Tech Meridian
← ENTITY INDEX
COMPANY · ENTITY #3451
Related event timeline, sources and context from the news index.
This arXiv paper (2610.41112v1) evaluates three calibration approaches—HiGPTQ, HiAutoRound, and HiFlatQuant—for the HiFloat4 (HiF4) 4-bit block floating-point format. The authors report that these methods narrow the accuracy gap to BF16 to under 1% and outperform NVFP4 across a range of models and downstream tasks compared with direct HiF4 conversion.
Recursive, a startup led by Richard Socher and co‑founded by Jeff Clune, is offering Flint, an open‑weight 40‑billion‑parameter coding model the company says was largely developed and optimized by its recursive AI system called the Eureka Machine. Recursive shared benchmark claims that Flint outperforms peers on coding tasks and has raised $650 million at a $4.65 billion valuation from investors including Google Ventures, Greycroft, Nvidia, and AMD Ventures.
Black Forest Labs published FLUX.3 Image Klein to the Hugging Face Hub and ships open checkpoints that run in Hugging Face Diffusers. The model supports text-to-image and multi-reference image-to-image generation, adds bounding-box layout control for region placement and image edits, includes a distilled checkpoint (1024×1024 ~2s on an H100, 16 steps by default), quantized single-file transformer weights (bf16, fp8r, nvfp4), and LoRA/DreamBooth training scripts.
The MLX array framework added Apple Silicon support and demonstrates running frontier LLMs locally and in a distributed setup on Mac Studio. Using the latest MLX release and RDMA over Thunderbolt, a four‑node Mac Studio cluster runs Muse Spark Open (816B, MoE, NVFP4) at >36 tokens/s; smaller models (Qwen 3.8 and Mistral Medium 3.5) can fit on a single 512GB Mac Studio. Benchmarks with mlx_lm.benchmark report up to 4.6× faster time‑to‑first‑token and ~1.6× higher tokens/s on M5 Ultra versus M3 Ultra, and modest peak power (~200–300W) for inference.
NVIDIA's TensorRT Edge-LLM ran Qwen3.6-27B on a single Jetson AGX Thor Developer Kit and achieved 52.33 tokens/sec in the MLPerf Inference v6.1 Edge Agentic performance workload, completing all 1,007 turns in 24 minutes 36 seconds — 6.4× faster than the llama.cpp Jetson reference (2h37m). The submission used NVFP4 quantization for weights/activations, FP8 for the KV cache, tree-based multi-token prediction, and KV-cache/recurrent-state reuse to accelerate long-context agent decoding.
Allow Google Analytics to measure visits? It stays off until you choose. You can change your choice at any time.