Instructions to use Tekimax/granite-ml-coder with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use Tekimax/granite-ml-coder with Transformers:

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

pipe = pipeline("text-generation", model="Tekimax/granite-ml-coder")
messages = [
    {"role": "user", "content": "Who are you?"},
]
pipe(messages)

# Load model directly
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("Tekimax/granite-ml-coder")
model = AutoModelForCausalLM.from_pretrained("Tekimax/granite-ml-coder")
messages = [
    {"role": "user", "content": "Who are you?"},
]
inputs = tokenizer.apply_chat_template(
	messages,
	add_generation_prompt=True,
	tokenize=True,
	return_dict=True,
	return_tensors="pt",
).to(model.device)

outputs = model.generate(**inputs, max_new_tokens=40)
print(tokenizer.decode(outputs[0][inputs["input_ids"].shape[-1]:]))

MLX

How to use Tekimax/granite-ml-coder with MLX:

# Make sure mlx-lm is installed
# pip install --upgrade mlx-lm

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

model, tokenizer = load("Tekimax/granite-ml-coder")

prompt = "Write a story about Einstein"
messages = [{"role": "user", "content": prompt}]
prompt = tokenizer.apply_chat_template(
    messages, add_generation_prompt=True
)

text = generate(model, tokenizer, prompt=prompt, verbose=True)

Notebooks
Google Colab
Kaggle
Local Apps Settings
LM Studio

vLLM

How to use Tekimax/granite-ml-coder with vLLM:

Install from pip and serve model

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

Use Docker

docker model run hf.co/Tekimax/granite-ml-coder

SGLang

How to use Tekimax/granite-ml-coder 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 "Tekimax/granite-ml-coder" \
    --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": "Tekimax/granite-ml-coder",
		"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 "Tekimax/granite-ml-coder" \
        --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": "Tekimax/granite-ml-coder",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

How to use Tekimax/granite-ml-coder with Pi:

Start the MLX server

# Install MLX LM:
uv tool install mlx-lm
# Start a local OpenAI-compatible server:
mlx_lm.server --model "Tekimax/granite-ml-coder"

Configure the model in Pi

# Install Pi:
npm install -g @mariozechner/pi-coding-agent
# Add to ~/.pi/agent/models.json:
{
  "providers": {
    "mlx-lm": {
      "baseUrl": "http://localhost:8080/v1",
      "api": "openai-completions",
      "apiKey": "none",
      "models": [
        {
          "id": "Tekimax/granite-ml-coder"
        }
      ]
    }
  }
}

Run Pi

# Start Pi in your project directory:
pi

Hermes Agent new

How to use Tekimax/granite-ml-coder with Hermes Agent:

Start the MLX server

# Install MLX LM:
uv tool install mlx-lm
# Start a local OpenAI-compatible server:
mlx_lm.server --model "Tekimax/granite-ml-coder"

Configure Hermes

# Install Hermes:
curl -fsSL https://hermes-agent.nousresearch.com/install.sh | bash
hermes setup
# Point Hermes at the local server:
hermes config set model.provider custom
hermes config set model.base_url http://127.0.0.1:8080/v1
hermes config set model.default Tekimax/granite-ml-coder

Run Hermes

hermes

MLX LM

How to use Tekimax/granite-ml-coder with MLX LM:

Generate or start a chat session

# Install MLX LM
uv tool install mlx-lm
# Interactive chat REPL
mlx_lm.chat --model "Tekimax/granite-ml-coder"

Run an OpenAI-compatible server

# Install MLX LM
uv tool install mlx-lm
# Start the server
mlx_lm.server --model "Tekimax/granite-ml-coder"
# Calling the OpenAI-compatible server with curl
curl -X POST "http://localhost:8000/v1/chat/completions" \
   -H "Content-Type: application/json" \
   --data '{
     "model": "Tekimax/granite-ml-coder",
     "messages": [
       {"role": "user", "content": "Hello"}
     ]
   }'

Docker Model Runner
How to use Tekimax/granite-ml-coder with Docker Model Runner:
```
docker model run hf.co/Tekimax/granite-ml-coder
```
Browse Quantizations to use this model in llama.cpp, Ollama, LM Studio, or any compatible app.

granite-ml-coder

A compact Python / machine-learning coding assistant, fine-tuned from ibm-granite/granite-3.1-1b-a400m-instruct. It writes runnable scikit-learn / pandas / NumPy code, explains ML pipeline steps, and reasons about everyday concepts like overfitting, cross-validation, and gradient descent.

It is small enough to run fully locally — on a laptop CPU, via Ollama, or quantized to GGUF — which makes it a good private copilot for data-science work where your data shouldn't leave the machine.

Built as the reference model for the TEKIMAX ML Model Workshop (fine-tune → host on HF/Ollama → build a production DL app).

Quick start

Transformers

import torch
from transformers import AutoModelForCausalLM, AutoTokenizer

tok = AutoTokenizer.from_pretrained("Tekimax/granite-ml-coder")
model = AutoModelForCausalLM.from_pretrained(
    "Tekimax/granite-ml-coder", dtype=torch.float32, attn_implementation="eager"
)

messages = [
    {"role": "system", "content": "You are an expert Python machine-learning engineer."},
    {"role": "user", "content": "Write a scikit-learn pipeline to classify the iris dataset and explain how you avoid overfitting."},
]
enc = tok.apply_chat_template(messages, add_generation_prompt=True,
                             return_tensors="pt", return_dict=True, enable_thinking=False)
out = model.generate(**enc, max_new_tokens=400, do_sample=True, temperature=0.7, top_p=0.9)
print(tok.decode(out[0][enc["input_ids"].shape[1]:], skip_special_tokens=True))

Ollama (recommended for local use)

ollama run tekimaxllc/granite-ml-coder "Write a Keras autoencoder for network-traffic anomaly detection"

GGUF / llama.cpp

A 4-bit Q4_K_M build (~378 MB) is available at Tekimax/granite-ml-coder-GGUF:

llama-cli -m granite-ml-coder-Q4_K_M.gguf -p "Write a sklearn pipeline"

Intended use

Drafting Python ML code (scikit-learn, pandas, NumPy, Keras) inside notebooks/IDEs
Explaining ML pipeline steps and concepts (overfitting, gradient descent, model choice)
A private, offline coding copilot for data-science tasks

Training


Base model	`ibm-granite/granite-3.1-1b-a400m-instruct` (Apache-2.0, IBM) — ~1.3B total / 400M active MoE
Data	`iamtarun/python_code_instructions_18k_alpaca`, filtered to ML/DS rows (≈2,341 examples)
Method	Full fine-tune, instruction format with the Granite chat template; loss computed on the assistant answer only (prompt tokens masked)
Schedule	2 epochs · effective batch size 16 · LR 2e-5 cosine · max_len 512
Hardware	Apple M2 Ultra, CPU (the MPS/Metal backend was unstable for fine-tuning on torch 2.12 — see the workshop appendix)
Result	training loss decreasing steadily, no NaN

Limitations

It's a 1B model. It learns style, format, and common patterns well, but is not a frontier model — it can produce incomplete or subtly wrong code, and it won't reliably "pick the best model" for hard problems. Treat its output as a fast first draft and verify before use.
English, Python-focused. Strongest on classic ML (sklearn/pandas); weaker on large, novel, or multi-file tasks.
Inherits any biases/limitations of the base model and the training dataset.

License

Apache-2.0, inherited from the base model and dataset.

Citation / credits

Base model: Granite
Dataset: iamtarun/python_code_instructions_18k_alpaca
Built with Hugging Face transformers Trainer; quantized with llama.cpp.

Downloads last month: 23

Safetensors

Model size

1B params

Tensor type

F32

MLX

Hardware compatibility

Quantized

Model tree for Tekimax/granite-ml-coder

Base model

ibm-granite/granite-3.1-1b-a400m-base

Finetuned

ibm-granite/granite-3.1-1b-a400m-instruct

Finetuned

(3)

this model

Quantizations

1 model

Tekimax
/

granite-ml-coder