The canonical repositories
- tiiuae/falcon-11B: the text checkpoint, raw safetensors weights in bfloat16.
- tiiuae/falcon-11B-vlm: the vision-language checkpoint, LLaVA-NeXT architecture.
- Both are public and not gated, so no access request is required.
What you are downloading
Expect roughly 22 GB of weights for an 11B model in bfloat16, plus tokenizer and configuration files. Quantised community builds are much smaller, in the 7-8 GB range at 4-bit, and they are separate repositories rather than versions of the official one.
Before you trust a quantisation
- Check the conversion names a specific source revision of tiiuae/falcon-11B.
- Prefer a build with a stated quantisation method and tool version.
- Download from the official repositories first if you are going to publish quality claims.
- Record which build you evaluated, because results do not transfer between them.
After the download
Loading instructions differ by runtime: transformers for the reference path, a serving engine for throughput, and llama.cpp or Ollama for quantised local use. Pick the runtime from your memory budget rather than the other way around.
Related pages
Sources
falcon2.lol is an independent third-party site. It is not affiliated with, endorsed by, sponsored by or operated by the Technology Innovation Institute, and it does not speak for TII. Every number on this page is attributed to the source listed above.