Tessera
Folgen

@qdrddr.bsky.social

https://targeterpro.com/linktr-ee-qdrddr

559 Posts87 Follower121 Folgt
@qdrddr.bsky.social · 2d ago

Why do ppl omit #quantization level when presenting #LLM bench? Shy? 🙂 It degrades intelligence. Even state-of-the-art methods inevitably trade accuracy for efficiency. Compress a high-rez photo: much smaller, some detail is inevitably lost, and some become unrecognizable. There’s no free lunch #AI

@qdrddr.bsky.social · 5d ago

Local #LLM proxy for #agents on #macOS. Switch between models and providers. Manage #API keys or #OAuth. #AI #Apple #OpenSource MIT lic github.com/nguyenphutro...

GitHub - nguyenphutrong/quotio: Stop juggling AI accounts. Quotio is a beautiful native macOS menu bar app that unifies your Claude, Gemini, OpenAI, Qwen, and Antigravity subscriptions – with real-tim...

Stop juggling AI accounts. Quotio is a beautiful native macOS menu bar app that unifies your Claude, Gemini, OpenAI, Qwen, and Antigravity subscriptions – with real-time quota tracking and smart au...

github.com

@qdrddr.bsky.social · 2w ago

FP64 → FP32 → FP16/BF16 → FP8 → FP4. Why stop? Why not natively trained ternary neural nets with weights {-1, 0, +1}? Instead of multiplying → sign changes, zeroing, & addition, eliminating floating-point GPU HW from the core weight ops. Why we heading toward a post-FP era for Models so slowly?

@qdrddr.bsky.social · 2w ago

@cloudflare.social released a #harness #agent, confusingly called #Cloudflare OS, that includes an agent runtime, a workspace, and additional abstraction layers. Heavily dependent on Cloudflare’s own infrastructure. Beware of vendor lock-in in disguise. #AI #LLM #Agents #OpenSource Apache 2.0 lic

GitHub - cloudflare/cloudflare-os: Agent workspace built on Cloudflare Workers for creating documents, building apps, and running agents with your company’s context and systems.

Agent workspace built on Cloudflare Workers for creating documents, building apps, and running agents with your company’s context and systems. - cloudflare/cloudflare-os

github.com

@qdrddr.bsky.social · 3w ago

#Firecrawl Anydoc fast #local Convertor: #Word, #PowerPoint, #Excel, #OpenDocument, #RTF, #EPUB, #CSV, and #PDF to clean #Markdown. Consume via #CLI or embed into your app: Built in #Rust, with #Node.js and #Python bindings. No LLM. For some complicated docs you still need an LLM #OpenSource MIT lic

GitHub - firecrawl/anydoc: Convert Word, PowerPoint, Excel, OpenDocument, RTF, EPUB, CSV, and PDF to clean Markdown. Built in Rust, with Node.js and Python bindings.

Convert Word, PowerPoint, Excel, OpenDocument, RTF, EPUB, CSV, and PDF to clean Markdown. Built in Rust, with Node.js and Python bindings. - firecrawl/anydoc

github.com

Cloudflare@cloudflare.social · 3w ago

gRPC support is now available in Cloudflare Workers. You can now route, transform, and serve gRPC services natively at the edge without maintaining dedicated proxy infrastructure or translation layers. Read the launch post: https://cfl.re/4fMsitT

Cloudflare Workers and Containers now support inbound TCP connections and gRPC

Cloudflare Workers now support inbound TCP connections via Spectrum, allowing direct socket forwarding to Durable Objects and Containers. Developers can run full-duplex gRPC applications or leverage a...

blog.cloudflare.com

@qdrddr.bsky.social · 3w ago

Fast local #PDF classification and data extraction, no #OCR. By firecrawl. Rust native, bindings for Python & TypeScript. 200 docs parsed in 0.470s. #OpenSource MIT lic. github.com/firecrawl/pd...

GitHub - firecrawl/pdf-inspector: Fast Rust library for PDF inspection, classification, and text extraction. Intelligently detects scanned vs text-based PDFs to enable smart routing decisions.

Fast Rust library for PDF inspection, classification, and text extraction. Intelligently detects scanned vs text-based PDFs to enable smart routing decisions. - firecrawl/pdf-inspector

github.com

@qdrddr.bsky.social · last mo.

#Offline meeting transcript meeting assistant #macOS #Windows #OpenSource MIT lic #AI #LLM #Whisper github.com/Zackriya-Sol...

GitHub - Zackriya-Solutions/meetily: Privacy first, AI meeting assistant with 4x faster Parakeet/Whisper live transcription, speaker diarization, and Ollama summarization built on Rust. 100% local pro...

Privacy first, AI meeting assistant with 4x faster Parakeet/Whisper live transcription, speaker diarization, and Ollama summarization built on Rust. 100% local processing. no cloud required. Meetil...

github.com

@qdrddr.bsky.social · last mo.

Proxy #MCP servers via @cloudflare.social #AI Controls. Instead of loading every tool into context, #CodeMode turns an agent's tool into an #API. #Agent chain & filter across dozens of tools in one pass, without a round-trip to model for every call. Cut #token #costs by ~93% #LLM #Agents

Cloudflare Dashboard | Manage Your Account

Log in to the Cloudflare dashboard. Make your websites, apps, and networks fast and secure. Build modern apps on our developer platform.

dash.cloudflare.com

@qdrddr.bsky.social · last mo.

Chunk Your #MCP Tools 🧩 Filter tools + trim schemas before inference ⚡ Fewer tokens → faster #LLMs, simpler tool selection. Embedded into your app or test via CLI 📦 Rust • Python • TypeScript • C • Go 📜 Apache 2.0 #OpenSource #AI #RAG Repost 🔁

GitHub - qdrddr/chunk-your-tools: Chunk MCP tools: Tool with required properties; optional Properties and Enums

Chunk MCP tools: Tool with required properties; optional Properties and Enums - qdrddr/chunk-your-tools

github.com

@qdrddr.bsky.social · last mo.

#Security code scanner from @vercel.com requires an #API key and back-end #harness #Agents such as #Claude, #Codex, #Pi; defaults to #Opus model. 🚧 Long runs with complex code might be costly 💸 with frontier #LLM models. #OpenSource Apache 2.0 lic #AI #BlueTeam

GitHub - vercel-labs/deepsec: Deepsec is a security harness for finding vulnerabilities in your codebase powered by coding agents

Deepsec is a security harness for finding vulnerabilities in your codebase powered by coding agents - vercel-labs/deepsec

github.com

@qdrddr.bsky.social · last mo.

#Inference proof of concept: Stream large int4-quantized #GLM model from disk, running on 25 GB of #RAM and x12 #CPU cores. Treat Disk and RAM as unified memory: Attention, shared experts, and embeddings stay in RAM; the rest of the model is on disk. Pure C. #OpenSource Apache 2.0 lic #AI #LLM

GitHub - JustVugg/colibri: Run GLM-5.2 (744B MoE) on a 25GB-RAM consumer machine — pure C, zero deps, experts streamed from disk. Tiny engine, immense model. 🐦

Run GLM-5.2 (744B MoE) on a 25GB-RAM consumer machine — pure C, zero deps, experts streamed from disk. Tiny engine, immense model. 🐦 - JustVugg/colibri

github.com

@qdrddr.bsky.social · last mo.

Even though this paper is about proving that #diffusion models can't simulate chain-reaction physics, no matter the compute, to an extent, it’s also a proof of no reasoning in #AI and a proof that there will be no #AGI any time soon. Conscious probability never. #noAGI #LLM #VLM

The Seriality Gap in Video Diffusion Models

When one ball strikes another, then another, video models should predict the consequences of each bounce. In controlled experiments on multi-ball hard-sphere dynamics, we find that the performance of ...

arxiv.org