Instructions to use henryhale/Kintu-2.0-30M with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use henryhale/Kintu-2.0-30M with Transformers:

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

pipe = pipeline("text-generation", model="henryhale/Kintu-2.0-30M")
messages = [
    {"role": "user", "content": "Who are you?"},
]
pipe(messages)

# pip install -U transformers accelerate
# Load model directly
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("henryhale/Kintu-2.0-30M")
model = AutoModelForCausalLM.from_pretrained("henryhale/Kintu-2.0-30M", device_map="auto")
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=256)
print(tokenizer.decode(outputs[0][inputs["input_ids"].shape[-1]:]))

Notebooks
Google Colab
Kaggle
Local Apps Settings

vLLM

How to use henryhale/Kintu-2.0-30M with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "henryhale/Kintu-2.0-30M"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "henryhale/Kintu-2.0-30M",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/henryhale/Kintu-2.0-30M

SGLang

How to use henryhale/Kintu-2.0-30M 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 "henryhale/Kintu-2.0-30M" \
    --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": "henryhale/Kintu-2.0-30M",
		"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 "henryhale/Kintu-2.0-30M" \
        --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": "henryhale/Kintu-2.0-30M",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Docker Model Runner
How to use henryhale/Kintu-2.0-30M with Docker Model Runner:
```
docker model run hf.co/henryhale/Kintu-2.0-30M
```
Browse Quantizations to use this model in llama.cpp, Ollama, LM Studio, or any compatible app.

Kintu-2.0-30M

A 30M-parameter Luganda chat model and the successor to Kintu-1.0-39M: Kintu-2.0-30M-Base fine-tuned on Luganda conversations.


Architecture	Qwen3-style decoder (`Qwen3ForCausalLM`): RMSNorm, SwiGLU, RoPE (θ = 1e6), QK-norm, grouped-query attention (8 query / 4 KV heads), tied embeddings
Size	30.0M parameters: 8 layers, hidden size 512, MLP 1,664
Tokenizer	Byte-level BPE, 6,400 tokens, trained on the pretraining data
Context	512 tokens

Training

Full fine-tuning on 59,798 Luganda prompt-reply pairs (the empathetic-dialogue and everyday-conversation subsets of henryhale/luganda-sft-translated) for 3 epochs : learning rate 5e-5 and 685 multi-turn Luganda converstations of henryhale/luganda-lima for 3 epochs : learning rate 2e-5with cosine decay, fp16 on one T4. Loss is only computed on the assistant's replies.

Usage

from transformers import AutoModelForCausalLM, AutoTokenizer

tok = AutoTokenizer.from_pretrained("henryhale/Kintu-2.0-30M")
model = AutoModelForCausalLM.from_pretrained("henryhale/Kintu-2.0-30M")
messages = [{"role": "user", "content": "Oli otya?"}]
enc = tok.apply_chat_template(messages, add_generation_prompt=True, return_tensors="pt", return_dict=True)
out = model.generate(**enc, max_new_tokens=100)
print(tok.decode(out[0, enc["input_ids"].shape[1]:], skip_special_tokens=True))

Samples

Sampled with temperature 0.85 and top-p 0.95:

💬 Oli otya?
🧠 Gezaako okukuumira ebirowoozo byo mu kamwa ko era weeyongere okunyweza omubiri gwo.

💬 Wasuze otya nno?
🧠 Wasuze otya naawe! Ndi muzadde omulungi nnyo! Ggwe osiibye otya?

💬 Osiibye otyanno ssebo?
🧠 Nsiibye bulungi. Oluggi luno lwakumala bbanga ki?

💬 Ndi musanyufu nnyo okulaba.
🧠 Nzikirizza kampuni ey'eby'enfuna n'eby'abakozi abakugu.

💬 Weebale nnyo okuyamba bulungi.
🧠 Ssanyu lyange. Ssanyu lyange. Olina omulimu?

Limitations

At 30M parameters it writes fluent Luganda but is often off-topic and gets facts wrong, so don't rely on it for information. It was trained on sequences of up to 512 tokens, and TinyStories translations are 68.5% of its pretraining tokens, so it leans towards simple stories.