AI Squawk the real-time terminal for AI news

15h ago TOOLING [EPLB] Support heterogeneous DeepSeek V4/V4.1 DSpark draft experts #

23h ago TOOLING [Model] Use FlashInfer for Gemma 4 with an NVFP4 KV cache when FA4 is unavailable #

25h ago TOOLING [ROCm][Docs] Add DeepSeek-V4.1-Flash MI355X DP2 AgentX preset with UMBP KV offload #

2d ago TOOLING [XPU] support adaptive verification for deepseek-v4 #

3d ago TOOLING [MM] Support mm_device_do_normalize for Gemma 4 Unified #

4d ago TOOLING [ROCm][UMBP] Support UMBP standalone mode for DeepSeek-V4.1-Flash #

5d ago TOOLING [DeepSeek-V4] Support sequence parallelism with pipeline parallelism for MegaMoE #

8d ago TOOLING [Model] Support data-parallel multimodal encoder for Gemma-4 #