Huawei Ascend
PulseAugur coverage of Huawei Ascend — every cluster mentioning Huawei Ascend across labs, papers, and developer communities, ranked by signal.
- developed by Huawei 100%
- used by DeepInfer: Open-Source Deep Learning Deployment Toolkit for Image-Guided Therapy 90%
- instance of Ascend 950 90%
- used by Meituan 90%
- developed Cann 90%
- partners with University of Science and Technology of China 90%
- used by LongCat-2.0 90%
- developed by Cann 70%
- affiliated with Cann 70%
- used by Tsinghua University 70%
- used by Shenzhen 70%
- instance of alphaXiv 70%
12 day(s) with sentiment data
-
Zhejiang University student's MoWorld 2.0 integrates 10+ physics elements, optimized for Huawei Ascend
Moxinc Technology, founded by Zhejiang University PhD student Chen Tianrun, has unveiled MoWorld 2.0, a new general-purpose 4D world model. This iteration enhances the modeling of physical laws, integrating over ten phy…
-
Moxin Technology unveils MoWorld 2.0 with physics simulation
Moxin Technology has released MoWorld 2.0, a 4D world model capable of simulating physics within its scenes. This advanced model incorporates elements like gravity, elasticity, and fluid dynamics into its 4D Gaussian re…
-
Huawei Ascend enables day-zero support for Moonshot AI's Kimi K3.1 Pro
Huawei's Ascend team has achieved day-zero inference support for Moonshot AI's Kimi K3.1 Pro, a 2.8-trillion-parameter model. This integration allows the Kimi K3.1 Pro model to run on Huawei's Atlas A3 and Ascend 950 Su…
-
DeepSeek plans V5 model training on Huawei Ascend, challenging US export controls
DeepSeek, a Chinese AI company, is reportedly planning to train its next-generation V5 foundation model using Huawei Ascend processors. This move challenges the assumption that China's AI development is constrained by e…
-
GPUWorks open-sources 40MB platform to pool diverse AI accelerators
GPUWorks has open-sourced a compact 40MB AI inference platform designed to pool various accelerators, including Nvidia, Huawei Ascend, T-Head PPU, and Hygon DCU, into a single cluster. This platform aims to streamline t…
-
DeepSeek V5 to leverage Huawei Ascend chips for pre-training
DeepSeek is planning to use Huawei's Ascend chips for the pre-training of its upcoming DeepSeek V5 large model. This marks a significant shift from previous uses of Ascend chips, which were primarily for inference tasks…
-
SAGE enhances Vision Transformers for universal vision recognition
Researchers have introduced SAGE, a Saliency-Annealed Gated Encoder designed to enhance Vision Transformers (ViTs) for universal vision recognition tasks. SAGE integrates a saliency-guided architecture that leverages DI…
-
DeepSeek open-sources DeepInfer, supporting Nvidia and Huawei chips
DeepSeek has released DeepInfer, an open-source inference engine designed to support both Nvidia and Huawei Ascend hardware. This move aims to facilitate cross-hardware deployment for Chinese enterprises, potentially ci…
-
MiniMax releases 428B coding model, but self-hosting remains costly
MiniMax has released the open weights for its 428B-parameter coding and agent model, M3.1 Flash. The model boasts a 1 million token context window and multimodal input capabilities, targeting coding and AI agent tasks. …
-
Huawei Launches Open Ascend Developer Base with CANNLab, CANNBot, and CANNJudge
Huawei has unveiled its open Ascend developer base, featuring CANNLab, CANNBot, and CANNJudge. CANNLab offers free cloud environments for developers, while CANNBot acts as a multi-agent assistant. CANNJudge provides cha…
-
DeepSeek open-sources DeepInfer inference engine with advanced caching
DeepSeek has open-sourced DeepInfer, an inference engine designed for production deployment. This toolkit, built upon vLLM with contributions from SGLang and TRRTLLM, offers advanced features like 72-hour prompt caching…
-
New AI method uses knowledge base for NPU kernel generation
Researchers have developed ASCENDEVOLVE, a novel approach to generating NPU kernels that goes beyond relying solely on experience. This new method integrates a large external knowledge base, derived from platform docume…
-
AI ethics, DeepMind's dual role, and China's chip independence debated
A series of Mastodon posts highlight ethical concerns regarding AI development, particularly focusing on Google DeepMind's work and its potential applications in warfare, contrasted with its contributions to protein str…
-
Preventing DeepInfer Latency Spikes on Mixed NVIDIA and Ascend Clusters
This article discusses strategies for preventing latency spikes in DeepInfer, an open-source deep learning deployment toolkit, particularly when operating on mixed clusters of NVIDIA and Huawei Ascend hardware. It highl…
-
Four chip platforms back StepFun's Step 5 Preview on launch day
Four domestic chip manufacturers, including Huawei Ascend, MetaX, Biren Technology, and Moore Threads, have announced Day-0 support for StepFun's Step 5 Preview. This coincides with the launch of Huawei's first 1,024-ca…
-
Intern Lumina U2: Diffusion LLM for Omni-Visual Understanding
Researchers have developed Intern Lumina U2, a novel diffusion large language model designed for omni-visual understanding and image generation. This model uniquely represents multi-codebook visual tokens in a fully dis…
-
Intern Lumina U2: Unified Omni-Visual LLM with Multi-Codebook Tokens
Researchers have introduced Intern Lumina U2, a novel diffusion large language model designed for comprehensive omni-visual understanding and generation across images, videos, and 3D assets. This model uniquely employs …
-
Anthropic launches Claude Haiku 5.5 with 75% cost reduction · 2 sources tracked
Anthropic has released Claude Haiku 5.5, positioning it as their fastest, most affordable, and capable small AI model to date. This new model is optimized for high-volume tasks such as summarization and classification, …
-
GLM-5.3-Flash model optimized for Huawei Ascend 310P hardware
A user has successfully optimized the GLM-5.3-Flash model, a 320-billion-parameter Mixture of Experts model, to run on Huawei Ascend 310P hardware. Initial performance was slow, but through various optimizations, the us…
-
Google AI chatbot AMIE shows promise in improving doctor-patient interactions · The Lancet study
A study published in The Lancet, co-led by Google and Beth Israel Deaconess Medical Center, suggests AI can improve patient-doctor relationships. The research utilized Google's diagnostic AI chatbot, AMIE, with 98 patie…