---

pipeline_tag: text-generation

language: por

license: apache-2.0

tags:

- trimmed

- granite

- granite-4.0

- gguf

- portuguese

library_name: gguf

base_model: alphaedge-ai/granite-4.0-350m-por-16384

base_model_relation: quantized

datasets:

- lbourdois/fineweb-2-trimming

---

granite-4.0-350m-por-16384-GGUF

This is a GGUF format conversion of alphaedge-ai/granite-4.0-350m-por-16384, a trimmed Portuguese language model based on ibm-granite/granite-4.0-350m.

The model vocabulary was reduced from 100,352 to 16,384 tokens using the trimming method, resulting in a 24.40% smaller model optimized for Portuguese.

Model Statistics

| Metric | Original | Trimmed | Reduction |

|--------|----------|---------|-----------|

| Vocabulary size | 100,352 tokens | 16,384 tokens | 83.67% |

| Model size | 352,379,904 params | 266,396,672 params | 24.40% |

Usage with llama.cpp

./llama-cli -m granite-4.0-350m-por-16384.gguf \
  -p "Qual é a capital do Brasil?" \
  -n 256

Or with the LlamaServer:

./llama-server -m granite-4.0-350m-por-16384.gguf --port 8080

Conversion Details

• Format: GGUF (GPT-Generated Unified Format)
• Tensor type: BF16
• Architecture: Granite (dense, no MoE)
• Context length: 32768 tokens
• Tokenizer: GPT-2 BPE (tied embeddings)

Citations

@misc{granite2025,
  author       = {IBM Research},
  title        = {Granite 4.0 Language Models},
  year         = {2025},
  howpublished = {https://github.com/ibm-granite/granite-4.0-language-models},
}

@misc{hf_blogpost_trimming,
  title={Introduction to Trimming}, 
  author={Loïck BOURDOIS and Tom AARSEN and Bram VANROY and Christopher AKIKI and Woojun JUNG and Manuel ROMERO and Prithiv SAKTHI},
  year={2026},
  url={https://huggingface.co/blog/lbourdois/introduction-to-trimming}, 
}