Crowd-v1
Crowd-v1 Trains 13B Tokens in a Week: Open-Source Collaboration Faces Its Test
Crowd-v1 is a community-trained 1B-parameter open model. Its 85 contributors processed 13B tokens in one week, but Qwen3-0.6B parity remains unverifie
Back to home
Qwen3
30 articles tagged with this topic
Tencent
Tencent’s DSpine Cuts LLM Inference Time, Showing Draft Models Still Have Room
Tencent’s DSpine injects prior-position information across model layers, improving parallel draft coherence and inference throughput on Qwen3.
D3D
D3D Cuts LLM Inference Time by Up to 7.48x, but Must Escape the Paper
Fudan University's D3D parallelizes draft generation and target-model verification, delivering average Qwen3 speedups of about 5x and a peak of 7.48x.
Cactus Compute
Mirage Brings Phone Vision AI Down to 22MB, Prioritizing Efficiency
Cactus Compute's 22MB Mirage runs vision tasks in 74ms on a Samsung Galaxy S26, proving efficient on-device AI while exposing accuracy gaps.
SFT
SFT Makes Base Models Conversational, but Data Quality Sets the Ceiling
SFT turns base models into conversational models, yet data quality, consistency, and cost matter more than parameter count.
whatmodel
32GB Mac matches GPT-5: Open-source leaderboard kills the cloud-only myth
Developer Jake Manger launched whatmodel.sh — a leaderboard that maps your RAM and GPU to runnable AI models. A 32GB Mac now matches GPT-5 scores.
Qwen
Can Qwen 3 27B Reverse Engineer Games? A Reddit Post Tests Local AI's Ceiling
Reddit user asks if Qwen 3 27B can reverse engineer software. The question itself is minor—but it reveals local AI's new frontier.
Gemma4
Local LLMs Now Deliver B+ Code — A 20-Year Engineer Tests Gemma4 vs Qwen3
A 20-year engineer tested Gemma4-31B and Qwen3.8-27B on real coding tasks: B+ and B-. Open-source local AI is past the demo stage.
Qwen3
Single 3090, 150K Context: Qwen 27B Hits a Local LLM Consumer Inflection Point
Reddit user runs Qwen 27B on a single RTX 3090 + 32GB RAM at 150K context, ~60 tokens/sec. Local LLM demand is hitting a consumer hardware threshold.
Qwen3
M3 Max MacBook Hits 110 tok/s on Qwen3 35B: Local AI Works, Tools Don't
Reddit benchmarks 4 local AI engines on a $2,498 M3 Max MacBook, hitting 110 tok/s on Qwen3 MoE — first consumer laptop at daily-usable speed.
AMD
Two $150 Secondhand GPUs Run 35B Models — Local AI Compute Costs Collapse
A Reddit user ran Qwen3 35B and other major open-source models on two retired AMD MI50 GPUs costing under $300 combined — local AI hardware costs are
Perplexity
Perplexity Turns Qwen3 Into a Decision Model — Open LLMs Pivot to Real Work
Perplexity open-sourced Decider 27B, a Qwen3 fine-tune built for judgment calls. Open-source AI is pivoting from sounding human to doing real work.
Qwen3
RTX 5060Ti Fails on Qwen3 27B — Local LLM Hardware Bar Higher Than Expected
A Reddit user spent a week trying to run Qwen3 27B on an RTX 5060Ti 16GB, no ideal config. How high is the hardware bar for local LLMs?
Strata
Consumer Laptop LLM Speed Doubles Again — But How Far From Replacing the Cloud?
Strata inference engine pushes Qwen3-8B on consumer laptops from 23 t/s to 51 t/s — over 2x llama.cpp. Is local AI hitting its productivity tipping po
Qwen3
Local LLMs still brutal: even pros can't solve dual-GPU VRAM imbalance
A Reddit LocalLLaMA user tried Qwen3 27B on dual RTX 4090s — over 1GB VRAM sat idle no matter how he tuned tensor-split. Local AI is still a geek toy.
Splash
Splash 1.1: Apple Silicon Now Runs 27B AI Models Smoothly
Splash 1.1.0 runs 27B models locally on top-spec Macs at ~50 tokens/sec. Local AI inference becomes practical—enterprises can skip cloud fees.
Qwen3
How Expensive Is Local Multi-Agent? One Consumer GPU Handles Only 8 27B Agents
LocalLLaMA benchmark shows Qwen3 27B + vLLM on a single consumer GPU runs only 8 simultaneous multi-agents. The year's hottest AI trend hides a steep
Qwen3
Qwen3 27B Runs CAD at Home — Open-Source LLMs Cross the 'Real Work' Line
This week a Reddit dev ran Qwen3 27B on an RTX 5090 to autonomously code CAD tools, designing a 3D-printable planter and MagSafe stand — all at home f
Qwen3
Qwen3 Fine-Tuning Cuts 40% of Inference Compute — A 'Think Less, Do More' Race
ThinkingCap and Swift — Qwen3 fine-tunes — each cut ~40% of tokens on Aider's code benchmark with similar pass rates, opening a 'think less' track.
Qwen3
Qwen3.8 Hits 65 tokens/s on a 12GB GPU — China's Open-Source LLMs Go Local
A developer hit 65 tokens/s on Alibaba's Qwen3.8-Flash-Next using a 12GB consumer GPU — a localization milestone for Chinese open-source LLMs.
CLM
Open-source CLM matches closed-source Jev on Agent decisions—capped at 8K
Open-source CLM runs 4-13x faster than closed-source Jev on Agent decisions, but context is capped at 8K and zero-shot breadth still trails Jev by ~4p
Tenstorrent
Tenstorrent Runs Qwen3.7-27B — Non-NVIDIA AI Chip Breaks Commercial Ice
Tenstorrent user shares inference data on QuietBox 2 running Qwen3.7-27B. First near-commercial benchmark from the non-NVIDIA camp, but still far from
Qwen3
A 24GB workstation card runs Qwen3 27B — local LLMs are finally viable
Reddit dev ran Qwen3 27B on a single 24GB workstation GPU, hitting 128K context and 60 tok/s. ~$2.8K hardware now handles mid-size LLMs locally.
Ornith
Ornith Runs 35B Coding Model on 8GB VRAM: Local AI's Sweet Spot Arrives
A Reddit user ran Ornith-1.5-35B-A3B on an 8GB RTX 3070 laptop at ~32 tokens/sec, completing agentic coding tasks end-to-end. Consumer hardware is now
Ninfer
Qwen3 Hits 220 tokens/sec on 5090 — Local AI Inflection Point Nears
New inference engine Ninfer pushed Qwen3 to 170 avg / 220 peak tokens/sec on RTX 5090 — over 2x faster than llama.cpp. Local AI is closing in on cloud
curvedinf
$6,500 Used GPUs Hit Top AI Speed — Solo Dev Rewrites Inference Economics
GitHub's curvedinf pushed Qwen3 27B from 15 to 972 token/s (65x) on $6,500 of used AMD MI100 GPUs. Enterprise AI's hardware bar just dropped.
Gemma4
Gemma4 vs Qwen3 Split Top AI Leaderboards — A Benchmarking Trust Crisis
Google's Gemma4 31B and Alibaba's Qwen3 27B received nearly opposite rankings on Artificial Analysis and LMArena — exposing systematic failure in AI b
Qwen3
No One Explains How to Run Qwen3 Locally—So a Reddit User Spent $100
A Reddit user is spending $100 to benchmark Qwen3 quantizations—exposing the open-source LLM ecosystem's failure to guide local-deployment users.
Qwen3
Qwen 27B Long-Context Speed Drops 65% — Local Players Distrust Benchmarks
AMD's 51.8 tokens/sec Qwen3 27B claim is ideal-only, Reddit user finds; real long-context drops 75→26 (–65%). Local LLMs now judged on stability.
DeepSeek
Open-Source LLMs Match GPT-5; Local Deployment ROI Drops to Two Months
NVIDIA's 748GB VRAM workstation and DeepSeek-V3/Qwen3 matching GPT-5 cut local LLM payback to two months. Hybrid is becoming the default architecture.