Falcon Mamba 7B Instruct
Falcon Mamba 7B Instruct is the earlier pure state-space Falcon model. The weights remain published, and the card points to Falcon3 Mamba as a newer release.
Training and release background
The model was trained on about 5.5 trillion tokens, mainly from filtered web data with code and other curated sources. Training length increased from 2,048 to 8,192 tokens. The card identifies TII Falcon-Mamba License 2.0 and provides CPU and GPU examples.Falcon Mamba 7B Instruct model card
Use it to reproduce an earlier design
Practical guidance. Keep this version when reproducing work that names the original Falcon Mamba checkpoint. For a new task, compare it with the later Falcon3 Mamba and hybrid H1 models. Preserve the version identifier so a model update does not change the experiment.
Specifications & access
| Detail | Value and conditions |
|---|---|
| Access | Downloadable |
| Model identifiers | tiiuae/falcon-mamba-7b-instruct |
| License | TII Falcon-Mamba License 2.0Falcon Mamba 7B Instruct model card |
| Release status | Earlier release; published weight access remains availableFalcon Mamba 7B Instruct model card |
Related models
- Falcon3 Mamba 7B Instruct
Technology Innovation Institute (TII) ยท Falcon Mamba
A 7B instruction-tuned language model that continues Falcon Mamba training with a pure Mamba1 architecture.
Launch date December 2024Downloadable