Falcon H1 34B Instruct
Falcon H1 34B Instruct is the largest original H1 release. The multilingual model combines attention and Mamba in a downloadable text-generation checkpoint.
The largest original H1 layout
The configuration specifies 72 layers, a hidden width of 5,120 and 262,144 positions. Base and instruction-tuned versions are published separately.Falcon-H1-34B-Instruct configurationFalcon H1 architecture and release
Use retrieval questions across the full document
Practical guidance. Place questions near the beginning, middle and end of a long source. Require the answer to identify supporting passages. This checks usable recall and grounding rather than treating the configuration's position count as a quality score.
Specifications & access
| Detail | Value and conditions |
|---|---|
| Access | Downloadable |
| Model identifiers | tiiuae/Falcon-H1-34B-InstructFalcon-H1-34B-Base |
| Context window | 262,144 tokens max_position_embeddings in the instruction checkpoint configuration; effective recall depends on the task. Falcon-H1-34B-Instruct configuration |
| License | Falcon-LLM LicenseFalcon-H1-34B-Instruct model card |
| Layers and width | 72 layers; hidden width 5120Falcon-H1-34B-Instruct configuration |
Related models
Falcon H1 7B Instruct is a multilingual hybrid model with a published 262,144-position configuration.
Launch date not confirmedDownloadable