Granite 4.0 350M
Granite 4.0 350M is a compact instruction-tuned text model. IBM reports 350M parameters for this checkpoint. It belongs to the Nano releases and is adapted from its matching base model.
Model design and training
The dense architecture uses rotary position embeddings and grouped-query attention. IBM applies supervised fine-tuning, reinforcement learning and model merging to the corresponding base. The resulting checkpoint follows instructions rather than serving only as a pretrained completion model.Granite 4.0 350M model card
Practical use
Practical guidance. Use the 350M release for narrow text tasks on constrained devices. Provide a small output schema and the evidence needed for the answer. Evaluate incorrect extractions and missed fields before expanding to open-ended assistance.
Specifications & access
| Detail | Value and conditions |
|---|---|
| Access | Downloadable |
| Model identifiers | ibm-granite/granite-4.0-350m |
| Context window | 32,768 tokens Published model context; deployment settings can impose a smaller limit. Granite 4.0 350M model card |
| License | Apache 2.0Granite 4.0 350M model card |
| Parameters | 350MGranite 4.0 350M model card |
| Architecture | 28 attention layers; hidden size 1,024.Granite 4.0 350M model card |
Related models
- Granite 4.0 H-350M
IBM ยท Granite 4.0 Nano
A 340M-parameter hybrid instruction model in IBM's compact Granite Nano text family.
Launch date not confirmedDownloadable