6 21 hours ago

1.42B open-weight Llama-architecture language and coding model with a custom 49k BPE tokenizer and 32K context window for fast CPU and mobile inference.