←
Ryu19,800 posts
Ryu avatar

Ryu

@_ryu15_

Local LLMを中心に、AI技術系の投稿とネタツイ Tesla v100 32GB×2枚と共に生きる。雰囲気Researcher 最近はFlash Nextちゃんと遊んでいます アイコンは羊野郎さん@maton_3110

◷ Joined August 2019▧ 762 Photos and videos
Ryu avatar
Ryu@_ryu15_·Sep 15

完成!!!!! 僕のVRAM 64GB ヘッドレスAIマシンです!! マザボ:ASUS Z170-A CPU: intel core i7 6700k GPU: Nvidia v100 32GB + PG500-216 32GB RAM: corsair ddr4 3200mhz 32gb PSU: CGR GDN-850 SSD: Fanxiang S501 512GB 総額25万円弱でした なお、メモリは去年買ったので今この構成作るともっと高いかも。

Post attachment
Ryu@_ryu15_Sep 15

ATX工業フレームマニュア

Post attachment
Ryu avatar
Ryu@_ryu15_·2h

12GB VRAMと32GB RAMでQwen3.8 flash next動かすエンジン!! 私はRAM64GB派閥だったけどサブPCも推論機化できたら激アツ!! VRAM12GBだと最良でも9Bとか35BA3Bとかが良かったけど、 12GBでいけるならQwenFlashNextはまじそれとは比べられんレベルで強いのでおすすめ。

Sync-Strata、一般公開しました🎉 github.com/hiruhiru2005/sync… 家庭用PCでQwen3.8 Flash-Next(125B MoE)を高速に動かせる、Strataベースのローカルアプリです RAMに合わせてSync-XS / S / Mの3モデルから選べます Windows / Linux対応 (注: 試験版のため、不具合が起きる可能性があります)

ヒル avatar
⟳ Ryu retweeted
ヒル@hiru_2005·4h

Sync-Strata、一般公開しました🎉 github.com/hiruhiru2005/sync… 家庭用PCでQwen3.8 Flash-Next(125B MoE)を高速に動かせる、Strataベースのローカルアプリです RAMに合わせてSync-XS / S / Mの3モデルから選べます Windows / Linux対応 (注: 試験版のため、不具合が起きる可能性があります)

32GB RAMでStrataが140tok/s (IQ3_S同等)! (Sync-Strata 独自エンジン・量子化) Strataを低RAM PC向けに改造したエンジン Sync-Strataと、それに対応したQwen3.8 Flash-Next量子化モデルSyncを開発しています。(1/3)

Post attachmentPost attachment
div (fp32) avatar
⟳ Ryu retweeted
div (fp32)@divagr1925·Nov 17

ngl Gemini 4 Carbon's way of writing is so REFRESHING to read, extremely parseable prose (I'm looking at you Astra)

Ryu avatar
Ryu@_ryu15_·3h

結局Astra6.2 とSol6.2のいい話何も聞かんけど安くなったが目玉なんかな? Solに至っては後退したベンチもあったけど、どうなんやろ...w Gemini4 Neonの噂も出てきたし、次モデルがまた2ヶ月後なら競争かなり落ちるな〜

clem 🤗 avatar
⟳ Ryu retweeted
clem 🤗@ClementDelangue·7h

Number 5 trending on the hub! huggingface.co/harness-hub/Q…

Post attachment

Introducing Qwen3.8-27B-pi-adapted 🤗 We adapted Qwen's model specifically for the pi coding agent using reinforcement learning and a LoRA adapter. The results, measured as pass@1: 🚀 SWE-bench Pro (Hard 51): 29.4% → 62.7% 💻 Terminal-Bench 2.1: 39.5% → 55.8% ⚡ 55–62% fewer tool calls per task This is the power of open models: you can adapt them to YOUR tools and YOUR use cases. Let’s build! 💛 Open weights: huggingface.co/harness-hub/Q… Methodology: huggingface.co/blog/harness-…

Post attachment
Elshayib avatar
⟳ Ryu retweeted
Elshayib@elshayib_·9h

Wait what? There's no way, if these Benchmarks turn out to be true Google will claim first place in the race with this Gemini Neon beast. What happened at Google man? Something clicked over there

Post attachment
younes avatar
⟳ Ryu retweeted
younes@yb2698·8h

Announcing harness adapters for @Alibaba_Qwen Qwen3.8-27B on @opencode harness! After yesterday's release of the Pi adapter, we now released to the community the OpenCode version, trained on the v2 version, ready to use for local inference. SWE-Bench Pro (Hard 51 subset) pass@1: 35.3% -> 51.0% | xhigh thinking mode 47.1% -> 54.9% Terminal Bench 2.1 pass@1: 47.0 -> 53.0 | xhigh thinking mode 55.4 -> 56.6 Generated tokens (content + reas content + tool calls): 6-8x fewer tokens per task ! huggingface.co/harness-hub/Q… 🧵

Post attachment
Ryu avatar
Ryu@_ryu15_·8h

Hy4 Flash 156B A12B +64B NgramならStrata的にまた使えるモデルになるじゃん! A12BだからアクティブがQwenFNより2倍大きいけど結構良さげ! 次のQwenFN枠として良さげかもしれない... 最小VRAM24GBとかからになってきそう

Hy4 flash is now open-weight, including Hy4 flash-FP8 and Hy4 flash-Base. 224B (12B active), 1M context, natively multimodal. Fast, efficient and built for everyday productivity. It's yours now. And stay tuned for Hy4 coming up next! 🤗: huggingface.co/tencent/Hy4-f… 🐙: github.com/Tencent-Hunyuan/H… 📄: hy.tencent.ai/research/hy4-f… Released under the Tencent Hy License. See full terms here: huggingface.co/tencent/Hy4-f…

Post attachment
NASA Solar System avatar
⟳ Ryu retweeted
NASA Solar System@NASASolarSystem·11h

Nov. 18, 2026, 5:16am EST The Voyager 1 spacecraft has reached a distance of one light-day from Earth.

AI Tracker Bot avatar
⟳ Ryu retweeted
AI Tracker Bot@aitrackerbot·13h

Gemini 4 Neon, the model we caught in the Gemini app on Monday, now has prices in Google Cloud's billing catalog 🔥 • $3 in / $15 out per 1M tokens, live as of 3 AM ET • That puts it between Hydrogen ($1.50 / $7.50) and Carbon ($4 / $20) • Batch and Flex at half price: $1.50 / $7.50 • Takes text, image, video and audio Not answering on Vertex yet. We're watching it closely.

GEMINI 4 NEON 👀 A model called gemini-4-neon has surfaced in the Gemini web app's code from Google. For now it's only a code reference, with no public access.

Post attachment
Ryu avatar
Ryu@_ryu15_·11h

OpenCodeの地元のGLM5.3 が10$で500$はワロタ なんで課金したら50倍の元取れるんだよw 今週まで!

F*CK IT $10 for $500 GLM 5.3 till end of the week

OpenCode avatar
⟳ Ryu retweeted
OpenCode@opencode·13h

F*CK IT $10 for $500 GLM 5.3 till end of the week

we're bored we're just gonna do phase 2 now $200 of GLM 5.3 usage for $10 have fun

Tencent Hy avatar
⟳ Ryu retweeted
Tencent Hy@TencentHunyuan·16h

Hy4 flash is now open-weight, including Hy4 flash-FP8 and Hy4 flash-Base. 224B (12B active), 1M context, natively multimodal. Fast, efficient and built for everyday productivity. It's yours now. And stay tuned for Hy4 coming up next! 🤗: huggingface.co/tencent/Hy4-f… 🐙: github.com/Tencent-Hunyuan/H… 📄: hy.tencent.ai/research/hy4-f… Released under the Tencent Hy License. See full terms here: huggingface.co/tencent/Hy4-f…

Post attachment
Ryu avatar
Ryu@_ryu15_·15h

なんかMuseが自分のブラウザの操作できる拡張機能でててアツいw これ使えば自分のブラウザ自体をMuseが操作するのでこれまでVM弾かれてたようなサイトも操作できるね!!! これですべてのブラウザ操作できるようになっちゃったおかげでMuseの価値また上がったな〜 以下は”4倍速”でddr4 32GB*2の中古を探させてるところ。(良いのはなかった...) めんどい登録作業とか、申請書類とかなんかすべての事務作業代替できる気がしてきた

0:31