Identity Matrix Outperforms DeepSeek's mHC in New Qwen3 Training Experiments
Tencent Hunyuan's Hy4 preview cited a Zhihu post challenging DeepSeek's mHC: experiments on Qwen3 models show Identity-based head compression outperforms the do…
Global AI Digest
-
-
Hy4-preview Compressed to ~200GiB GGUF with Mixed-Precision Quantization, Accuracy Nearly Unchanged
AngelSlim releases Hy4-preview GGUF, compressing 1.5TB to ~200GiB using mixed-precision STQ1_0/IQ2_XXS. Benchmarks show minimal accuracy loss.
-
OpenAI Plans to Block Cursor Users From Accessing Its Models, Cursor Says
Cursor says OpenAI intends to block Cursor users from accessing OpenAI models in three months, while noting OpenAI models account for only about 5% of Cursor us…
-
MiniMax H3 Max Creates an Infinite AI-Generated Twitch Stream
A creator hooked MiniMax H3 Max, an AI video model that generates footage faster than real time, to a Twitch livestream, creating an endless 'interdimensional c…
-
OpenAI Ends Cursor Partnership After SpaceX Acquisition, Cutting Model Access on Nov 12
OpenAI announced it is ending its partnership with Cursor after Cursor's acquisition by SpaceX. Under the proposal, Cursor's direct access to OpenAI models will…
-
Midjourney Begins Testing V8.2 Edit Model with Instruction-Based Editing and Multi-Image References
Midjourney announced it is beginning to test its first V8.2 edit model, which adds instruction-based editing, up to 4 image references, brush-based inpainting a…
-
Tencent Hunyuan Hy4-preview Runs in vLLM from Day One on NVIDIA GPUs
Tencent Hunyuan's Hy4-preview is supported in vLLM from day 0, verified on NVIDIA GPUs. The model features 770B total parameters with 49B active, 256 routed exp…
-
Tencent Open-Sources Hy4 Preview: 770B-Parameter MoE with 1M Context
Tencent has open-sourced Hy4 preview, a flagship mixture-of-experts model with 770B total parameters, 49B active per token, and 1M context length. It outperform…
-
Hy4 Preview Released: 770B-Parameter Open-Source Model with 1M Context
Hy4 preview is now available, featuring 770B total parameters with 49B active, a 1M context window, and an open-source, productivity-focused design at a consist…
-
Perplexity Launches Portable Computer, a Fully Local AI Runtime on NVIDIA DGX Spark
Perplexity announced Portable Computer, a fully local version of its Computer product that runs entirely on NVIDIA DGX Spark with no cloud dependency.
