VIDRAFT POCKET has released a family of 180 billion parameter Mixture-of-Experts (MoE) models designed to run on consumer hardware, including laptops with as little as 8 GB of VRAM. These models utilize the GGUF format and standard runtimes like llama.cpp, Ollama, and LM Studio, requiring no cloud dependency or modified software. The flagship POCKET-Darwin-180B model achieves 87.65% quality retention at 4-bit quantization, and the collection has garnered over 3.15 million downloads on Hugging Face. AI
IMPACT Enables running large language models on consumer hardware, potentially democratizing access to advanced AI capabilities.
RANK_REASON Release of a specific model family designed for consumer hardware, not a frontier model release from a major lab.
- Hugging Face
- llama.cpp
- LM Studio
- NVIDIA H100
- Ollama
- POCKET-26B
- POCKET-Darwin-180B
- pocketpal
- VIDRAFT POCKET
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →