Top AI Repos — open-source AI, indexed and scored
Top AI Repos tracks AI repositories on GitHub and answers two different questions about each one: is it moving right now, and would you bet a product on it.
Top AI Repos tracks AI repositories on GitHub and answers two different questions about each one: is it moving right now, and would you bet a product on it.
Evidence-backed AMD Strix Halo local-AI setup and benchmarks: Qwen3.8, Ollama, llama.cpp, Vulkan/ROCm, large GGUFs, and cross-OEM results.
| Date | Stars |
|---|---|
| 2026-08-05 | 257 |
| 2026-08-06 | 260 |
| 2026-08-08 | 261 |
| 2026-08-09 | 262 |
| 2026-08-10 | 264 |
| 2026-08-11 | 266 |
| 2026-08-12 | 267 |
| 2026-08-14 | 268 |
| 2026-08-15 | 270 |
| 2026-08-16 | 272 |
| 2026-08-17 | 274 |
| 2026-08-18 | 280 |
| 2026-08-19 | 283 |
| 2026-08-20 | 285 |
| 2026-08-21 | 288 |
| 2026-08-22 | 291 |
| 2026-08-23 | 292 |
| 2026-08-24 | 294 |
| 2026-08-26 | 296 |
| 2026-08-27 | 300 |
| 2026-08-28 | 303 |
| 2026-08-29 | 307 |
| 2026-08-30 | 310 |
| 2026-08-31 | 313 |
| 2026-09-01 | 314 |
| 2026-09-03 | 316 |
| 2026-09-04 | 318 |
| 2026-09-05 | 320 |
| 2026-09-06 | 323 |
| 2026-09-07 | 324 |
| 2026-09-08 | 325 |
| 2026-09-11 | 326 |
| 2026-09-14 | 327 |
| 2026-09-15 | 330 |
| 2026-09-17 | 332 |
| 2026-09-18 | 333 |
| 2026-09-19 | 336 |
| 2026-09-22 | 341 |
| 2026-09-23 | 343 |
| 2026-09-25 | 346 |
| 2026-09-26 | 345 |
| 2026-09-28 | 346 |
| 2026-09-29 | 347 |
| 2026-09-30 | 350 |
| 2026-10-01 | 351 |
| 2026-10-02 | 350 |
| 2026-10-05 | 351 |
| 2026-10-06 | 352 |
| 2026-10-07 | 353 |
| 2026-10-09 | 354 |
| 2026-10-11 | 356 |
| 2026-10-13 | 358 |
| 2026-10-14 | 358 |
| 2026-10-15 | 357 |
| 2026-10-16 | 359 |
| 2026-10-17 | 360 |
| 2026-10-18 | 360 |
| 2026-10-19 | 361 |
| 2026-10-20 | 362 |
| 2026-10-21 | 362 |
| 2026-10-22 | 362 |
| 2026-10-23 | 363 |
| 2026-10-27 | 365 |
| 2026-10-28 | 366 |
| 2026-10-31 | 367 |
| 2026-11-04 | 369 |
Today
+2 stars today
This week
+3 stars this week
This month
+18 stars this month
Momentum
26.0
growth rate 0.82%/day
# Strix Halo Guide: AMD Ryzen AI MAX+ 395 Local LLM Setup & Benchmarks **Maintenance update: October 21, 2026.** Beginner-route evidence, model-answer corrections, current release observations and 495 dispatch status: [MAINTENANCE_REVIEW_2026-10-21.md](MAINTENANCE_REVIEW_2026-10-21.md). Historical measurements and the broad October 1 evidence review keep their original dates. Get a local AI model running on your AMD Strix Halo / Ryzen AI MAX+ 395 (Radeon 8060S) under Ubuntu 24.04, with settings measured on real hardware and a check that it really runs on the GPU. Then pick the right model for your work and make it faster. **Start here → [Get local AI running, step by step](LOCAL_CHAT_START.md)** · Ubuntu 24.04 · 128GB-class systems · from BIOS to your first answer and a GPU check, then chat in your browser if you want it. Built on [real-world results from 10 Strix Halo systems plus 3 independent sources](SYSTEM_EVIDENCE_MATRIX.md), with measured settings, model comparisons and documented fixes. **Setup status:** GPU answers and browser chat passed an isolated hardware check on October 15. A complete run on a freshly installed Strix Halo is not yet verified; [see what has been checked](LOCAL_CHAT_START.md#version-and-security-status). ## What This Gives You - **A clear setup route.** The settings that matter on this chip, from the BIOS memory option to the one Ollama setting without which measured builds quietly run on the CPU, in one [step-by-step start route](LOCAL_CHAT_START.md) that shows the expected result of every step. - **A model that fits your work.** Which model to run for chat, coding, images or long context, with the speed each one reached here: [Best Known Profiles](BEST_KNOWN_PROFILES.md). - **Know what's tested.** See which settings have been tested, what is still experimental, and which [known failures you can avoid](#things-that-dont-work-dont-waste-your-time). If something goes wrong, use the [troubleshooting steps](#troubleshooting). Every headline number links to its [raw evidence](#headline-evidence) or carries an explicit caveat. ## Start Here | I want to… | Start here | |---|---| | Get local AI running | [Install, get an answer and check the GPU](LOCAL_CHAT_START.md) | | Choose a model or make it faster | [Best Known Profiles](BEST_KNOWN_PROFILES.md) and [Current Models](CURRENT_MODELS.md) | | Fix something that is not working | [Troubleshooting](#troubleshooting) | | Decide what to buy | [Buyer Use Cases](BUYER_USE_CASES.md) and [Buying Guide](#buying-guide) | Also: [website](https://strixhaloguide.com/) · [Qwen3.8](QWEN38_STRIX_HALO.md) · [reproduce a result](REPRODUCIBILITY.md) · [vendor/reviewer brief](ONE_PAGE_BRIEF.md). ## Quick Start (6 Steps) New here? Use the [complete start route](LOCAL_CHAT_START.md): it has the commands, what each one should print, and what to do when it does not. The six points below are the short overview for anyone who wants to do, or understand, each step by hand: 1. **BIOS:** use UMA 512MB if available, or the vendor minimum such as 2GB. Leave IOMMU enabled/default; the desktop benchmark alternative has [suspend/NPU caveats](#step-12-choose-the-iommu-policy). 2. **OS:** install Ubuntu 24.04 LTS. X11 is needed only by desktop tools that require it. 3. **Memory:** the recorded 128GB profile uses `amdgpu.gttsize=131072 ttm.pages_limit=31457280 amdgpu.cwsr_enable=0`. Do not reuse these limits on 64/96/192GB systems; see [the manual memory section](#step-32-configure-grub-boot-parameters). 4. **Drivers:** use Mesa/RADV and preserve the existing power policy by default. `tuned` is an opt-in reproduction profile. 5. **Ollama:** the setup script does steps 3–5 for you on the 128GB-class profile. It sets `OLLAMA_VULKAN=1`, `OLLAMA_IGPU_ENABLE=1` and `HIP_VISIBLE_DEVICES=-1`; reboot first if it asks you to. 6. **Check:** chat with the model using `ollama run qwen3.6:35b-a3b`, then run the [known-answer, GPU and loopback-listener checks](LOCAL_CHAT_START.md#step-5-get-
Excerpt of 272,032 characters
Read on GitHubWould you bet a product on this? Bounded 0–100 and slow moving.
matched fp:c2c9717690b7c47e, llm:Repo description and README: guide to running large language models locally on AMD Strix Halo/Ryzen AI MAX+ 395, covers Ollama, llama.cpp, Vulkan/ROCm experiments, benchmarks, GGUF models, setup and reproducibility.
matched fp:c2c9717690b7c47e, llm:Repo description and README: guide to running large language models locally on AMD Strix Halo/Ryzen AI MAX+ 395, covers Ollama, llama.cpp, Vulkan/ROCm experiments, benchmarks, GGUF models, setup and reproducibility.
matched fp:c2c9717690b7c47e, llm:Repo description and README: guide to running large language models locally on AMD Strix Halo/Ryzen AI MAX+ 395, covers Ollama, llama.cpp, Vulkan/ROCm experiments, benchmarks, GGUF models, setup and reproducibility.
matched fp:c2c9717690b7c47e, llm:Repo description and README: guide to running large language models locally on AMD Strix Halo/Ryzen AI MAX+ 395, covers Ollama, llama.cpp, Vulkan/ROCm experiments, benchmarks, GGUF models, setup and reproducibility.