Falcon E 3B Instruct
Falcon E 3B Instruct is the larger Falcon Edge model for English text. It is a transformer trained for ternary weights and is available in several forms for inference and further training.
Choose the appropriate weight form
The model card documents BitNet, prequantized and bfloat16 variants. Fine-tuning uses the prequantized revision with onebitllms. Its configuration has 32 layers and a hidden width of 2,048. This differs from ordinary post-training quantization of a generic 3B checkpoint.Falcon-E-3B-Instruct model cardFalcon-E-3B-Instruct configuration
Measure the local workflow
Practical guidance. Use the same hardware and input lengths when comparing the 1B and 3B Edge releases. Check whether the application is limited by loading, prompt processing or token generation. Each stage can affect a local user's experience differently.
Specifications & access
| Detail | Value and conditions |
|---|---|
| Access | Downloadable |
| Model identifiers | tiiuae/Falcon-E-3B-InstructFalcon-E-3B-Base |
| Context window | 32,768 tokens max_position_embeddings in the published instruction configuration. Falcon-E-3B-Instruct configuration |
| License | Falcon-LLM LicenseFalcon-E-3B-Instruct model card |
| Weight representation | Ternary values: -1, 0 and 1Falcon Edge training and available variants |
Related models
- Falcon E 1B Instruct
Technology Innovation Institute (TII) ยท Falcon Edge
Falcon E 1B Instruct is a ternary-weight language model for local text generation.
Launch date not confirmedDownloadable