Falcon LLM open-source large model

説明を展開

Falcon LLM was released by Abu Dhabi's Technology Innovation Institute (TII) in May 2023, starting from Falcon-40B and expanding through 2024-25 to include 180B, Mamba-7B and other variants — all open-sourced under Apache 2.0. Falcon has consistently held a Top-5 non-US/non-China spot on the Hugging Face Open LLM Leaderboard. Training compute is a hybrid of AWS plus local supercomputing. Assessment: Falcon is one of the few non-US/non-China open large models that can compete with Llama / Qwen, and it is essential to the UAE's "AI sovereignty" narrative. But Falcon lags US/China frontier models (GPT-4o, Claude 3.5, Qwen 2.5) by roughly a generation, and its GenAI commercial ecosystem (API consumption, developer community) trails Llama by a wide margin. The "sovereign large model" value is diplomatic, not commercial.

参考出典

• TII Falcon LLM announcement · 2023-05-25

同じ地域で掘り下げる