Havuz

llama.cpp fork for significantly improved performance on Ampere generation Nvidia cards (especially RTX 3090 / 3090 Ti): example: 110+ tok/s

Slotstream, büyük bir MoE modelini SSD’den parça parça yükleyip yoğun kullanılan expert’leri bellekte önbelleğe alarak Apple Silicon Mac’lerde çalıştıran yerel bir inference engine’dir.

Neden yükseliyor · README, 48 GB M5 Pro üzerinde belirli koşullarda 15.86 tok/s ölçümü ve prompt işleme iyileştirmeleri yayımlıyor; son 28 günde 271 commit ve 5 farklı yazar da yoğun geliştirme faaliyetine işaret ediyor. Ancak proje yalnızca 30 günlük ve mevcut sürüm için nitelikli tam yanıt hızı ölçümünün hâlâ beklediği belirtiliyor.

halogen-flash-server, Qwen3.8-Flash-Next modelini AMD Strix Halo (gfx1151) donanımında çalıştırmak için özelleştirilmiş bir inference server'dır.

Neden yükseliyor · 32 günlük repo için 722 yıldız ve son 28 günde 99 commit dikkat çekiyor. README, belirli donanım üzerindeki ölçümleri ve ölçüm koşullarını açıklıyor; ancak performans iddiaları proje tarafından sunulmuş ve bağımsız doğrulama bilgisi burada yok. 99 commit'in tek yazardan gelmesi geliştirme hızını gösterirken bakım kapasitesine dair kanıtı sınırlıyor.

37 tok/s from Qwen3-Next-80B (48 GB) on a 32 GB laptop, 24 without drafts vs 2.8 for llama.cpp; reading 62 vs 16 tok/s. Faster than llama.cp

Bilingual English and Mandarin text to speech in C++ and GGUF. Voice design, cloning, direction and experimental voice conversion. Vulkan ba

A catalog of every MLP neuron and attention head of Qwen3-8B and Gemma 4 12B, read from the weights through two Jacobian lenses (R and S), f

Fully uncensored, capability-enhanced abliteration of Qwen3.6-27B. NVFP4 + z-lab DFlash speculative decoding (n=12) on the unified ghcr.io/a

Evidence-backed AMD Strix Halo local-AI setup and benchmarks: Qwen3.8, Ollama, llama.cpp, Vulkan/ROCm, large GGUFs, and cross-OEM results.

🎬 Generate images from any camera viewpoint via 3D interactive control. Drag the camera in 3D space or use sliders to set azimuth/elevation/

The fastest way to run Qwen3.8 27B on Strix Halo (gfx1151)

Open-source agent execution-recovery controller for coding and tool-using AI agents.

MemPrivacy is a privacy-preserving personalized memory management framework for edge-cloud agents.

Open-source deep research MCP. Qwen3-30B-A3B-Thinking via OpenRouter, cited web synthesis for Claude Code, Codex, Cursor, Hermes and any MCP

Serving Qwen3.8-27B-FP8 on a single DGX Spark (GB10): 7.88 to 58.5 tok/s single-stream from decode strategy alone, weights untouched. Specul

Serving 4-bit Qwen3.8-27B on a single DGX Spark (GB10): 75 tok/s single-stream, 246 tok/s aggregate at 8-way concurrency. NVFP4 vs MixedInt4

FlowState QMD — Anticipatory Memory for AI Agents. Hermes 2026 Hackathon Entry.

OpenAI-compatible inference server for Qwen3.8 on one RTX 5090 / RTX PRO 6000: the 27B at ~170 tok/s (peaks ~370), Flash-Next (125B MoE) at

Durable local inference for Oh My Pi: NInfer + Qwen3.8 27B on one RTX 3090/4090/5090, with restart-resumable OpenAI Responses state.

Build & prompt Qwen3.x agents in their native dialect — extracted from qwen-code & Qwen-Agent

An assistant that lives next to your cursor on Windows. It sees your screen, talks with you, draws on things, solves the math you write, and

AI coding agent harness for any hardware, local or cloud (OpenAI-compatible engines). Tuned for AMD Strix Halo: Qwen3.8 125B on the iGPU, NP

A local AI agent built from scratch in Python with Ollama: conversation memory, system prompts, and multi-step tool calling.

Crow — a local AI agent, not a chat box: 30 tools, MCP, memory, skills, goals, subagents, vision, and image generation and editing with Qwen

llama.cpp speculative decoding measured on one RTX 3090, Qwen3.6-35B-A3B UD-Q4_K_XL, commit 3737e4137. Published figures are re-derived from

Multimodal AI studio powered by Qwen3.6-35B-A3B. End-to-end web app exposing visual reasoning, image captioning, and document understanding

Qwen3.8-Flash-Next 177B MoE (IQ3_S 3.44bpw) on RTX 5090 Laptop 24GB: full-model quality @ 97.7 tok/s · dual-lane 256K full-context unlocked

Local classification, routing, and scoring for AI agents, with option probabilities. Text + images, Python, MCP and agent skills.

Qwen3.8-27B on one laptop CPU: up to 2.52 token/s, 8 GB tested, no runtime accuracy loss. Native OpenAI-compatible API with function tools;

The best model under 200B at near 10 token/s exact batch throughput on a single laptop CPU, with an automatic 8 GB RAM path. Native C, no GP

Think only when the tool call looks unsure: confidence-gated reasoning for small local models. Code, BFCL results and paper.

Asking for What Was Never Requested: horizontal and vertical proactivity in LLM agents. Need-graph metrics (no LLM judge) and Q&D, which tra

Deploy Qwen3.6-35B-A3B (Q4_K_XL) + MTP speculative decoding on a single NVIDIA L4 24GB — GCP g2-standard-8 — via the official llama.cpp Dock

Train JEV-style System 1 decision models on Qwen3.5 and Qwen3.8-27B: 24-slot decision head, KL distillation, calibration and serving. 在 Qwen

Run stock llama.cpp with the stock Unsloth Qwen3.6-35B-A3B 4-bit on a 16 GB Mac.

One command spins up Qwen3.8-27B (abliterated, NVFP4) on a rented Blackwell GPU at ~150 tok/s and wires it into your local coding agent via

Docker compose configs for serving Qwen3.8-27B on a DGX Spark (GB10): vLLM+MTP current, SGLang+DSPARK + FP8 rollback stacks

Kata builds the best AI agent for a subnet through open competition — so anyone can mine that subnet with a proven, optimized agent.

A 6.87 GiB GGUF build of Qwen 3.8 27B — full 27B quality on 8 GB RAM. 96.7% HumanEval, 92.9% MMLU retention, no distillation.

Qwen3.6 vLLM Toolkit — Launcher + Templates, Optimized for 2×24 GB. vLLM launcher with an optimized hybrid chat template drawing from the be