Instructions to use webAI-Official/granite-4.2-8b-builder-lora-peft with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use webAI-Official/granite-4.2-8b-builder-lora-peft with PEFT:

from peft import PeftModel
from transformers import AutoModelForCausalLM

base_model = AutoModelForCausalLM.from_pretrained("ibm-granite/granite-4.2-8b")
model = PeftModel.from_pretrained(base_model, "webAI-Official/granite-4.2-8b-builder-lora-peft")

Transformers

How to use webAI-Official/granite-4.2-8b-builder-lora-peft with Transformers:

# Use a pipeline as a high-level helper
from transformers import pipeline

pipe = pipeline("text-generation", model="webAI-Official/granite-4.2-8b-builder-lora-peft")
messages = [
    {"role": "user", "content": "Who are you?"},
]
pipe(messages)

# Load model directly
from transformers import AutoModel
model = AutoModel.from_pretrained("webAI-Official/granite-4.2-8b-builder-lora-peft", device_map="auto")

Notebooks
Google Colab
Kaggle
Local Apps Settings

vLLM

How to use webAI-Official/granite-4.2-8b-builder-lora-peft with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "webAI-Official/granite-4.2-8b-builder-lora-peft"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "webAI-Official/granite-4.2-8b-builder-lora-peft",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/webAI-Official/granite-4.2-8b-builder-lora-peft

SGLang

How to use webAI-Official/granite-4.2-8b-builder-lora-peft with SGLang:

Install from pip and serve model

# Install SGLang from pip:
pip install sglang
# Start the SGLang server:
python3 -m sglang.launch_server \
    --model-path "webAI-Official/granite-4.2-8b-builder-lora-peft" \
    --host 0.0.0.0 \
    --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "webAI-Official/granite-4.2-8b-builder-lora-peft",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker images

docker run --gpus all \
    --shm-size 32g \
    -p 30000:30000 \
    -v ~/.cache/huggingface:/root/.cache/huggingface \
    --env "HF_TOKEN=<secret>" \
    --ipc=host \
    lmsysorg/sglang:latest \
    python3 -m sglang.launch_server \
        --model-path "webAI-Official/granite-4.2-8b-builder-lora-peft" \
        --host 0.0.0.0 \
        --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "webAI-Official/granite-4.2-8b-builder-lora-peft",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Docker Model Runner
How to use webAI-Official/granite-4.2-8b-builder-lora-peft with Docker Model Runner:
```
docker model run hf.co/webAI-Official/granite-4.2-8b-builder-lora-peft
```

granite-4.2-8b-builder-lora-peft

PEFT LoRA adapter for the builder persona, trained on ibm-granite/granite-4.2-8b. A llama.cpp version of the same weights is at webAI-Official/granite-4.2-8b-builder-lora-GGUF.

Usage (Transformers + PEFT)

import torch
from transformers import AutoModelForCausalLM, AutoTokenizer
from peft import PeftModel

base = AutoModelForCausalLM.from_pretrained("ibm-granite/granite-4.2-8b", torch_dtype=torch.bfloat16, device_map="auto")
model = PeftModel.from_pretrained(base, "webAI-Official/granite-4.2-8b-builder-lora-peft")
tokenizer = AutoTokenizer.from_pretrained("webAI-Official/granite-4.2-8b-builder-lora-peft")

messages = [{"role": "user", "content": "Hello!"}]
inputs = tokenizer.apply_chat_template(messages, add_generation_prompt=True, return_tensors="pt").to(model.device)
print(tokenizer.decode(model.generate(inputs, max_new_tokens=256)[0][inputs.shape[-1]:], skip_special_tokens=True))

Call model.merge_and_unload() to fold the adapter into the base weights. The adapter was saved with PEFT 0.21.0; the tokenizer config uses the Transformers v5 format.

Files

adapter_model.safetensors, adapter_config.json: LoRA weights (fp32) and config
tokenizer.json, tokenizer_config.json, chat_template.jinja: tokenizer and chat template used in training
persona.json, run_config.json, train_metrics.json: persona settings, training config, and training metrics

Adapter details

Type: LoRA, rank 16, alpha 32, dropout 0.05
Target modules: q_proj, k_proj, v_proj, o_proj, gate_proj, up_proj, down_proj
Thinking mode: thinking
Training: 2776 examples, 3 epochs, lr 0.0001, cosine schedule, max seq len 8192, packing
Final step 522; train loss 0.3201; eval loss 0.3395

Downloads last month: 14

Model tree for webAI-Official/granite-4.2-8b-builder-lora-peft

Base model

ibm-granite/granite-4.1-8b-base

Finetuned

ibm-granite/granite-4.2-8b

Adapter

(5)

this model