Technology Innovation Institute (TII), Abu Dhabi, has released Falcon-H1R-7B, a 7B parameter reasoning specialized model that matches or exceeds many 14B to 47B reasoning models in math, code and general benchmarks, while staying compact and efficient. It builds on Falcon H1 7B Base and is available on Hugging Face under the Falcon-H1R collection. Falcon-H1R-7B is interesting because it combines 3 design choices in 1 system, a hybrid Transformer along with Mamba2 backbone, a very long context that reaches 256k tokens in standard vLLM deployments, and a training recipe that mixes supervised long form reasoning with reinforcement learning using GRPO. Hybrid…

NVIDIA has just released its new streaming English transcription model (Nemotron Speech ASR) built specifically for low latency voice agents and live captioning. The checkpoint nvidia/nemotron-speech-streaming-en-0.6b on Hugging Face combines a cache aware FastConformer encoder with an RNNT decoder, and is tuned for both streaming and batch workloads on modern NVIDIA GPUs. Model design, architecture and input assumptions Nemotron Speech ASR (Automatic Speech Recognition) is a 600M parameter model based on a cache aware FastConformer encoder with 24 layers and an RNNT decoder. The encoder uses aggressive 8x convolutional downsampling to reduce the number of time steps, which directly lowers…

In deep learning, classification models don’t just need to make predictions—they need to express confidence. That’s where the Softmax activation function comes in. Softmax takes the raw, unbounded scores produced by a neural network and transforms them into a well-defined probability distribution, making it possible to interpret each output as the likelihood of a specific class.  This property makes Softmax a cornerstone of multi-class classification tasks, from image recognition to language modeling. In this article, we’ll build an intuitive understanding of how Softmax works and why its implementation details matter more than they first appear. Check out the FULL CODES here.…

In this tutorial, we build a genuinely advanced Agentic AI system using LangGraph and OpenAI models by going beyond simple planner, executor loops. We implement adaptive deliberation, where the agent dynamically decides between fast and deep reasoning; a Zettelkasten-style agentic memory graph that stores atomic knowledge and automatically links related experiences; and a governed tool-use mechanism that enforces constraints during execution. By combining structured state management, memory-aware retrieval, reflexive learning, and controlled tool invocation, we demonstrate how modern agentic systems can reason, act, learn, and evolve rather than respond in a single pass. Check out the FULL CODES here. !pip -q…

Liquid AI has introduced LFM2.5, a new generation of small foundation models built on the LFM2 architecture and focused at on device and edge deployments. The model family includes LFM2.5-1.2B-Base and LFM2.5-1.2B-Instruct and extends to Japanese, vision language, and audio language variants. It is released as open weights on Hugging Face and exposed through the LEAP platform. Architecture and training recipe LFM2.5 keeps the hybrid LFM2 architecture that was designed for fast and memory efficient inference on CPUs and NPUs and scales the data and post training pipeline. Pretraining for the 1.2 billion parameter backbone is extended from 10T to…

Marktechpost has released AI2025Dev, its 2025 analytics platform (available to AI Devs and Researchers without any signup or login) designed to convert the year’s AI activity into a queryable dataset spanning model releases, openness, training scale, benchmark performance, and ecosystem participants. Marktechpost is a California based AI news platform covering machine learning, deep learning, and data science research. What’s new in this release The 2025 release of AI2025Dev expands coverage across two layers: Release analytics, focusing on model and framework launches, license posture, vendor activity, and feature level segmentation. Ecosystem indexes, including curated “Top 100” collections that connect models to…

Question: Imagine your company’s LLM API costs suddenly doubled last month. A deeper analysis shows that while user inputs look different at a text level, many of them are semantically similar. As an engineer, how would you identify and reduce this redundancy without impacting response quality? What is Prompt Caching? Prompt caching is an optimization technique used in AI systems to improve speed and reduce cost. Instead of sending the same long instructions, documents, or examples to the model repeatedly, the system reuses previously processed prompt content such as static instructions, prompt prefixes, or shared context. This helps save both…

In this tutorial, we build an advanced multi-agent incident response system using AgentScope. We orchestrate multiple ReAct agents, each with a clearly defined role such as routing, triage, analysis, writing, and review, and connect them through structured routing and a shared message hub. By integrating OpenAI models, lightweight tool calling, and a simple internal runbook, we demonstrate how complex, real-world agentic workflows can be composed in pure Python without heavy infrastructure or brittle glue code. Check out the FULL CODES here. !pip -q install "agentscope>=0.1.5" pydantic nest_asyncio import os, json, re from getpass import getpass from typing import Literal from pydantic…

Zlab Princeton researchers have released LLM-Pruning Collection, a JAX based repository that consolidates major pruning algorithms for large language models into a single, reproducible framework. It targets one concrete goal, make it easy to compare block level, layer level and weight level pruning methods under a consistent training and evaluation stack on both GPUs and TPUs. What LLM-Pruning Collection Contains? It is described as a JAX based repo for LLM pruning. It is organized into three main directories: pruning holds implementations for several pruning methods: Minitron, ShortGPT, Wanda, SparseGPT, Magnitude, Sheared Llama and LLM-Pruner. training provides integration with FMS-FSDP for…

Tencent Hunyuan researchers have released HY-MT1.5, a multilingual machine translation family that targets both mobile devices and cloud systems with the same training recipe and metrics. HY-MT1.5 consists of 2 translation models, HY-MT1.5-1.8B and HY-MT1.5-7B, supports mutual translation across 33 languages with 5 ethnic and dialect variations, and is available on GitHub and Hugging Face under open weights. Model family and deployment targets HY-MT1.5-7B is an upgraded version of the WMT25 championship system Hunyuan-MT-7B. It is optimized for explanatory translation and mixed language scenarios, and adds native support for terminology intervention, contextual translation and formatted translation. HY-MT1.5-1.8B is the compact…