Led the release of the Falcon-H1 series (May 2025), featuring a novel hybrid Transformer–SSM architecture and achieving state-of-the-art performance across multiple scales (0.5B to 34B).
Released Falcon-Edge (May 2025), a series of powerful, universal, fine-tunable 1.58-bit language models.
Launched Falcon 3 (December 2024), including base and Instruct models from 1B to 10B, plus an advanced 7B Mamba variant.
Released Falcon Mamba 7B (August 2024), the first strong attention-free 7B language model, with a technical report published in October 2024.
Published key preprints on arXiv, including 'Falcon-H1' (arXiv’25) and 'Falcon Mamba' (arXiv’24).
Co-organized the NeurIPS 2025 E2LM Competition (June 2025) on early-stage training evaluation of LLMs.
Served on program committees for ECAI 2024, CIKM 2023/2024, ECML-PKDD 2020, and as a reviewer for journals such as TKDE, Information Sciences, etc.
Previously served as Web Chair for MDM 2020 and on the junior organizing committee for JDSE 2019.