 Announce Type: new Abstract: Humanoid locomotion on granular terrain remains a significant challenge due to its complex foot-terrain interaction dynamics that are difficult to model

safetyarxiv-cs-ro

RedKnot-MLA: Multi-Head Offline-Online Reuse for DeepSeek-V4 Long-Context Serving

model-releasesarxiv-cs-ai

arXiv:2609.07008v1 Announce Type: new Abstract: Multi-head latent attention (MLA) exposes many logical query heads through one packed latent KV stream. This representation is memory efficient, but it

Sources: Sam Altman told OpenAI employees that the company is considering slowing cutting-edge AI development, and he hopes other AI companies will do the same (Bloomberg)

Bloomberg: Sources: Sam Altman told OpenAI employees that the company is considering slowing cutting-edge AI development, and he hopes other AI companies will do the same — OpenAI is considering slowi

local-aitechmeme

A Measurement Study of LLM Inference Trade-offs Across Edge Continuum Hardware

model-releasesarxiv-cs-lg

arXiv:2609.08307v1 Announce Type: cross Abstract: Large language models (LLMs) are increasingly used as backends for intelligent web services, but serving them across the edge continuum requires balan

CoSA: Correlation-Guided Change A ttention with Learnable Residual Gating for Remote Sensing Change Detection

model-releasesarxiv-cs-cv

arXiv:2609.08914v1 Announce Type: new Abstract: Pixel-level annotation of fixed traffic-camera imagery is expensive, while crosswalk models trained from street-level imagery face a substantial viewpoi

FALCON-S: Fixed-wing ground-effect Aerodynamics Simulator and Flight Control Learning Suite

arXiv:2609.06046v1 Announce Type: cross Abstract: We present a modular, high-fidelity simulation framework for the development and benchmarking of flight control strategies in fixed-wing aerial robots

Safety

If they truly believe the risk is that high, and I think they do, then we need 100x more research and understanding of it. That requires the…

If they truly believe the risk is that high, and I think they do, then we need 100x more research and understanding of it. That requires them to openly share their models, datasets, training code, and

safetyclem-delangue--x

MolParser-Mobile: Ultrafast OCSR System for Large-Scale Chemical Literature Mining

arXiv:2609.05807v1 Announce Type: new Abstract: Optical Chemical Structure Recognition (OCSR) is a fundamental component of chemical literature mining, enabling molecular database construction, reacti

hardwarearxiv-cs-cv

RGB Input Pipelines: Throughput, GPU Memory, and Transformation Coverage

model-releasesarxiv-cs-cv

arXiv:2609.06635v1 Announce Type: cross Abstract: An image-augmentation pipeline must deliver a complete batch before a model can use it. We compare seven input paths from five libraries, starting wit

RoLA: Rotary-Positioned Low-Rank Linear Attention for Efficient Diffusion Transformers

arXiv:2609.06712v1 Announce Type: new Abstract: Diffusion Transformers (DiTs) achieve strong video generation quality, but their dense spatiotemporal self-attention scales quadratically with sequence

hardwarearxiv-cs-cv

Scalability Analysis of Distributed Kolmogorov-Arnold Network Training on High-Performance Computing Systems

model-releasesarxiv-cs-lg

arXiv:2609.07740v1 Announce Type: cross Abstract: Kolmogorov-Arnold Networks (KANs) replace the fixed activation functions and linear weights of Multi-Layer Perceptrons (MLPs) with learnable univariat

TBDub: Production-Oriented Visual Dubbing

arXiv:2609.06144v1 Announce Type: new Abstract: Visual dubbing must synchronize mouth motion with replacement speech while preserving identity, appearance, and temporal consistency. Although X-Dub pro

hardwarearxiv-cs-cv

GTIG AI Threat Tracker: From Prompting to Autonomy – The Evolution of Adversarial AI

model-releasesgoogle-cloud-ai

Executive Summary Since the release of our May 2026 report detailing adversarial misuse of artificial intelligence (AI), Google Threat Intelligence Group (GTIG) has observed forward leaning adversarie

Open-source AI developer Mistral closes €3B funding round

model-releasessiliconangle

French artificial intelligence lab Mistral AI SAS today announced that it has raised €3 billion, or about $3.49 billion, in funding. Samsung Electronics Co. Ltd. led the investment. It was joined by m

Other investors in the round include Benchmark, Bessemer, Kleiner Perkins, Greylock, Lightspeed, Altimeter, Bond Capital, Meritech, Atreides…

model-releasescognition-ai--x

Other investors in the round include Benchmark, Bessemer, Kleiner Perkins, Greylock, Lightspeed, Altimeter, Bond Capital, Meritech, Atreides, Valor, T. Rowe Price, Lux, 8VC, D1, 137 Ventures, DST, Pos

Qwen3.8-Flash-Next in llama.cpp vs SGLang vs FreeToken: 35s vs 258s to first token at full context. My findings on new PRs coming to engines.

Hey guys, After my CPU-only to 96GB VRAM test, I tested Qwen3.8-Flash-Next across llama.cpp, SGLang and FreeToken on the same workstation. This time I wanted to see what changes when you keep the hard

A Sim-to-Real Study of Surface-Code Decoder Benchmarking

arXiv:2609.04557v1 Announce Type: cross Abstract: Quantum error-correction decoders are typically benchmarked against synthetic circuit-level noise, under the assumption that a decoder's ranking under

ExLlamaV3 is underrated

I moght get shit on for posting this but, I feel like i don't see this being talked enough and it feels like such a waste of a good piece of software. Exl3 is incredible, albeit only if you have NVIDI

I Ran 112 MiniMax H3 Tests on an RTX 3090 — Searching for the “Golden” Settings

hardwarer-stablediffusion

I ran a 112-case MiniMax H3 quality sweep on an RTX 3090 last night trying to find the best combos . Kept the prompt, seed, reference image, RefMods, Sage setup, sampler, and scheduler stayed constant

LayerStoRm open-source expert streaming: 1M context GLM-5.3-Flash [UD-Q4_K_XL] at 24.5 tok/s @8k on just 2× RTX 5090 + 2× RTX 5080 (186 GiB MoE on 96 GB VRAM)

LayerStoRm: GLM-5.3-Flash UD-Q4_K_XL (186 GiB) at 1M context on 2× RTX 5090 + 2× RTX 5080 (96 GB VRAM total) using RAM for the pinned experts. LayerStoRm is a (still experimental) MIT-licensed continu

little-coder vs just Pi

Hi, I use hermes as a harness, and I am pleased with it, but sometimes Hermes's context size is a bit too much for my system, so I wanted to delegate or straight up use a lighter option for coding tas

hardwarer-localllama

Measuring proximity to standard planes during fetal brain ultrasound scanning

arXiv:2404.07124v2 Announce Type: replace-cross Abstract: This paper presents a pipeline designed to bring ultrasound (US) plane pose estimation closer to clinical use, demonstrating the feasibility o

hardwarearxiv-cs-ai

48 tg/s 440 prefill on my grandma's cluster (2xP40) (sort of)

TL;DR: switching KV cache to f16 may give a boost in speed if using MTP and ngrams. I have a self-built 'AI mega-cluster' with 2x P40s on a cheap Chinese motherboard and a Xeon CPU (around $1,100 to b

6 Sept 2026

I built an LLM benchmark harness that lets you browse and compare how models answered each question

Hello everyone. I felt current LLM benchmark harnesses hand you headline numbers but offer no tooling to see how models actually answered each question (they dump everything to a JSONL or Parquet file

6 Sept 2026

vllm + p2p driver hack + qwen 3.8 27B vs llamacpp + qwen flash next ?

Hi everyone I'm running four rtx 4090, 64GB ram, on a threadripper pro motherboard so all PCIe x16 ports, as a homelab machine for coding. I was migrating from vllm + qwen 3.8 27B (fp8+256k kv cache)

6 Sept 2026

My only real use case for a local AI use is document management, how much VRAM do I realistically need for a good experience?

I just want to use paperless-ai and be able to ask questions relative to it. Bonus points if I could use it with home assistant but that's not the focus. I just can't see needing a 32 GB VRAM GPU for

5 Sept 2026

NInfer vs llama.cpp vs vLLM: quality + speed comparison for Qwen3.8-27B NVFP4 on RTX 5090

I've been running Qwen3.8-27B as a local inference server for a production content intelligence pipeline (HVAC industry stuff, lots of long-context retrieval and structured extraction). I have been wa

5 Sept 2026

BASP: Communication-Efficient Batch-Aware Sequence Parallelism for LLM Training