Models & Labs
Granite 4.2 LLMs Introduced by IBM and Hugging Face
Hugging Face Bloghigh confidence
Why it matters
- →Granite 4.2 models enhance reasoning capabilities with a unique training strategy.
- →The agentic RL feature allows models to perform complex tasks in real environments.
- →OpenAI-compatible endpoints and native tool calling increase accessibility for developers.
IBM and Hugging Face have released Granite 4.2, a new family of reasoning-focused language models available in three sizes: 3B, 8B, and 30B. These models are trained on 15 trillion tokens and feature a five-phase training strategy that extends their context window to 512K tokens. The 8B and 30B models include agentic reinforcement learning, allowing them to perform tasks like code editing and web searching in real environments. Released under the Apache 2.0 license, Granite 4.2 models support native tool calling and are compatible with OpenAI endpoints, making them a versatile option for developers.
Read originalMore from Hugging Face Blog
© Hugging Face BlogResearchresearch
4-bit Model Outperforms Full-Precision Original
Hugging Face's latest research introduces Quantization-Aware Healing (QAH), a method that allows a compressed, 4-bit model to outperform its full-precision counterpart. By distilling directly from the original, pre-compression model, QAH avoids the limitations of traditional quantization-aware training. This approach not only enhances accuracy but also improves training stability, as demonstrated by a GPT-OSS 120B model that excels on 7 out of 9 benchmarks. The innovation lies in using a full-size, full-precision teacher to guide the smaller, quantized student, resulting in a model that is both efficient and highly capable.
Hugging Face Blog
© Hugging Face BlogCoding Toolscoding
Gradio Introduces AI Workflow Automation
Gradio's new feature, gr.Workflow, transforms AI workflows by allowing users to create complex pipelines with a simple drag-and-drop interface. This innovation enables users to build, run, and deploy AI applications seamlessly, turning each workflow into a REST API with minimal effort. By integrating with Hugging Face Inference Providers and Gradio Spaces, users can execute multi-step processes and access intermediate results in real-time. This development significantly lowers the barrier for developers to create and deploy sophisticated AI applications, making it easier to experiment and iterate on ideas.
Hugging Face Blog
More in Models & Labs
Models & Labsmodels
llama.cpp b10618 Release Enhances Parsing
The b10618 release of llama.cpp tackles a crucial parsing issue, specifically improving the handling of hyphens in character classes. This update ensures that generated tool-call grammars are parsed correctly, enhancing the software's reliability. With new parser and integration tests included, the release verifies these improvements effectively. While it doesn't introduce major new features, this update strengthens llama.cpp's core functionality, making it more dependable for developers working on different operating systems and hardware configurations.
llama.cpp Releases
Models & Labsmodels
llama.cpp b10620 Release Expands Platform Support
The b10620 release of llama.cpp marks another step in broadening its platform reach, now supporting systems like Ubuntu with Vulkan and ROCm 7.14, alongside Windows with CUDA 13. This update underscores llama.cpp's adaptability, making it a go-to tool for developers working across various hardware configurations, from macOS Apple Silicon to Windows arm64. While the release doesn't introduce new groundbreaking features, it reinforces llama.cpp's role as a flexible inference runtime. By ensuring compatibility with more systems, llama.cpp becomes increasingly accessible to developers, allowing them to leverage its capabilities regardless of their hardware setup.
llama.cpp Releases
Models & Labsmodels
llama.cpp releases version 0.3.0
The latest llama.cpp release, version 0.3.0, brings a notable expansion in platform compatibility and functionality. This update enhances support for macOS, Linux, Windows, and openEuler, accommodating architectures like Apple Silicon and Vulkan. Developers will find the inclusion of CUDA 13 on Windows, albeit in preview, and ROCm 7.14 on Ubuntu particularly useful. While the release doesn't introduce groundbreaking features, it solidifies llama.cpp's role as a flexible tool for developers working across different computing environments. The update ensures that llama.cpp remains a reliable choice for those needing robust support across multiple systems.
llama.cpp Releases