granite-3.1-8b-finetuned

Granite 3.1-8B Fine-tuned Model (SafeTensors Format)

This directory contains your fine-tuned Granite 3.1-8B model in SafeTensors format.

Contents

• model-00001-of-00004.safetensors through model-00004-of-00004.safetensors - Model weights split into 4 shards
• model.safetensors.index.json - Index file mapping tensors to shard files
• config.json - Model configuration
• tokenizer.json, tokenizer_config.json - Tokenizer files
• Other config files for generation, special tokens, etc.

Size

• Total: ~15.2 GB

Usage

Upload to HuggingFace

You can upload this entire directory to HuggingFace Hub:

huggingface-cli upload your-username/granite-3.1-8b-finetuned . --repo-type model

Use with Transformers

from transformers import AutoModelForCausalLM, AutoTokenizer

model = AutoModelForCausalLM.from_pretrained("path/to/this/directory")
tokenizer = AutoTokenizer.from_pretrained("path/to/this/directory")

Use with vLLM, TGI, or other frameworks

Point the framework to this directory path.

Model Info

• Base Model: IBM Granite 3.1-8B Instruct
• Fine-tuning: LoRA adapters merged with base model
• Precision: FP16
• Architecture: GraniteForCausalLM