Falcon H1 1.5B Instruct
Falcon H1 1.5B Instruct is a multilingual hybrid model with downloadable weights. It is the wider, shallower member of the two original 1.5B H1 designs.
The standard 1.5B layout
The configuration has 24 layers and a hidden width of 2,048, with 131,072 positions. It differs from the 66-layer Deep release despite their similar nominal parameter counts.Falcon-H1-1.5B-Instruct configuration
Compare equal-sized models on latency
Practical guidance. Compare this checkpoint with 1.5B Deep using the same prompt length and precision. Include time to first token as well as answer quality. Similar parameter totals do not make different layer layouts behave identically.
Specifications & access
| Detail | Value and conditions |
|---|---|
| Access | Downloadable |
| Model identifiers | tiiuae/Falcon-H1-1.5B-InstructFalcon-H1-1.5B-Base |
| Context window | 131,072 tokens max_position_embeddings in the instruction checkpoint configuration; effective recall depends on the task. Falcon-H1-1.5B-Instruct configuration |
| License | Falcon-LLM LicenseFalcon-H1-1.5B-Instruct model card |
| Layers and width | 24 layers; hidden width 2048Falcon-H1-1.5B-Instruct configuration |
Related models
A 1.5B multilingual hybrid language model with a deeper, narrower design than standard Falcon H1 1.5B.
Launch date not confirmedDownloadable