ggml has released v1.0.0, marking the first formal versioned milestone for the tensor library underpinning the llama.cpp ecosystem. Its core value is a dependency-free C/C++ implementation spanning x86, ARM, RISC-V and more, with 2- to 8-bit integer quantization plus MXFP4 and NVFP4 microscaling formats and zero runtime memory allocation. For local-inference and edge-deployment developers, this signals long-term ecosystem stability.
MXFP4
4-bit microscaling floating-point format defined by OCP, using shared exponents to limit low-precision accuracy loss.
Curated Intelligence
Dutch chip startup Axelera AI has disclosed the architecture of Titania, its third-generation AIPU targeting data center inference, with mass production slated for Q3 2027 and over $300 million in orders already booked. Optimized for sparse MoE models, the chip delivers 2.5 PFLOPS at FP8, and a single 4U eight-card server can run a 1.6T MoE model locally, with the company claiming 7x per-watt throughput versus current data center servers.
OpenAI released gpt-oss-120b and gpt-oss-20b this week, marking its first open-weight LLM since GPT-2 in 2019. The article analyzes the architectural evolution of these models, highlighting how MXFP4 quantization optimizations enable local execution on single GPUs, while comparing width-versus-depth trade-offs against previous generations.