Google has recently released Genie 2, their most advanced large-scale foundation model capable of generating consistent and playable worlds for a minute. This advancement could be particularly beneficial for game developers, but potentially pose threats for current jobs.
DeepThought-8B: A Transparent Reasoning ModelDeepThought-8B is a transparent reasoning model constructed upon LLaMA-3.1, equipped with test-time compute scaling capabilities. It utilizes approximately 16GB of VRAM and demonstrates competitive performance with models that possess 70 billion parameters. The model’s weights are open-sourced, along with inference scripts. However, it’s still 2Rs in strawberry ;). Ollama is converting the model, so it will soon be available there, now you can get it from Hugging Face.
Ollama has introduced a new embedding model from Snowflake: snowflake-arctic-embed2. This model supports multilingual capabilities. https://ollama.com/library/snowflake-arctic-embed2