🚀 AIME G500 Blackwell Pro AI Workstation 🚀
The AIME G500 is the most powerful HPC workstation worldwide, designed as maintainable high-end workstation with enough cooling and PSU capacity to host up to four high-end GPUs.
AIME
AIME develops AI-machines for deep learning: Multi-GPU workstations & HPC servers. We also provide GPU cloud compute, deep learning services & consulting.
AIME boosted:
Mistral Code is an AI-powered coding assistant that bundles powerful models, an in-IDE assistant, local deployment options, and enterprise tooling into one fully supported package, so developers can 10x their productivity with the full backing of their IT and security teams.
Mistral Code builds on the proven open-source project "Continue", reinforced with the controls and observability that large enterprises require; private beta is open today for VSCode.
AIME boosted:
Mistral released Devstral, an agentic LLM for software engineering tasks. Devstral is built under a collaboration between Mistral AI and All Hands AI. It outperforms all open-source models on SWE-Bench Verified by a large margin and was released under the Apache 2.0 license.
AIME boosted:
We updated our GPU Benchmark blog article, adding RTX 5090 and RTX Pro 6000 Blackwell WS.
With the AIME G500 workstation you can run two of the RTX Pro 6000 Blackwell WS GPUs - this makes it the world's most powerful workstation!
Link to workstation: https://www.aime.info/de/shop/product/aime-g500-workstation/?pid=G500-2XR6BW-75X-192R-2T-N-N-N
Link to benchmark article: https://www.aime.info/blog/en/deep-learning-gpu-benchmarks/
AIME boosted:
Qwen WorldPM-72B (World Preference Modeling) demonstrates that preference modeling follows similar scaling laws as language modeling.
Through large-scale training on 15M preference data, they reveal that preference models can learn unified preference representations.
AIME boosted:
Stanford released The 2025 AI Index Report.
The AI Index offers one of the most comprehensive, data-driven views of artificial intelligence.
Recognized as a trusted resource by global media, governments, and leading companies, the AI Index equips policymakers, business leaders, and the public with rigorous, objective insights into AI’s technical progress, economic influence, and societal impact.
AIME boosted:
xDiT is a scalable inference Engine for diffusion transformers (DiTs). Think of it as vLLM for Pixels (IMages, Video).
xDiT is an inference engine designed for the parallel deployment of DiTs on a large scale. xDiT provides a suite of efficient parallel approaches for Diffusion Models, as well as computation accelerations.
AIME boosted:
ACE-Step is an open-source foundation model for music AI that empowers artists, producers, and creators to generate music from text prompts.
Current music generation methods often sacrifice speed, coherence, or control. ACE-Step changes the game by combining the best of diffusion-based generation, advanced compression, and transformer technology.
AIME boosted:
TNG Tech created a LLM-Chimera by merging DeepSeek-R1 and DeepSeek-V3, combining the reasoning capabilities of R1 with the token efficiency improvements of V3.
It merges weights from both source models to balance performance in terms of reasoning, efficiency, and instruction compliance. It is released under the MIT License and is intended for research and commercial use.
In benchmarks, it appears to be as smart as R1 but much faster, using 40% fewer output tokens.
AIME boosted:
Vidi: Large Multimodal Models for Video Understanding and Editing
Bytedance, the company behind TikTok, released a new Large Multimodal Model family called Vidi that addresses video editing challenges by enabling temporal retrieval—identifying time ranges in hour-long videos linked to text queries.
Vidi outperforms proprietary models like GPT-4o and Gemini in retrieval tasks, demonstrating superior efficacy in video editing scenarios.
AIME boosted:
Alibaba released Qwen3 LLM
Their new dense and Mixture-of-Experts (MoE) models (0.6B, 1.7B, 4B, 8B, 14B, 32B, 30B-A3B, 235B-A22B) enable seamless switching between reasoning-intensive tasks (math, coding) and efficient general-purpose chat.
Quantization, Training and Framework integration for RAG, agents, and application-specific use cases are documented. Models and documentation are publicly available, emphasizing accessibility and cost-effective scaling.
AIME boosted:
S3MOT: Monocular 3D Object Tracking with Selective State Space Model
A new study introduces advancements in monocular 3D multi-object tracking (MOT) to address challenges in spatiotemporal association from 2D video streams.
Paper: https://lnkd.in/eXjpBhq7
Code: https://lnkd.in/eivtvnZh
AIME boosted:
Microsoft released a 1-Bit-DeepSeek derivate which runs on CPU.
AIME boosted:
Google released Gemma 3 QAT Models, introducing Quantization-Aware Training (QAT), enabling state-of-the-art AI performance on consumer-grade GPUs.
The new models reduce memory requirements by up to 75% while maintaining high quality, making it possible to run larger models on devices with limited VRAM.
Key highlights:
* Gemma 3 27B model can run on a single NVIDIA RTX 3090 GPU
* Official int4 and Q4_0 models
* Integration with popular tools like Ollama, vLLM
AIME boosted:
Skywork-OR1 (Open Reasoner 1) is a new SOTA 32B model family with open weights, training code and training data from China which includes two general-purpose reasoning models: Skywork-OR1-7B and Skywork-OR1-32B as well as a math-specialized model, skywork-OR1-Math-7B.
These models are fine tunings of deepseek-ai/DeepSeek-R1-Distill-Qwen-7B and deepseek-ai/DeepSeek-R1-Distill-Qwen-32B.
They show state-of-the-art reasoning performance at every model size.
AIME boosted:
Another open source model from China: Kimina-Prover Preview is "the first large formal reasoning model that can reason in a human-like way and prove mathematical theorems rigorously in the Lean 4 language."
AIME boosted:
HiDream-I1 is a new open-source image generative foundation model from China with 17B parameters that achieves state-of-the-art image generation quality within seconds.
AIME boosted:
Meta released Llama 4, a collection of mixture of experts (MoE) using alternating dense and MoE layers for inference efficiency.
17 billion parameter models offer a context window of 10M.
AIME boosted:
Bytedance (TikTok) released a paper on Human Image Animation: "DreamActor-M1: Holistic, Expressive and Robust Human Image Animation with Hybrid Guidance"
Find paper and videos here: https://grisoon.github.io/DreamActor-M1/
AIME boosted:
Deepseek dropped new model weights for V3.
https://huggingface.co/deepseek-ai/DeepSeek-V3-0324/tree/main
Client Info
Server: https://mastodon.social
Version: 2025.04
Repository: https://github.com/cyevgeniy/lmst