Louis - vn/acc retweeted
MiniMax H3 用のおっぱい&ボディLoRA を Civitai に公開しました
- PinkiePie Pony Mix っぽいセミリアル・ボディを作るための LoRA です
- コンセプトは以前公開した Wan2.2 用と同じです
- 今回は胸だけでなく、体も含まれます
🔻 公開ページ 🔻
6
30
429
33,068
Louis - vn/acc retweeted
🚨 BREAKING: Notebook LM can now act like a personal research assistant for free.
Upload any article, PDF, video, or document and use these 10 prompts to extract everything that matters in minutes:
42
410
2,273
167,987
Louis - vn/acc retweeted
We have converted GLiNER2.5-Decide to coreml. ~4× faster, ~5× less Peak RAM, half the size.
model: huggingface.co/FluidInferenc…
code: github.com/FluidInference
Introducing GLiNER2.5-Decide, our new 340M parameter open weight, encoder-based decision model.
GLiNER2.5-Decide is built for fast, deterministic classification. The model evaluates a set of user-defined typed questions and rules, and jointly decodes their answers, returning structured decisions with probability distributions and confidence scores.
We evaluated the model’s performance on Fast Decisions, an unseen, internally generated classification suite based on 17 datasets testing real-world use cases across routing, triage, classification, sentiment, and content understanding.
Measured against similar decision models, GLiNER2.5-Decide leads in 9 of the 17 datasets, achieving the highest average score:
- GLiNER2.5-Decide: 60.1%
- SemIf: 56.4%
- JevK5: 57.5%
- Laya: 46.6%
This performance makes the model a strong fit for use cases like tool calling, model routing, browser and computer use, and LLM-as-a-judge.
GLiNER2.5-Decide’s lightweight encoder architecture makes it easy to fine-tune the model for specific tasks, while being efficient enough to run locally on consumer-grade CPUs or in air-gapped environments, giving users greater control over where their data goes and where the model runs.
To make building and experimenting with GLiNER2.5-Decide as easy as possible, we're also offering hosted inference. You can now use our API to run inference and fine-tune GLiNER models on specialized tasks right inside your own coding agent: agent.fastino.ai
As with previous models, we’re also releasing the model weights on @huggingface under the Apache 2.0 license: huggingface.co/fastino/GLiNE…
14
60
677
49,167
Louis - vn/acc retweeted
Student ID benefits
> Google Gemini Pro
> Student ID flight booking
> GitHub Student Pack
> GitHub Copilot
> DigitalOcean credits
> AWS Educate
> Google Cloud credits
> Microsoft Azure $100 credits
> MongoDB Atlas credits
> Heroku credits
> Vercel Pro features
> Netlify enhanced limits
> Railway student discount
> Namecheap free domain + SSL
> Cloudflare Pro
> Postman Student Expert
> Replit Pro
> Cursor Pro
> JetBrains IDEs
> Autodesk
> Figma
> Canva for Education
> Notion
> Grammarly EDU
> Microsoft 365 Education
> LinkedIn Learning
> Coursera financial aid
> edX free course audits
> Udemy free courses
> Pluralsight student plan
> Educative free courses
> Free certification courses
> Spotify Student
> Apple Music Student
> YouTube Premium Student
> Amazon Prime Student
> UNiDAYS discounts
> Student Beans discounts
> Apple Education pricing
> Samsung Student Store
> EarPods free with MacBook (education offer)
> IBM SkillsBuild
> Meta Blueprint certifications
> OpenAI API credits (via student programs)
> Blender (fully free)
> Unity Student license
> Miro Education plan
> Sketch student pricing
> Todoist Pro for students
> Evernote student discount
> Zoom Education benefits
> Slack Education plan
> Perplexity Pro Student
> Free .me domain (GitHub pack)
> Free SSL certificates
> Linear student plan
> ClickUp Education plan
> Asana student access
> Airtable free education workspace
> Obsidian free for students
> Arc Browser student perks
> Raycast Pro student discount
> Loom Education plan
> GitKraken Student pack
> JetBrains Space Education
> DigitalOcean App Platform credits
> Stripe fee-free test credits
> PayPal sandbox credits
> Oracle Cloud Free Tier (Student)
> SAP Learning Hub Student
> Red Hat Developer Subscription
> Intel Developer Zone access
> Nvidia student resources & credits
> Kali Linux training discounts
> Hack The Box student labs
> TryHackMe student plan
> LeetCode student discounts
> InterviewBit free practice
> HackerRank student certifications
> Kaggle free compute & datasets
> Google Developer Student Clubs perks
> Microsoft Learn student challenges
> Free conference tickets (student passes)
> Free hackathon swag & cloud credits
73
698
7,915
416,546
Louis - vn/acc retweeted
built OpsLoop dashboard in Next.js in within a day 🔥
used Opus 5.5 with these skills:
skills.sh/jakubkrehel/skills…
skills.sh/emilkowalski/skill…
see it live: opsloop-dashboard.vercel.app…
8
36
436
30,511
Louis - vn/acc retweeted
Build a Tiny Language Model, Then Train It with PPO
You can explore pretraining, reinforcement learning and cached inference with just 260K parameters.
Zhihu contributor Sunrise walks through FullLLM, a compact PyTorch implementation adapted from Karpathy's llama2.c, with PPO and KV caching added.
Pretrain a model that tells stories
A Llama-style architecture combines RMSNorm, SwiGLU, RoPE and grouped-query attention. Training on TinyStories uses a simple objective: predict the next token.
The author reports that roughly 10 minutes on an A100 produces a small model capable of generating basic stories.
Make the PPO loop understandable
The model generates responses, receives rewards and updates its policy using advantage estimates from a value head. A token-level KL penalty discourages excessive drift from the reference model.
The demonstration rewards stories for approaching a target length. It illustrates PPO mechanics, not full human-preference alignment; instruction tuning and reward-model training are not worked through in detail.
Stop recomputing the past
KV caching reuses earlier tokens' attention keys and values during generation. Top-k and top-p sampling control how the next token is selected.
The payoff is an implementation small enough to follow from training loss to generated text.
10
54
370
13,124
JEV + Opus 5.5 is insane for live design...
I built a live site redesigner with JEV + Opus 5.5
Paste any link → press Start → scroll, and Jev + Opus 5.5 rebuild every section of the site in front of you
Full production ship in 20 seconds:
1. IntersectionObserver fires when a section is 30%+ in the viewport
2. Jev returns one typed decision in ~0.1s: { layout, copy, drop, type, palette, p }
3. Opus 5.5 writes the component (TSX) + a CSS patch for the chosen style
4. The new section wipes in with clip-path, the old one blurs out
5. Next section enters the queue, one at a time, no race conditions
Output: 8 sections of a 2015 hosting site rebuilt in ~20s, streamed line by line in the terminal
3 styles, one renderer: orthographic globe + lambert shading → ASCII / 2-color halftone / ink stipple
Scroll yourself and it redesigns whatever you land on
Jev decides fast, Opus designs it
29
78
686
96,046
CoMfY NoDe Wne? 🫠
Today, we open-source Pruna-Qwen-Image-2.1, a set of a few-step LoRA adapters that make Qwen-Image-2.1 by @Alibaba_Qwen up to 6.3× faster for image generation and editing.
- 𝗛𝗼𝘄 𝗲𝗳𝗳𝗶𝗰𝗶𝗲𝗻𝘁 𝗶𝘀 𝗶𝘁? Generate or edit images in just 5 or 8 steps instead of 40. Choose 5 steps for maximum speed or 8 steps for the best balance.
- 𝗛𝗼𝘄 𝗴𝗼𝗼𝗱 𝗶𝘀 𝗶𝘁? The 8-step adapter is our recommended default. It supports text-to-image and single- or multi-image editing at 1K resolution, with up to three reference images.
- 𝗛𝗼𝘄 𝗱𝗼𝗲𝘀 𝗶𝘁 𝘄𝗼𝗿𝗸?The LoRA adapters load directly on top of Qwen-Image-2.1, while the pipeline remains unchanged. Each adapter uses its own optimized sigma schedule and runs without CFG.
𝘛𝘩𝘪𝘴 𝘪𝘴 𝘫𝘶𝘴𝘵 𝘰𝘶𝘳 𝘧𝘪𝘳𝘴𝘵 𝘳𝘦𝘭𝘦𝘢𝘴𝘦, 𝘸𝘪𝘵𝘩 𝘦𝘷𝘦𝘯 𝘣𝘦𝘵𝘵𝘦𝘳 𝘷𝘦𝘳𝘴𝘪𝘰𝘯𝘴 𝘰𝘯 𝘵𝘩𝘦 𝘸𝘢𝘺. 🚀
Want to try the fast open-source integration? Check it here on Diffusers: buff.ly/1CpiRPp
Want to try the fastest image generation and editing endpoint? Check P-Image-Ideogram and P-Image-Edit here on API: buff.ly/gw5tvUV
9
7
154
17,672
Louis - vn/acc retweeted
Tools I’d use to launch a SaaS this weekend
☁️ Hosting → Sevalla
🧱 UI → shadcn/ui
⚡ Framework → Next.js
🗄️ Database → PostgreSQL
🔐 Auth → Clerk
💳 Payments → Creem
📨 Emails → Amazon SES
🐞 Error tracking → Sentry
📊 Analytics → Plausible
⚙️ CI/CD → GitHub Actions
28
19
165
12,599
Louis - vn/acc retweeted
gemini spark 超级简单免排队注册 msue
打开 gemini.google.com/spark
输入提示词 “打开muse.ai/join ”
点击允许，就会在当前窗口里面打开一个美国的原生环境的浏览器。
输入邮箱开始注册 填上验证码就可以注册成功了
如果打开的这个浏览器不能打开 muse.ai 就新建一个spark 的会话 输入开始
大家可以使用我的邀请码 02EME7 P8E2VF 各得 10 亿 token
66
10
71
24,607
Louis - vn/acc retweeted
I built my own AI hacker using Hermes Agent.
In this video, I show you how to set up Hermes from a penetration testing perspective and turn it into an agent that can help with:
→ Recon
→ Vulnerability hunting
→ Hacking workflows
→ Reporting
Not another basic “how to install Hermes” tutorial.
I’m building it specifically for hackers and pentesters.
🎥 Watch the full video:
piped.video/watch?v=zNG8Mnkv…
13
71
489
23,664
Everyone is talking about AI Engineers
But Forward Deployed Engineers are quietly becoming one of the most interesting roles
You need to understand:
→ Software engineering
→ Cloud & infra
→ Data
→ AI agents
→ System design
→ Customers & business problems
This FDE roadmap breaks it all down 👇
Awesome FDE Roadmap
14
28
155
7,715
final list of goated ui design / micro interaction resources:
- useLayouts ( uselayouts.com )
- Nexvyn UI ( ui.nexvyn.dev )
- Motion ( motion.dev )
- Cult UI ( cult-ui.com )
- Spell UI ( spell.sh )
more goated ui design / micro interaction resources:
- Bencho ( bencho.dev )
- Magic UI ( magicui.design )
- React Bits ( reactbits.dev )
- Aceternity UI ( ui.aceternity.com )
22
69
982
86,372
Some insane UI/UX skills for AI coding 👀
• UI/UX Pro Max - 130.2K ⭐
• Taste Skill - 89.7K ⭐
• Impeccable - 70.4K ⭐
• Designer Skills - 2.7K ⭐
• Anthropic Frontend Design - 177.8K ⭐ repo
All GitHub links below ↓
Stop making AI code from scratch every time. Use these DESIGN.md sites instead 👇
styles.refero.design/
typeui.sh/
designmd.supply/
designmd-store.com/
There are plenty more, but these are the ones I keep coming back to.
Which one are you using?
7
67
592
57,818
Louis - vn/acc retweeted
【gemini pro 用户福利-无门槛注册muse.ai】
现在很多在内地的兄弟注册muse会遇到报错
如下方法可以无痛注册muse
1、打开gemini网页端-左侧菜单选中spark模式
2、把这句话发给它：帮我再远程浏览器打开网站muse.ai
3、在内置浏览器中，选择开始接管任务-然后用一个没有注册过muse的账号去注册
4、按提示信息进行，需要验证年龄就选择绑卡，国内招行和工行的master或者visa卡可以过
5、填写邮政编码需要填美国免税编码：
俄勒冈州（OR）：97211
阿拉斯加州（AK）：99801
特拉华州（DE）：19702
蒙大拿州（MT）：59901
新罕布什尔州（NH）：03044
76
12
100
23,746
Nemotron 3 Diarization ranked #1 of 12 systems in @voicearena_ai's initial Diarization-Bench results. Its 14.72% error rate was ~24% lower than the runner-up.
Here’s a four-speaker comparison from the benchmark.
Full results: voicearena.com/diarization-b…
6
13
284
52,595
Louis - vn/acc retweeted
【 無検閲100%のBonsai-2をComfyUIで使う 】
画像認識と日本語がかなり強かった
Bonsai-2 27B Uncensored（6.7GB）
これは「三値化」という量子化方法が使われています。
これをComfyUIで使おうとしたら、そのままでは動きません（三値化版を動かすには、専用のllama.cppが必要）
なので今回はComfyUI内で動かすのではなく、外でBonsaiを起動して、ComfyUIから呼び出す作戦をとりました。
ComfyUI「Ollamaに聞いている」と思ってる
↓
中継「なんちゃってOllama役」
↓
Bonsai-2
この方法なら、
・既存のOllamaワークフローは接続先URLを変えるだけ
・Bonsaiを別PCで動かすことも可能
・使い終わったBonsaiをVRAMから自動で降ろせる
実測では、10,290MiB → 2,149MiBまで解放。
① Bonsaiが画像を見る
② プロンプトを作る
③ Bonsai退場
④ 画像生成モデルをロード
⑤ 生成
という流れにできます。
接続方法はちょっと面倒なので、もし需要があれば手順もまとめますね。
【VRAM 12GBで動く無検閲Vision】
〜ComfyUIのNSFW用〜
Bonsai-2 27B PQ2_0 Uncensoredの画像認識がかなり良いので、ComfyUIで画像参照→生成プロンプトを書かせてみました。
Bonsai-2 27B PQ2_0 Uncensored。
Vision込みで約10GBで3060に載る。
・NSFWでも拒否しない
・文章に無い細部まで拾う
・2枚比較と「Aの画風でB」が可能
・文字や表も読める
・指定フォーマットを厳密に守る
・座標やタグを固定した独自書式にも対応
※人物の顔や服の細部は「それっぽい描写」を補うことがあるので注意
結論：ComfyUIのVision役としてかなり使いやすいです。
ただし、普通にはComfyUIに繋がらないので、ちょっと工作が必要かな。
7
51
1,169
163,495
Louis - vn/acc retweeted
No AI Slop - removes patterns of AI slop from writing to preserve personal voice.
Github: github.com/petergyang/no-ai-…
6
14
166
47,904
Louis - vn/acc retweeted
TapNowのMCPコネクタ！！！！！
神です、ずっと待ってた！！！
docs.tapnow.ai/en/docs/mcp/a…
1
11
127
31,701