Falcon-Edge: A series of powerful, universal, fine-tunable 1.58bit language models.
Hugging Face Blog
Falcon-Edge released language models in 1.58-bit ternary format available in 1 billion and 3 billion parameter sizes for both inference and fine-tuning. The models were pre-trained on 1.5 trillion tokens and released with bfloat16 variants, pre-quantized BitNet weights, and a Python package called onebitllms to enable community fine-tuning. The release enables developers to deploy compressed models on edge devices while maintaining competitive performance on standard benchmarks.
