Search: “r-localllama”

Yoyo I'm back with updates to the fastest inference engine with minimal quality loss for Qwen3.8-27B. The last few weeks I've been optimizing decode speed and I don't think it can be pushed further, u

Kaitchup just posted results of his benchmarks for Qwen3.8 27B for quants from different labs, Q4 to Q1, . All the details are hidden behind the paywall, but high level result is visible and looks lik

Hello! I've just uploaded a recent checkpoint of my model trained from scratch: https://huggingface.co/piotr-ai/polanka_3.7b_exp_wip_260901 It was pre-trained, mid-trained, and fine-tuned on a single

I was browsing HF for small LLMs and run into this model. It does not seem to be a fine tune - the model has its own architecture. https://huggingface.co/XHToken/Spark-X2.5-1.7B https://huggingface.co

I only started using some vLLM forks recently in a 4 x 48GB 4090 system. DS4F - ~5000pp/180tg (DSpark) Qwen3.8 Flash next - ~7500pp/135tg (MTP) This is amazing, like having the API in my house. But it

I am absolutely blown away. Yes my setup is crap but the fact that it managed to do this in a single take is unbelievable (and I'm a developer). Hardware used: - Windows PC with 4070ti (12GB VRAM, 32G

if you are on Qwen Flash Next make sure to update your build often https://preview.redd.it/zoq3j9gnjvmh1.jpg?width=900&format=pjpg&auto=webp&s=bb21bc405696df94fd5ce38c22f4ca6dc596135f merged already (

I posted a new gfx906 based llama.cpp fork a few days ago. One of the main points of critique was that i did not provide sufficient numbers for the gains to be achieved. -- TL;DR: After switching our

Vellium is an open-source, local-first desktop app for AI chat, character roleplay and long-form writing. Recent updates have focused mostly on making local voice and model setups easier to use. Live

I used a famous Simon Willison's pelican riding a bicycle prompt on the biggest local LLMs that can run on 128GB Apple Silicon. U used quantizations by Unsloth. Qwen3.8 Flash-Next gives a lot of detai

I've made a modification of llama.cpp MTP for people that want to run models like QWEN 27B on 16GB and similar setup, the focus is reducing the memory cost of MTP allowing more speed for less ctx cost

I haven't had a chance to test it yet, but it looks very promising. It seems to speed up MTP for MoE models across different draft widths (especially greater than 1). Check the benchmarks. submitted b

I'm very happy with Qwen-3.8-27B, I run it on my work machine and switched for major part of real coding tasks from cloud subscriptions to it. I use one dedicated headless RTX 3090, and get maybe 1000

According to this article Samsung has locked up the 70% of it's future ram production in contracts to companies like Microsoft, Google, and Nvidia. Everyone knows this is driving the ram price increas

I keep seeing demos of AI agents building scenes in Blender through BlenderMCP, so I tried it myself. I ran both models locally for this and picked the GLM 5.3 family(Q4 quant) because videos of it do

TL;DR: this paper proposes a method to fix hallucination rates to very low levels or zero by disabling neurons which contribute to hallucination. This discovery has been out for a while now, but it ha

Hi all, I have recently been experimenting with different LLM set ups and after everyone was raving about how good Qwen 3.8 27b was, I was inspired to try and get it deploying. After some battling wit

Just ordered a TB5 enclosure plus 4TB SSD to prepare for M5 Ultra 512GB. I am planning to download the following models in advance. Would like to ask the M3 Ultra 512GB owners whether the quants are t

Repo: https://github.com/Saren-Arterius/qwen3.8-Flash-DGX-AutoRound I forked from blazux/qwen3.8-Flash-DGX and stole all the ideas from previous qwen 3.5 122b recipes, and created a hybrid model with

When using kvarn quants at low context depth tg speed is similar to llama.cpp on and equivalent qx_x quant. However, as context depth grows kvarn tanks your tg speed. This fork optimises kvarn to have

Hello! I was looking into quantizing models and i saw how Hy4 was shrunk from 1.5 TB to 200GB with high retention in benchmarks (98% i think). I was wondering if: a) it would be worth it to attempt th

Hey guys, I tested Qwen3.8 Flash with llama.cpp from CPU-only to the full 96GB of my RTX PRO 6000. Short version: CPU-only reached 8.34 tok/s at a 2K prompt Full 96GB reached 109.07 tok/s At 245K cont

## Qwen3.8-Flash-Next-NVFP4 (inferact) vs Qwen3.8-27B-FP8 (qwen) Slammed with work and no time to pretty this up. Qwen wrote most of this but I checked the data. All tests done on the same rig, same p

Mac Studio, 96GB unified memory (M3 Ultra). I want the largest usable Qwen3.8-Flash-Next setup, and I'd rather not burn 100GB of bandwidth on the wrong download. Here's my math. Please tell me which p

Stumbled around a post about optimizing new Qwen up to 178t/s with a patched version of sglang : https://github.com/jpezzulli/sglang-rtxpro6000 I managed to reproduce results (kudos to jpezzulli, whom

TL;DR: Ninfer/DS4 but for RDNA4 Highly custom kernels built for RDNA4, applied to vLLM-Radiance to greatly improve Qwen3.8 Flash Next speeds. This is mostly for dual R9700s with preferably 48GB of RAM

TL;DR - yeah it works, no wild skills or patching required. 37-50 tg/s when generating text, spiking over 60tg/s when writing code and mtp hits hard. Your results can be better if you have a better mo

Interesting new paper from Alexia Jolicoeur-Martineau (of Tiny Recursive Model fame) and collaborators. They seem to be able to replace quadratic attention with sliding window attention + attention si

SlopTV: a YouTube live stream where the chat writes the programming. You type 'capybara dj underwater rave', an LLM inflates it into a 400-word structured video prompt, one of my 5090s renders 15 seco

https://huggingface.co/Nanbeige/Nanbeige4.2-3B-DSpark 4b model for the gpu poor that I think is stronger than qwen 3.5 9b now faster! I was getting roughly 35 t/s AR which was this models downside, it

Sori-1B is a 1B-parameter audio-language model built by a single SNU researcher whose core claim to fame is that its decoder is trained entirely from scratch on audio-paired text — no text-only pretra

A hallucinated event, Crono awakens in his modest bedroom of 2095... I am using local models since 2025 January. My daily driver is qwen 3.6 35B A3B, which works pretty well for coding tasks, but I al

With b10726, the default --lazy-mode change keeps the 51B-parameter PLE n-gram embedding table of Qwen 3.8 Flash Next on disk: it is mmap'd and its rows are read on demand during inference, even with

I used Claude Code to help write this pipeline - Gemma4 to turn an idea into a short story, then IndexTTS 2.5 running on an NVIDIA 3080 to use voices from LibreVox and VTCK voices to pin as character

Hi All, I am currently contemplating an upgrade from my 5060ti 16gb. I am getting ~40t/s with 130k context on Qwen 3.8 IQ3_S HF quant. I am running llama.cpp on linux. Objective is to increase context

So, ideally for this thread we exclude the ones that everyone on here is already well aware of and discussing on here a lot, like N-gram, quantization improvements, MTP, D-flash, and D-spark, since th

When using two GPU with llama.cpp with --split-mode layer you usually provide a --tensor-split value to fit the proportion of the allocation on each GPU, say you have 16GB + 12GB something like --tens

I'm running Muse Glimmer 30B EXL3-SC 3.00bpw H4, fully resident on my 12GB VRAM GPU at 100K context with Q8_O KV cache. It's a joy to use a dense 30B model at this size and still get 30 tok/s on a VRA

(not written by Claude, all errors and crappy text are result of too little coffee on a Sunday morning ;) Our home server is a 2018 Thinkstation P520, bought for about 600eur in 2023. It's been upgrad

Hey guys I have been working on a open source LLM development repo. Anyone can train any amount of parameter against any arch (GQA, MLA, Dense...) When I tried to implement MLA I read the technical re

I’ve been experimenting with MiniMax H3 and finally have video generation working in TensorSharp. TensorSharp started primarily as a local GGUF/LLM inference engine, so getting a video-generation pipe

Since I spent the time to figure it out and it is not like it will make me any money ever. I think I'd share with you all what I managed to cobble together. First and foremost here the relevant bits a

The intelligence in my context is an aggregate index, I called the Agentic Coding Index, across most relevant agentic coding benchmarks: SWE-bench Pro, DeepSWE v1.1, Terminal-Bench (v4, v3, v2.1), Cod

I built SpeakoFlow Mini, an Apache-2.0 fine-tune of Qwen3.5-0.8B for dictation cleanup. It is not a chat model or a general rewriter. It takes speech-to-text output, applies corrections the speaker ac

I see this kind of approach quite often with local AI models. For example, when optimizing models for an RTX 3090, people sometimes convert them from FP8 to INT8 so they can run efficiently on Ampere

I am doing a lot of translation work with different languages, and German is just exceptionally well done by Qwen 3.8 27B. It is lengths ahead of GPT-5.6 and even Fable 5. With the frontier models I o