Granite 4.0 H-350M
Granite 4.0 H-350M is a compact instruction-tuned text model. IBM reports 340M parameters for this checkpoint. It belongs to the Nano releases and is adapted from its matching base model.
Model design and training
The hybrid architecture combines Mamba-2 with attention and uses no positional embeddings. Post-training includes supervised fine-tuning, reinforcement learning and model merging. This checkpoint generates text; it does not itself contain the speech components used by other Granite products.Granite 4.0 H-350M model card
Practical use
Practical guidance. The hybrid 350M model needs a compatible runtime even though the weights are small. Compare cold-start time and sustained request latency with the dense 350M checkpoint. This distinction matters for short tasks where loading and scheduling can exceed generation time.
Specifications & access
| Detail | Value and conditions |
|---|---|
| Access | Downloadable |
| Model identifiers | ibm-granite/granite-4.0-h-350m |
| Context window | 32,768 tokens Published model context; deployment settings can impose a smaller limit. Granite 4.0 H-350M model card |
| License | Apache 2.0Granite 4.0 H-350M model card |
| Parameters | 340MGranite 4.0 H-350M model card |
| Architecture | Four attention layers and 28 Mamba-2 layers; hidden size 768.Granite 4.0 H-350M model card |
Related models
- Granite 4.0 350M
IBM ยท Granite 4.0 Nano
A 350M-parameter instruction model in IBM's compact Granite Nano text family.
Launch date not confirmedDownloadable