Instructions to use mlx-community/granite-20b-code-base-4bit with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use mlx-community/granite-20b-code-base-4bit with Transformers:

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

pipe = pipeline("text-generation", model="mlx-community/granite-20b-code-base-4bit")

# Load model directly
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("mlx-community/granite-20b-code-base-4bit")
model = AutoModelForCausalLM.from_pretrained("mlx-community/granite-20b-code-base-4bit")

MLX

How to use mlx-community/granite-20b-code-base-4bit with MLX:

# Make sure mlx-lm is installed
# pip install --upgrade mlx-lm
# if on a CUDA device, also pip install mlx[cuda]

# Generate text with mlx-lm
from mlx_lm import load, generate

model, tokenizer = load("mlx-community/granite-20b-code-base-4bit")

prompt = "Once upon a time in"
text = generate(model, tokenizer, prompt=prompt, verbose=True)

Notebooks
Google Colab
Kaggle
Local Apps
LM Studio

vLLM

How to use mlx-community/granite-20b-code-base-4bit with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "mlx-community/granite-20b-code-base-4bit"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "mlx-community/granite-20b-code-base-4bit",
		"prompt": "Once upon a time,",
		"max_tokens": 512,
		"temperature": 0.5
	}'

Use Docker

docker model run hf.co/mlx-community/granite-20b-code-base-4bit

SGLang

How to use mlx-community/granite-20b-code-base-4bit 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 "mlx-community/granite-20b-code-base-4bit" \
    --host 0.0.0.0 \
    --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "mlx-community/granite-20b-code-base-4bit",
		"prompt": "Once upon a time,",
		"max_tokens": 512,
		"temperature": 0.5
	}'

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 "mlx-community/granite-20b-code-base-4bit" \
        --host 0.0.0.0 \
        --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "mlx-community/granite-20b-code-base-4bit",
		"prompt": "Once upon a time,",
		"max_tokens": 512,
		"temperature": 0.5
	}'

MLX LM

How to use mlx-community/granite-20b-code-base-4bit with MLX LM:

Generate or start a chat session

# Install MLX LM
uv tool install mlx-lm
# Generate some text
mlx_lm.generate --model "mlx-community/granite-20b-code-base-4bit" --prompt "Once upon a time"

Docker Model Runner
How to use mlx-community/granite-20b-code-base-4bit with Docker Model Runner:
```
docker model run hf.co/mlx-community/granite-20b-code-base-4bit
```

mlx-community/granite-20b-code-base-4bit

The Model mlx-community/granite-20b-code-base-4bit was converted to MLX format from ibm-granite/granite-20b-code-base using mlx-lm version 0.13.0.

Use with mlx

pip install mlx-lm

from mlx_lm import load, generate

model, tokenizer = load("mlx-community/granite-20b-code-base-4bit")
response = generate(model, tokenizer, prompt="hello", verbose=True)

Downloads last month: 14

MLX

Hardware compatibility

Quantized

Datasets used to train mlx-community/granite-20b-code-base-4bit

Evaluation results

pass@1 on MBPP
self-reported

43.800
pass@1 on MBPP+
self-reported

51.600
pass@1 on HumanEvalSynthesis(Python)
self-reported

48.200
pass@1 on HumanEvalSynthesis(Python)
self-reported

50.000
pass@1 on HumanEvalSynthesis(Python)
self-reported

59.100
pass@1 on HumanEvalSynthesis(Python)
self-reported

32.300
pass@1 on HumanEvalSynthesis(Python)
self-reported

40.900
pass@1 on HumanEvalSynthesis(Python)
self-reported

35.400
pass@1 on HumanEvalSynthesis(Python)
self-reported

17.100
pass@1 on HumanEvalSynthesis(Python)
self-reported

18.300
pass@1 on HumanEvalSynthesis(Python)
self-reported

23.200
pass@1 on HumanEvalSynthesis(Python)
self-reported

10.400
pass@1 on HumanEvalSynthesis(Python)
self-reported

25.600
pass@1 on HumanEvalSynthesis(Python)
self-reported

18.300
pass@1 on HumanEvalSynthesis(Python)
self-reported

23.200
pass@1 on HumanEvalSynthesis(Python)
self-reported

23.800