MarkTechPost news
The latest MarkTechPost news, : Zyphra releases ZHA, a sparse attention method for frozen models.
MarkTechPost publishes technical tutorials and guides for AI developers. Recent coverage includes a demonstration of running the Underdog Saluki 27B quantized model and a tutorial on multimodal data augmentation using the AugLy library. These resources focus on practical implementation details for quantized models and adversarial robustness benchmarking.
12 stories mentioning MarkTechPost since 25 September 2026, newest first, each with its sources and discussion. Follow to see new ones on your front page.
-
Zyphra releases ZHA, a sparse attention method for frozen models
Zyphra has released Zyphra Hierarchical Attention (ZHA), a sparse attention method designed to reduce the computational cost of long-context processing. Unlike existing sparse methods that still scan the entire context…
1 sourceMarkTechPost -
Meet LEANN: An Open-Source Vector Index that Shrinks a 60M-Chunk RAG Index from 201 GB to 6 GB
UC Berkeley’s Sky Computing Lab introduced LEANN, an open-source vector index that reduces storage needs by recomputing embeddings at query time instead of storing them. A 60 million-chunk Wikipedia index shrinks from…
1 sourceMarkTechPost -
MarkTechPost publishes tutorial on running Underdog Saluki 27B quantized model
MarkTechPost published a technical tutorial demonstrating how to run Underdog Saluki 27B, a 7.9 GB quantized version of Qwen3.8-27B that claims to preserve tool-calling ability at roughly two bits per weight. Both…
1 sourceMarkTechPost -
DeepSeek open-sources DeepInfer inference framework for DeepSeek-V4.1-Flash
DeepSeek has open-sourced DeepInfer, its internal AI inference framework, under the Apache 2.0 license, making it available to developers. The framework powers DeepSeek-V4.1-Flash and runs on NVIDIA SM100 GPUs and…
7 sourcesLatent SpaceThe DecoderMarkTechPostbloomberg.com +3 more -
Datalab releases OmniParseBench, an open OCR benchmark with 16,288 tests
Datalab has released OmniParseBench, an open-source benchmark designed to evaluate Optical Character Recognition (OCR) parsers. The dataset consists of 16,288 pass/fail tests derived from 2,937 pages across 2,343…
1 sourceMarkTechPost -
Alibaba’s QwenGyre speeds up long-horizon agent training by up to 1.85x
Alibaba’s Qwen team introduced QwenGyre, an elastic reinforcement learning framework designed for extreme-long-horizon agent training. It targets tasks running for hours and processing nearly 1M tokens per rollout,…
1 sourceMarkTechPost -
Datalab releases OmniExtractBench to audit PDF extraction accuracy
Datalab launched OmniExtractBench, an open benchmark for structured document extraction from PDFs into JSON. It pools 620 documents from four existing datasets, including regulatory filings and synthetic tests, to…
1 sourceMarkTechPost -
Liquid AI launches d1 decision model for structured tasks with zero output tokens
Liquid AI released d1, a decision model designed for structured choices like classification, ticket routing, and moderation. Unlike text-generating models, d1 returns calibrated probabilities across predefined outcomes…
1 sourceMarkTechPost -
Alibaba releases Qwen-Audio-3.1-Realtime for voice agents at 85% lower prices
Alibaba’s Qwen team launched Qwen-Audio-3.1-Realtime, a full-duplex speech model designed for voice agents that can reason, call tools, and manage turn-taking in real-time conversations. The model is part of a…
1 sourceMarkTechPost -
Sarvam AI launches Saaras V4 speech-to-text model for 22 Indian languages and global English
Sarvam AI released Saaras V4, a speech-to-text model supporting all 22 scheduled Indian languages plus global English accents. The encoder-decoder system uses a 3B-parameter hybrid state-space language model…
1 sourceMarkTechPost -
MarkTechPost publishes tutorial on multimodal data augmentation with AugLy library
MarkTechPost published a tutorial demonstrating end-to-end multimodal data augmentation and adversarial robustness benchmarking using the AugLy library. The tutorial covers image, text, and audio augmentation workflows…
1 sourceMarkTechPost -
Fastino Releases GLiNER2.5-Decide: A 340M Open-Weight Decision Model That Runs on CPU
The model takes text and a schema of typed questions and returns answers with probability distributions, confidence scores, and constraint-feasibility metadata. It is built on a DeBERTa-v3-large encoder, fine-tuned…
1 sourceMarkTechPost
Questions about MarkTechPost
What is the latest MarkTechPost news?
As of 3 November 2026: Zyphra releases ZHA, a sparse attention method for frozen models. Zyphra has released Zyphra Hierarchical Attention (ZHA), a sparse attention method designed to reduce the computational cost of long-context processing.
What is MarkTechPost?
MarkTechPost publishes technical tutorials and guides for AI developers. Recent coverage includes a demonstration of running the Underdog Saluki 27B quantized model and a tutorial on multimodal data augmentation using the AugLy library. These resources focus on practical implementation details for quantized models and adversarial robustness benchmarking.