Falcon E 1B Instruct
Falcon E 1B Instruct is a ternary-weight language model for local text generation. Its BitNet design uses weights with three values. The release also includes base and fine-tuning variants.
Choose the appropriate weight form
The release's detailed table reports about 1.8B parameters for the checkpoint named 1B. It provides native BitNet, bfloat16 and prequantized forms. The prequantized form is intended for further training with TII's onebitllms tools.Falcon-E-1B-Instruct model cardFalcon Edge training and available variants
Measure the local workflow
Practical guidance. Compare the actual file format and runtime before comparing memory. A bfloat16 copy does not have the same storage cost as the ternary artifact. Test short interactive requests on the intended CPU, including the first model load.
Specifications & access
| Detail | Value and conditions |
|---|---|
| Access | Downloadable |
| Model identifiers | tiiuae/Falcon-E-1B-InstructFalcon-E-1B-Base |
| Context window | 32,768 tokens max_position_embeddings in the published instruction configuration. Falcon-E-1B-Instruct configuration |
| License | Falcon-LLM LicenseFalcon-E-1B-Instruct model card |
| Weight representation | Ternary values: -1, 0 and 1Falcon Edge training and available variants |
Related models
- Falcon E 3B Instruct
Technology Innovation Institute (TII) ยท Falcon Edge
A 3B English language model trained with ternary weights for local text generation.
Launch date not confirmedDownloadable