billyG88

What compression looks like on @vllm_project. Same Gemma 4 31B. Red Hat AI's quantized version runs at nearly 2x tokens/sec, half the memory, 99%+ accuracy retained. Open source. Quantized with LLM Compressor. Links in comments. 🙏 @_soyr_ for the 2-minute demo.

🚀 New TTS Model: MOSS-TTS-Nano, a Real-time TTS that runs fully on CPU! 💡I'm always looking for good TTS models to run on an average PC (w/o a GPU) and this is one of them. OpenMOSS is a new 0.1B multilingual model: • No GPU needed, pure CPU inference • Real-time speech generation + streaming • 20 languages supported • 48 kHz stereo quality • Ultra-lightweight & easy local/web deployment Huge value for edge devices, local demos, web apps, and lightweight products where GPU isn’t an option. Part of the MOSS-TTS family 👏

Say hello to MOSS-TTS-Nano 🚀 0.1B multilingual TTS from MOSI.AI and OpenMOSS. Designed for realtime speech generation without a GPU. Runs directly on CPU, keeping the deployment stack simple enough for local demos, web serving, and lightweight product integration. Part of the MOSS-TTS family alongside the 1.7B and 8B flagship models. 🤖 modelscope.cn/models/openmos… 🌍 modelscope.ai/models/openmos… 💻 github.com/OpenMOSS/MOSS-…

started with a raspberry pi, now i run an entire AWS region at home

Externalization in LLM Agents: A Unified Review of Memory, Skills, Protocols and Harness Engineering arxiv.org/abs/2604.08224

@Hesamation I wonder how many AI avenues could benefit from 1T $$$ in investment…

DHH is in. Karpathy is in. Andrew Ng is in. Terence Tao is in. Linus Torvalds is in. John Carmack is in. Tony with an opinion still believes AI is just a next-token predictor with no real future.

MiniMax m2.7 at 83gb. wow.

🎉Congratulations to the @MiniMax_AI team on the launch of MiniMax M2.7! MiniMax M2.7 is now available with NVIDIA GPU accelerated endpoints ready to try out with claws including NemoClaw and @OpenClaw. 🦞 📝Get started with our technical guide: developer.nvidia.com/blog/minimax-m… and see how you can begin experimenting for free at build.nvidia.com/minimaxai/mini…. What will you build this weekend? Share in comments. 👇

Trending #1 on @huggingface. The hardest one yet.

As promised, here's a recording of my 30-min keynote and the subsequent Q&A for the inaugural late interaction retrieval (LIR) workshop, cc @bclavie @antoine_chaffin. The talk is admittedly advanced, as it's directed at an expert IR community. But hopefully still broadly useful!

This 2 hour Stanford lecture shows exactly how Stanford trains it's engineers to build AI systems. It's more practical than every Claude tutorial & prompting threads you've seen. Bookmark & give it 2 hours, no matter what. It'll be the most productive thing you do this weekend.

I'm sorry George.

@pfau AXAXAXAXAXAXXAAXXAXAXAXAXAXAXAX

At the current growth rate of 3x every quarter, Anthropic's revenue is on track to surpass Google in Q4 this year, Amazon in Q1 next year and the entire United States federal government somewhere in Q2/Q3.

This is really 2019 pure science fiction.

This 15-minute talk by the creator of Pydantic on how to correctly use MCPs will teach you more about making your AI tools actually work together than everything you've scrolled past this year. Bookmark this & watch, no matter what. Then read the guide below by @eng_khairallah1

training in unlsoth - learning which quant size to train - training parameters based on data size - merging LoRA - exporting to gguf / quantized - merging ggufs - running on llama.cpp vs. ollama fun stuff. lots of wasted time but learning by doing is best IMO

Tomorrow, we're launching the world's most advanced computer-use dataset. Stay tuned :) @markov__ai

I actually think this is the best subscription out now I just used $10 of opus last night or I could have a month of GO $10 a month with generous limits on open models The best of both worlds for qwen 27b + Hermes

BREAKING: Wan2.7-Video by @Alibaba_Wan is now #1 on Video-to-Video Arena with an Elo of 1337! This establishes a new state of the art with video editing models Huge congrats to the @Alibaba_Wan team!

💥 MiniMax M2.7 is NOW running locally — 229B open source model on CPU + GPU ♠ and everyone can do it with llama.cpp 🔹229B MoE model, 256 experts, 8 active per token 🔹IQ4_XS quant brings it down to 100GB — no need for multiple GPUs 🔹Split across H100 VRAM + system RAM via llama.cpp GPU offloading 🔹OpenAI compatible API endpoint — drop-in replacement for any client 🔹SWE-Bench Pro 56.2 — within 2 points of GPT-5.4 🔹MLE-Bench Lite 66.6% medal rate — #2 only to Opus 4.6 and GPT-5 🔹Modified MIT license — run it, build on it, ship it 🔥 Watch the full video below:👇