Engines and Tools

llama-modchip v2.4.0 Released for CPU Inference of Large MoE

Version v2.4.0 of llama-modchip brings CPU inference for massive MoE models like DeepSeek-V4-Flash on 64GB RAM.

Image, Video and Audio

Underdog Woof: Lightweight NumPy-Only English TTS Model

Explore Underdog Woof, a lightweight CPU-friendly English TTS model by ConwayResearch requiring only NumPy for high-a ...

Engines and Tools

Open WebUI v0.13.1 Released with Security Fixes and Tether

Open WebUI v0.13.1 has been released, featuring security fixes, the Tether desktop app, coding agent support, and per ...

New Models

Qwen3.8-27B-DSpine-b16 Draft Model for Speculative Decoding: 8GB+ VRAM

Explore tencent/Qwen3.8-27B-DSpine-b16, a draft model for speculative decoding using the DSpine method to accelerate ...

New Models

Arete-27B Text Generation Model: 12GB+ VRAM, GGUF Builds

Nous Research has released Arete-27B, a 27B text generation model aligned via SFT to fix biases and reduce propaganda ...

Engines and Tools

ggml-org Releases tinyLFM3-Omni-for-testing-gguf for CI

ggml-org released tinyLFM3-Omni-for-testing-gguf on Hugging Face to test llama.cpp multimodal and CI pipelines.

Engines and Tools

Unsloth v0.1.916-beta Released: Performance & Multi-GPU Updates

Unsloth v0.1.916-beta is released with dual-GPU acceleration, faster image/video generation, UI improvements, and inf ...

New Models

ByteDance-Seed Releases Cola2 Language Model

ByteDance-Seed has released Cola2, a new language model combining semantic compression via a continuous latent space ...

Engines and Tools

Ollama v0.40.15 Released with Parallel Vision GGUF Support

Ollama v0.40.15 is out, featuring parallel request handling for recurrent and vision GGUF models, Qwen3 YaRN scaling, ...

Community

Pyramid-JiT: A 2B Text-to-Image Model Trained for $21K

Explore Pyramid-JiT, a 2B text-to-image model by Linum AI trained for $21K without a VAE. Read community reactions an ...