Tag
#local-ai
2posts

AI & Agents9 min read
Gemma 12B vs DeepSeek 763B MoE: Extreme Local AI Benchmark on an RTX 4070
A head-to-head empirical battle between Gemma 4 12B Heretic (50 tok/s in VRAM) and DeepSeek-V4.1-Flash 763B MoE (3.1 tok/s over NVMe SSD) on a 12GB GeForce GPU.
#deepseek#gemma#local-ai

AI & Agents13 min read
CosyVoice Explained: Zero-Shot Voice Cloning & Streaming TTS
How Alibaba's CosyVoice achieves zero-shot multilingual voice cloning, bi-streaming at 150ms latency, and pronunciation inpainting using flow matching and LLMs.
#cosyvoice#text-to-speech#voice-cloning