Search AI Products and News

Baidu officially released the ERNIE Bot 4.5 series models and made them fully open source. Users can experience this latest open-source technology immediately through ERNIE Bot (https://yiyan.baidu.com). This series includes multiple parameter configurations, such as Mixture of Experts (MoE) models with activated parameters of 47B and 3B, as well as dense models designed with 0.3B parameters, totaling ten different models. In terms of training and inference, the ERNIE 4.5 series models use PaddlePaddle deep learning.

On June 30, Huawei officially announced the open sourcing of the Pangu dense model with 7 billion parameters, the PanguPro MoE model with 72 billion parameters, and the model inference technology based on Ascend. This open-source initiative is a key step in Huawei's strategy to build an Ascend ecosystem, aiming to promote research and innovation in large model technology, accelerate the application of artificial intelligence across industries, and create value.

As enterprises accelerate the adoption of AI technologies, effectively managing complex AI workflows has become a key challenge. Latest industry analysis indicates that building a robust AI orchestration layer is an effective solution to address prompt chaos. The Dilemma in Choosing AI Orchestration Frameworks: Currently, there are many AI orchestration tools and frameworks on the market, including well-known solutions like LangChain, which presents a challenge for enterprise technology selection.

Family members, today I must tell you about a super cool new achievement in the academic world - Flow-GRPO! This thing is something special; it's like injecting a super evolution agent into image generation models, directly elevating them from 'Bronze' to 'King'. Do you want to know how it does that? Quickly grab your stools and listen carefully! The Growing Pains of Image Generation Models: Nowadays, image generation models, such as those based on flow matching, have a solid theoretical foundation and generate high-quality results.

Developed by TNG Technology Consulting, the DeepSeek R1T Chimera model has officially launched on the OpenRouter platform, providing global developers with efficient and powerful inference capabilities. This new open-source model combines the excellent inference capabilities of DeepSeek R1 with the high performance of V3-0324, marking another significant breakthrough in the balance of performance and efficiency in open-source AI technology. The following is compiled by AIbase.

At SenseTime's Technology Exchange Day on April 10th, the company launched its latest multimodal large model, SenseNova V6, and the SenseCore 2.0 system. This new version aims to integrate various information formats, including text, images, and videos, to provide users with a more natural and richer interactive experience. The SenseNova V6 series includes four versions, with SenseNova V6Pro being the most notable.

Meta has released its latest open-source AI model, Llama 4, marking another significant advancement in the field of artificial intelligence. Llama 4 comes in two versions, Scout and Maverick, designed to enhance AI model capabilities and performance. Meta states that Llama 4 is a multimodal large language model capable of processing various data types, including text, images, video, and audio, and can freely convert between these formats. Notably, the Llama 4 series is the first...

Ant Group's Ling team recently published a preprint on arXiv titled "Every FLOP Matters: Scaling a 300-billion parameter Mixture-of-Experts LING model without high-end GPUs," detailing two novel large language models: Ling-Lite and Ling-Plus. These models incorporate several innovations enabling efficient training on low-performance hardware, significantly reducing training costs.

ByteDance's Doubao large model team recently announced a breakthrough in addressing key bottlenecks in Mixture-of-Experts (MoE) architecture, open-sourcing a significant optimization technology called COMET. This technology dramatically improves large model training efficiency, achieving a remarkable 1.7x speedup and a 40% reduction in training costs. Image Note: Image generated by AI, image licensing provider Midjourney. COMET has been deployed in ByteDance's multi-thousand-GPU cluster training, resulting in millions of GPU hours saved.

DeepSeek announced its second-day product of the open-source week: the first open-source EP communication library for MoE models, enabling full-stack optimization for mixed-expert model training and inference. DeepEP is a high-efficiency communication library specifically designed for Mixture-of-Experts (MoE) and Expert Parallelism (EP). It aims to provide high-throughput and low-latency many-to-many GPU kernel communication, commonly known as MoE routing and aggregation. DeepEP not only supports low-precision operations such as FP8 but also integrates with the DeepSeek-V3 paper.