AlphaSignal

Finally found a way to explain what we do in a single image.

You can run open-source models like DeepSeek V3 for 90% cheaper in 2 min. Inference.net gives you access to top open models through a simple API that runs on unused GPU capacity. They built their own scheduler to grab idle GPUs across data centers and pass the savings straight to you. Features: ? OpenAI-compatible endpoints. ? Set it up in minutes and scale forever ? Supports 125K context, real-time + batch inference ? OpenAI SDK-compatible—just change 2 lines of code ? Supports LangChain, LlamaIndex, and more Try it here: https://lnkd.in/gCP-QHQi

Alibaba just released TaoAvatar on Hugging Face, enabling real-time lifelike full-body talking avatars for AR. The system uses 3D Gaussian Splatting to generate realistic avatars from a single image. It supports full-body animation with 15 FPS rendering on consumer hardware. Check out the paper and dataset: https://lnkd.in/gcS-ebEQ and https://lnkd.in/gw5ntPrJ

StarVector app is now available on Hugging Face. You can generate SVG code directly from images and text. The model comes in two sizes: a lightweight 1B parameter version and a more capable 8B parameter version. Both convert raster images to scalable vector graphics. Try the 1B model here: https://lnkd.in/gCzdEbvj or the 8B model here: https://lnkd.in/gp59efnR

Big News. Google just released the new Gemini 2.5, some are already calling it "the best multimodal model in the world". It ranked #1 on LMArena by +39 ELO points, with significant improvements across the board in multimodal reasoning, coding and STEM. > 18.8% on Humanity's Last Exam > 63.8% on SWE-Bench Verified (agentic coding) > GPQA Diamond and AIME 2025 (STEM) > Long-context and visual reasoning (1M context window) Try it now → https://goo.gle/4c2HKjf

Alibaba just released TaoAvatar on Hugging Face, enabling real-time lifelike full-body talking avatars for AR. The system uses 3D Gaussian Splatting to generate responsive avatars from a single image. It processes at 30 FPS with realistic body movements that match audio input. Check out the paper and dataset: https://lnkd.in/gcS-ebEQ and https://lnkd.in/gw5ntPrJ

Wow. Deepseek just dropped a ~700GB model, MIT licensed. https://lnkd.in/e-wFX9Zi

Meta just released Gemma, a minimal Python library for fine-tuning Gemma 3 models. It's 100% open-source with comprehensive docs covering LoRA, PEFT, multimodality, tokenization, and model sharding techniques. No proprietary components. Check out the documentation here: https://lnkd.in/gmFv92BD

Alibaba just released LHM on Hugging Face, enabling 3D human reconstruction from a single image in seconds. LHM generates high-quality, animatable human models with realistic textures. The model works with diverse poses and clothing types in any environment. Check it out here: https://lnkd.in/gPZtcT_Q

Microsoft just released Data Formulator, an open-source no-code tool for AI-powered data analysis and visualization. The LLM-powered tool transforms raw data into insights and creates visualizations with simple prompts. No coding skills required. Check it out here: https://lnkd.in/g6veuFsu