Instructions to use ibm-granite/granite-switch-4.1-8b-preview with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use ibm-granite/granite-switch-4.1-8b-preview with Transformers:

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

pipe = pipeline("text-generation", model="ibm-granite/granite-switch-4.1-8b-preview")
messages = [
    {"role": "user", "content": "Who are you?"},
]
pipe(messages)

# Load model directly
from transformers import AutoModelForCausalLM
model = AutoModelForCausalLM.from_pretrained("ibm-granite/granite-switch-4.1-8b-preview", dtype="auto")

Notebooks
Google Colab
Kaggle
Local Apps Settings

vLLM

How to use ibm-granite/granite-switch-4.1-8b-preview with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "ibm-granite/granite-switch-4.1-8b-preview"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "ibm-granite/granite-switch-4.1-8b-preview",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/ibm-granite/granite-switch-4.1-8b-preview

SGLang

How to use ibm-granite/granite-switch-4.1-8b-preview 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 "ibm-granite/granite-switch-4.1-8b-preview" \
    --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": "ibm-granite/granite-switch-4.1-8b-preview",
		"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 "ibm-granite/granite-switch-4.1-8b-preview" \
        --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": "ibm-granite/granite-switch-4.1-8b-preview",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Docker Model Runner
How to use ibm-granite/granite-switch-4.1-8b-preview with Docker Model Runner:
```
docker model run hf.co/ibm-granite/granite-switch-4.1-8b-preview
```

granite-switch-4.1-8b-preview / BUILD.md

antonpuz

Updating GraniteSwitch 4.1 8b

c739556 verified 13 days ago

preview code

raw

history blame contribute delete

2.99 kB

Granite Switch Composed Model

Base Model

Identifier: ibm-granite/granite-4.1-8b
Model type: granite
Architectures: GraniteForCausalLM
Hidden size: 4096
Hidden layers: 40
Attention heads: 32
Vocab size: 100352

Embedded Adapters

Total adapters: 12

#	Name	Technology	Control Token	Token ID	Rank	Alpha	Target Modules	Source
1	citations	lora	`<\|citations\|>`	100352	16	32	k_proj, o_proj, q_proj, v_proj	ibm-granite/granitelib-rag-r1.0
2	query_rewrite	alora	`<\|query_rewrite\|>`	100353	32	32	down_proj, gate_proj, k_proj, o_proj, q_proj, up_proj, v_proj	ibm-granite/granitelib-rag-r1.0
3	query_clarification	alora	`<\|query_clarification\|>`	100354	32	64	down_proj, gate_proj, k_proj, o_proj, q_proj, up_proj, v_proj	ibm-granite/granitelib-rag-r1.0
4	hallucination_detection	lora	`<\|hallucination_detection\|>`	100355	16	32	down_proj, gate_proj, k_proj, o_proj, q_proj, up_proj, v_proj	ibm-granite/granitelib-rag-r1.0
5	answerability	alora	`<\|answerability\|>`	100356	16	32	down_proj, gate_proj, k_proj, o_proj, q_proj, up_proj, v_proj	ibm-granite/granitelib-rag-r1.0
6	factuality-detection	alora	`<\|factuality-detection\|>`	100357	32	64	k_proj, q_proj, v_proj	ibm-granite/granitelib-guardian-r1.0
7	policy-guardrails	alora	`<\|policy-guardrails\|>`	100358	16	32	k_proj, o_proj, q_proj, v_proj	ibm-granite/granitelib-guardian-r1.0
8	factuality-correction	alora	`<\|factuality-correction\|>`	100359	32	64	k_proj, q_proj, v_proj	ibm-granite/granitelib-guardian-r1.0
9	guardian-core	alora	`<\|guardian-core\|>`	100360	16	64	k_proj, o_proj, q_proj, v_proj	ibm-granite/granitelib-guardian-r1.0
10	uncertainty	alora	`<\|uncertainty\|>`	100361	32	64	down_proj, gate_proj, k_proj, o_proj, q_proj, up_proj, v_proj	ibm-granite/granitelib-core-r1.0
11	requirement-check	alora	`<\|requirement-check\|>`	100362	16	64	k_proj, o_proj, q_proj, v_proj	ibm-granite/granitelib-core-r1.0
12	context-attribution	lora	`<\|context-attribution\|>`	100363	16	32	k_proj, q_proj, v_proj	ibm-granite/granitelib-core-r1.0

Composition Details

base_param_count: 8,380,551,168
composed_param_count: 9,568,112,640
Param delta: +14.17%
compose_settings:
- adapter_substitute_token_ids:
  - 100264
  - 100264
  - 100264
  - 100264
  - 100264
  - 27
  - 27
  - 27
  - 27
  - 27
  - 27
  - 100264
- target_model: "granite-4.1-8b"
adapter_sources:
- "ibm-granite/granitelib-rag-r1.0": "2f0b2c79c6731068625aca8045c2eb2e8912b353"
- "ibm-granite/granitelib-guardian-r1.0": "773b254e98f993a605ec4b6259634906e0e64e8e"
- "ibm-granite/granitelib-core-r1.0": "d0a2a96a4cd07e96f0fe7ca29a42bfe088299d43"