Falcon H1 7B Instruct
Falcon H1 7B Instruct is a multilingual hybrid model with a published 262,144-position configuration. It is separate from the later H1R 7B reasoning model.
H1 and H1R are different releases
The instruction model uses 44 layers and a hidden width of 3,072. H1R was later trained from H1-7B-Base with reasoning data and reinforcement learning, rather than being a new name for this Instruct checkpoint.Falcon-H1-7B-Instruct configurationFalcon H1R 7B model card
Compare reasoning separately
Practical guidance. For a task that needs a short direct answer, compare this checkpoint with H1R at the same output budget. Check answer completeness and latency. A reasoning model's extra output is useful only when it improves the required result.
Specifications & access
| Detail | Value and conditions |
|---|---|
| Access | Downloadable |
| Model identifiers | tiiuae/Falcon-H1-7B-InstructFalcon-H1-7B-Base |
| Context window | 262,144 tokens max_position_embeddings in the instruction checkpoint configuration; effective recall depends on the task. Falcon-H1-7B-Instruct configuration |
| License | Falcon-LLM LicenseFalcon-H1-7B-Instruct model card |
| Layers and width | 44 layers; hidden width 3072Falcon-H1-7B-Instruct configuration |
Related models
- Falcon H1R 7B
Technology Innovation Institute (TII) ยท Falcon H1R
Falcon H1R 7B is a reasoning-focused adaptation of Falcon-H1-7B-Base.
Launch date not confirmedDownloadable